Chips and China

Intel may not be the most obvious place to start when it comes to the China chip sanctions announced by the Biden administration three weeks ago (I covered the ban in the Daily Update here and here); the company recently divested its 3DNAND fab in Dalian, and only maintains two test and assembly sites in Chengdu. Sure, there is an angle about Intel’s future as a foundry and its importance in helping the United States catch up in terms of the most advanced processes currently dominated by Taiwan’s TSMC, but when it comes to exploring the implications and risks of these sanctions I am much more interested in Intel’s past.

Start with the present, though: two weeks ago Intel CEO Pat Gelsinger announced a restructuring of the company, with the goal of putting more distance between its design and manufacturing teams. From the Wall Street Journal:

Intel Corp. plans to create greater decision-making separation between its chip designers and chip-making factories as part of Chief Executive Pat Gelsinger’s bid to revamp the company and boost returns. The new structure, which Mr. Gelsinger disclosed in a letter to staff on Tuesday, is designed to let Intel’s network of factories operate like a contract chip-making operation, taking orders from both Intel engineers and external chip companies on an equal footing. Intel has historically used its factories almost exclusively to make its own chips, something Mr. Gelsinger changed when he launched a contract chip-making arm last year.

Back in 2018 I wrote about Intel and the Danger of Integration:

It is perhaps simpler to say that Intel, like Microsoft, has been disrupted. The company’s integrated model resulted in incredible margins for years, and every time there was the possibility of a change in approach Intel’s executives chose to keep those margins. In fact, Intel has followed the script of the disrupted even more than Microsoft: while the decline of the PC finally led to The End of Windows, Intel has spent the last several years propping up its earnings by focusing more and more on the high-end, selling Xeon processors to cloud providers. That approach was certainly good for quarterly earnings, but it meant the company was only deepening the hole it was in with regards to basically everything else. And now, most distressingly of all, the company looks to be on the verge of losing its performance advantage even in high-end applications.

That article was primarily about Intel’s reliance on high margin integrated processors and its unwillingness/inability to become a foundry serving 3rd-party customers, and how smartphones provided the volume for modular players like TSMC to threaten Intel’s manufacturing dominance. However, it’s worth diving into the implications of Intel’s integrated approach relative to TSMC’s modular approach, because it offers lessons for the long road facing China when it comes to building its own semiconductor industry, highlights why the U.S. is itself vulnerable in semiconductors, and explains why the risk for Taiwan has increased significantly.

TSMC’s Depreciation

Fabs are incredibly expensive to build, while chips are extremely cheap; to put it in economic terms, fabs entail massive fixed costs, while chips have minimal marginal costs. This dynamic is very similar to software, which is why venture capital rose up to support chip companies like Intel, and then seamlessly transitioned to supporting software (Silicon Valley, which is today known for software, is literally named for the material used for chips).

One way to manage these costs is to build a fab once and then run it for as long as possible. TSMC’s Fab 2, for example, the company’s sole 150-millimeter wafer facility, was built in 1990, and is still in operation today. That is one of seven TSMC fabs that are over 20 years old, amongst the company’s 26 total (several more are under construction, including the one in Arizona). The chips in these fabs don’t sell for much, but that’s ok because the fabs are completely depreciated: almost all of the revenue is pure profit.

This may seem like the obvious strategy, but it’s a very path dependent one: TSMC was unique precisely because they didn’t design their own chips. I explained the company’s origin story in Chips and Geopolitics:

A few years later, in 1987, Chang was invited home to Taiwan, and asked to put together a business plan for a new government initiative to create a semiconductor industry. Chang explained in an interview with the Computer History Museum that he didn’t have much to work with:

I paused to try to examine what we have got in Taiwan. And my conclusion was that [we had] very little. We had no strength in research and development, or very little anyway. We had no strength in circuit design, IC product design. We had little strength in sales and marketing, and we had almost no strength in intellectual property. The only possible strength that Taiwan had, and even that was a potential one, not an obvious one, was semiconductor manufacturing, wafer manufacturing. And so what kind of company would you create to fit that strength and avoid all the other weaknesses? The answer was pure-play foundry…

In choosing the pure-play foundry mode, I managed to exploit, perhaps, the only strength that Taiwan had, and managed to avoid a lot of the other weaknesses. Now, however, there was one problem with the pure-play foundry model and it could be a fatal problem which was, “Where’s the market?”

What happened is exactly what Christensen would describe several years later: TSMC created the market by “enabl[ing] independent, nonintegrated organizations to sell, buy, and assemble components and subsystems.” Specifically, Chang made it possible for chip designers to start their own companies:

When I was at TI and General Instrument, I saw a lot of IC [Integrated Circuit] designers wanting to leave and set up their own business, but the only thing, or the biggest thing that stopped them from leaving those companies was that they couldn’t raise enough money to form their own company. Because at that time, it was thought that every company needed manufacturing, needed wafer manufacturing, and that was the most capital intensive part of a semiconductor company, of an IC company. And I saw all those people wanting to leave, but being stopped by the lack of ability to raise a lot of money to build a wafer fab. So I thought that maybe TSMC, a pure-play foundry, could remedy that. And as a result of us being able to remedy that then those designers would successfully form their own companies, and they will become our customers, and they will constitute a stable and growing market for us.

It worked. Graphics processors were an early example: Nvidia was started in 1993 with only $20 million, and never owned its own fab.1 Qualcomm, after losing millions manufacturing its earliest designs, spun off its chip-making unit in 2001 to concentrate on design, and Apple started building its own chips without a fab a decade later. Today there are thousands of chip designers in all kinds of niches creating specialized chips for everything from appliances to fighter jets, and none of them have their own fab.

By creating this new market TSMC ended up with a massive customer base; moreover, most of those customers didn’t need cutting edge chips, but rather the same chip that they started with for as long as they made the product into which that chip went. That, by extension, meant that all of those old foundries had a customer base, enabling TSMC to make money on them long after they had been paid off.

Intel’s Margins

Intel’s path, though, preceded TSMC’s, which is to say that of course Intel both designed and manufactured their own chips (“real men have fabs”, as AMD founder Jerry Sanders once famously put it); to put it another way, the entire reason why Chang saw a market in being just a manufacturer was because every company that proceeded TSMC had done both out of necessity, because a company like TSMC didn’t exist.

And, it’s worth noting, there was no reason for TSMC to exist: Intel’s chips, for the two decades it existed before TSMC, were never good enough: every generation would result in such massive leaps in performance that it simply wouldn’t have made sense to keep the old assembly lines around. Still, this stuff was expensive, which is where being integrated helped.

This was the other way to manage the cost of cutting edge fabs: because Intel was at the cutting edge, it would charge a huge premium for its chips (and thus have the highest margins in the industry that I referenced earlier). At the beginning, when fabs were cheaper, Intel was happy to sell off its old equipment and make a few extra bucks on the back end. Over the last decade, though, as equipment became more and more expensive, and as Intel’s leadership started to care more about finances than about engineering, it increasingly became a priority to re-use equipment to the greatest extent possible. This wasn’t easy, I would note: Intel would stick with (relatively) outdated equipment in not just one fab but also in the fabs it built around the world.

This is where the integration point was critical: because Intel both designed and manufactured its chips, the latter could call the shots for the former; chips had to be designed to work with Intel manufacturing, not the other way around, and this extended to not just the designs themselves but all of the tooling that went into it. Intel, for example, used its own chip design software, and favored suppliers who would do what Intel told them to, and then hand the equipment off to Intel to do with it as they saw fit. Intel would then get everything to work in one fab, and Copy Exactly! that fab in another location: everything was identical, down to the position of the toilets in the bathrooms.

As I noted in the conclusion of Intel and the Danger of Integration, Intel’s strategy worked phenomenally well, right up until it didn’t:

What makes disruption so devastating is the fact that, absent a crisis, it is almost impossible to avoid. Managers are paid to leverage their advantages, not destroy them; to increase margins, not obliterate them. Culture more broadly is an organization’s greatest asset right up until it becomes a curse. To demand that Intel apologize for its integrated model is satisfying in 2018, but all too dismissive of the 35 years of success and profits that preceded it. So it goes.

So it goes, indeed — or rather, the correct conjugation is the past tense: so went Intel’s manufacturing advantage.

ASML’s Rise

I mentioned TSMC’s Fab 2 earlier and its 150-millimeter wafers; that is 1980’s era technology. The 1990s brought 200-millimeter wafers (which are used in seven of TSMC’s fabs). It was the transition to today’s 300-millimeter fabs in the early 2000’s, though, that marked the rise of ASML.

Intel’s partner in the lithography space — the use of light to draw transistors on wafers — was Nikon, and Nikon’s approach to 300-millimeter wafers was to scale up its 200-millimeter process. There was a downside to this approach, though: because the wafers were larger they had to move more slowly (more mass means more force, unless acceleration is decreased). This was fine with Intel, though: they were their own only customer, and their margins were plenty high enough to handle a decrease in throughput (indeed, Intel was well-known for running their machines well below capacity).

Lower speed wasn’t fine for TSMC and Samsung, the other up-and-comer in the space: like any challenger they were operating on much lower margins, and they didn’t want a decrease in throughput — the entire point of larger wafers was to increase the number of chips that could be produced, not to give away that gain by running everything more slowly. ASML saw the opportunity and designed an entirely new process around 300-millimeter wafers, creating dual wafer stage technology that aligned and mapped one wafer while another was being exposed.

TSMC and ASML were already close, in part because both were part of the Philips family tree (Philips was the only external investor in TSMC, which licensed Philips technology to start, and ASML was a joint venture of Philips and ASMI). What was more important is that both were ignored by the dominant players in the industry: the big chip makers, from Intel to Motorola to Texas Instruments, were matched up with Nikon and Canon; the former didn’t want equipment from a new entrant, and the latter didn’t have capacity for a foundry that was not only working on low margins but also, as part of its cost consciousness, wanted to learn how to service the machines themselves (the Japanese companies preferred to deliver black boxes that their own technicians would service).

ASML’s 300-nanometer process, though, required a reworking on the fab side as well. Now TSMC and ASML weren’t simply stuck together like two kids picked last at recess: they were deeply enmeshed in the process of working through the new process’s bugs, designing new fabs to support it, and maximizing output once everything was working. This increase in output had another side effect: TSMC started to make a bit more money, which it started pouring into its own research and development. It was TSMC that pushed ASML towards immersion lithography, where the space between the lens and the wafer was filled with a liquid with a higher refraction index than air. Nikon would eventually be forced to respond with its own lithography machines, but they were never as good as ASML’s, which meant that even Intel had to come calling as a customer.

ASML, meanwhile, had been working for years on a true moonshot: extreme ultraviolet lithography. Here is the Brookings Institution’s description of the process:

A generator ejects 50,000 tiny droplets of molten tin per second. A high-powered laser blasts each droplet twice. The first shapes the tiny tin, so the second can vaporize it into plasma. The plasma emits extreme ultraviolet (EUV) radiation that is focused into a beam and bounced through a series of mirrors. The mirrors are so smooth that if expanded to the size of Germany they would not have a bump higher than a millimeter. Finally, the EUV beam hits a silicon wafer — itself a marvel of materials science — with a precision equivalent to shooting an arrow from Earth to hit an apple placed on the moon. This allows the EUV machine to draw transistors into the wafer with features measuring only five nanometers — approximately the length your fingernail grows in five seconds. This wafer with billions or trillions of transistors is eventually made into computer chips.

An EUV machine is made of more than 100,000 parts, costs approximately $120 million, and is shipped in 40 freight containers. There are only several dozen of them on Earth and approximately two years’ worth of back orders for more. It might seem unintuitive that the demand for a $120 million tool far outstrips supply, but only one company can make them. It’s a Dutch company called ASML, which nearly exclusively makes lithography machines for chip manufacturing.

It’s not just ASML, though: that mirror is made by Zeiss, and the laser is made by TRUMPF using carbon dioxide sources pioneered by Access Laser (a U.S. company later acquired by TRUMPF). They are the two most important of over 800 suppliers for EUV, but it’s the end users that are equally essential.

When TSMC Passed Intel

In 2012 Intel, TSMC, and Samsung all invested in ASML to help the company finish the EUV project that had started 11 years earlier: there were very real questions about whether or not ASML would ever ship, or die trying, while it was clear that immersion lithography was reaching the limits of what was possible. The investment amounts are interesting in retrospect:

Company Intel TSMC Samsung
Investment in stock 15% for $3.1 billion 5% for $1.03 billion 3% for $630 million
Investment in R&D $1 billion $345 million $345 million

Intel, despite investing the most (and having contributed a big chunk of the underlying technology), was convinced it could stick with immersion lithography as it transitioned first to 10-nanometer and then 7-nanometer chips. Yes, those were awfully small lines to be drawing with a light source that was 193-nanometers in width, but it wasn’t clear that EUV yields were going to be high enough, and besides, Intel had a lot of lithography equipment that, if used for one or two more generations, would make for some very fat margins. That was more of a priority for Intel than technological leadership, even as decades of said leadership had created the arrogance to believe that Intel could use quad-patterning — i.e. doing four exposures on a single wafer — to create those ever thinner lines.

TSMC, on the other hand, had three reasons to commit to EUV:

  • First, TSMC had a multi-decade relationship with ASML that included two significant process transitions (to 300-millimeter wafers and immersion lithography).
  • Second, because TSMC was a foundry, it needed to manufacture smaller lots of much greater variety; this meant that fiddly multi-pattern approaches that took many runs to improve yields didn’t make sense. EUV’s 13.5 nanometer light offered the potential for much simpler designs that fit TSMC’s business model.
  • Third, Apple was willing to pay to have the fastest chips in the world, which meant that TSMC had a guaranteed first customer with massive volume whenever it could get EUV working.

In the end, TSMC started using EUV for non-critical layers at 7 nanometers, and for critical layers at 5 nanometers (in 2020); Intel, meanwhile, failed for years to ship 10 nanometer chips (which are closer to TSMC’s 7 nanometer chips), and had to completely rework its 7 nanometer process to incorporate EUV. Those chips are only starting mass production this fall — the same time period when TSMC is shipping new 3 nanometer chips. Intel, by the way, is a customer for TSMC’s 3nm process: the company’s performance was falling too far behind AMD, which abandoned its own fabs in 2009 and has been riding TSMC’s improvements (along with its own new designs) for the last five years.

China’s Integrated Path

Only now, 3,500 words in, do I turn to China, and the country’s path forward to building the sort of advanced chips that the U.S. has just cut off access to. That, though, is the point: the chip industry’s path to today is China’s path to the future.

This is a daunting challenge: it’s not just that China needs to re-create TSMC, but also ASML, Lam Research, Applied Materials, Tokyo Electronic, and all of the other pieces of the foundry supply chain. And, to go one layer deeper, not only does China need to re-create ASML, but also Zeiss, and TRUMPF, and Access Laser, and all of the other pieces of the global supply chain, much of which is not located in China. China’s manufacturing prowess is centered on traditionally labor-centric components; even though Chinese labor is now much more expensive than it was, and automation much more common, path dependency matters, and China’s capability is massive but in some respects limited.

Globalization made all of those Chinese factories extremely valuable, because the world was China’s market. At the same time, globalization also meant that China could buy high-precision capital-intensive goods abroad: it didn’t need to build them itself to get the benefits immediately. By the same token high-precision capital-intensive goods are exactly what Western countries like the U.S., Germany, Netherlands, Japan and Taiwan invested in, in part because they couldn’t compete with China on labor. To put it another way, the principles of comparative advantage governed an infinite number of decisions on the margins that led to the U.S. government having the ability to impose these sanctions on China; the realities of semiconductor manufacturing, where every paradigm shift costs massive amounts of money, years in R&D, and the willingness of partners to take the leap with you, are a further manifestation of comparative advantage: it simply makes the most sense for one company to do lithography, and another to lead the world in fabrication.

In other words, China is going to need to build up these capabilities from the ground up, and it’s going to be a long hard road. Moreover, China will not have the benefit of partnership and distributed expertise that have driven the last decade of innovation: in some respects China is going to need to be Intel, doing too much on its own.

That said, the country does have three big advantages:

  • First, it is much easier to follow a path than to forge a new one. China may not be able to make EUV machines, but at least they know they can be made.
  • Second, China has benefited from all of the technological sharing to date: Semiconductor Manufacturing International Corporation (SMIC) has successfully manufactured 7nm chips (using ASML’s immersion lithography machines), and Shanghai Micro Electronics Equipment (SMEE) has built its own immersion lithography machines. Granted, those 7nm chips almost certainly had poor yields, and the trick is for SMIC to use SMEE on the cutting edge, but that leads to the third point:
  • China has unlimited money and infinite motivation to figure this out.

Money is not a panacea: you can’t simply spend your way to faster chips, but instead must move down the learning curve on both the foundry and equipment level. Money does, though, pay for processes that don’t have great yields: the problem for Intel at 7 nanometer, for example, wasn’t that they couldn’t make chips, but that they couldn’t get yields high enough to make them economically. That won’t be a concern for China when it comes to chips for military applications.

What is more meaningful, though, will be the alignment of China’s private sector behind China’s chip companies: TSMC didn’t only need ASML, it also needed Apple and AMD and Nvidia, end users who were both willing to pay for performance and also work deeply with TSMC to figure out generation after generation of faster chips. Tencent and Alibaba and Baidu will now join Huawei in being the China chip industry’s most demanding customers, in the best possible sense.

China’s Trailing Edge

There is one more advantage China has: remember all of those old fabs that TSMC is still operating? It turns out that as more and more products incorporate microprocessors, trailing edge chips are exploding in demand. This was seen most clearly during the pandemic when U.S. automakers, who foolishly canceled their chip orders when the pandemic hit, suddenly found themselves at the back of the line as demand for basic chips skyrocketed.

In the end it was China that picked up a lot of the slack: the company’s commitment to building its own semiconductor industry is not a new one (just much more pressing), and part of the process of walking the path I detailed above is building more basic chips using older technologies. China’s share of >45 nanometer chips was 23% in 2019, and probably over 35% today; its share of 28-45 nanometer chips was 19% in 2019 and is probably approaching 30% today. Moreover, these chips still make up most of the volume for the industry as a whole: when you see charts like this, which measure market share by revenue, keep in mind that China has achieved 9% market share with low-priced chips:

China's increasing share of chips by revenue

The Biden administration’s sanctions are designed to not touch this part of the industry: the limitations are on high end fabs and the equipment and people that go into them, not trailing edge fabs that make up most of this volume. There is good reason for this: these trailing edge factories are still using a lot of U.S. equipment; for most equipment makers China is responsible for around a third of their revenue. That means cutting off trailing edge fabs would have two deleterious effects on the U.S.: a huge number of the products U.S. consumers buy would falter for lack of chips, even as the same U.S. companies that have built the advantage the administration is seeking to exploit would have their revenue (and future ability to invest in R&D) impaired.

It’s worth pointing out, though, that this is producing a new kind of liability for the U.S., and potentially more danger for Taiwan.

Go back to Intel’s strategy of selling off and/or reusing its old fabs, which again, made sense given the path Intel started on decades ago: that means that Intel, unlike TSMC, doesn’t have any trailing edge capacity (outside of what it acquired in the Tower Semiconductor deal). Global Foundries, the U.S.’s other foundry, had the same model as Intel while it was the manufacturing arm of AMD; Global Foundries acquired trailing edge capacity with its acquisition of Chartered Semiconductor, but there is a reason why the U.S. >45 nanometer market share was only 9% in 2019 (and likely lower today), and 28-45 nanometer market share was a mere 6% (and again, likely lower today).

Again, these aren’t difficult chips to make, but that is precisely why it makes little sense to build new trailing edge foundries in the U.S.: Taiwan already has it covered (with the largest marketshare in both categories), and China has the motivation to build more just so it can learn.

What, though, if TSMC were taken off the board?

Much of the discussion around a potential invasion of Taiwan — which would destroy TSMC (foundries don’t do well in wars) — centers around TSMC’s lead in high end chips. That lead is real, but Intel, for all of its struggles, is only 3~5 years behind. That is a meaningful difference in terms of the processors used in smartphones, high performance computing, and AI, but the U.S. is still in the game. What would be much more difficult to replace are, paradoxically, trailing node chips, made in fabs that Intel long ago abandoned.

China meanwhile, has had good reason to keep TSMC around, even as it built up its own trailing edge fabs: the country needs cutting edge chips, and TSMC makes them. However, if those chips are cut off, then what use is TSMC to China? This isn’t a new concern, by the way; I wrote after the U.S. imposed sanctions on Huawei:

I am, needless to say, not going to get into the finer details of the relationship between China and Taiwan (and the United States, which plays a prominent role); it is less that reasonable people may disagree and more that expecting reasonableness is probably naive. It is sufficient to note that should the United States and China ever actually go to war, it would likely be because of Taiwan.

In this TSMC specifically, and the Taiwan manufacturing base generally, are a significant deterrent: both China and the U.S. need access to the best chip maker in the world, along with a host of other high-precision pieces of the global electronics supply chain. That means that a hot war, which would almost certainly result in some amount of destruction to these capabilities, would be devastating…one of the risks of cutting China off from TSMC is that the deterrent value of TSMC’s operations is diminished.

My worry is that this excerpt didn’t go far enough: the more that China builds up its chip capabilities — even if that is only at trailing nodes — the more motivation there is to make TSMC a target, not only to deny the U.S. its advanced capabilities, but also the basic chips that are more integral to everyday life than we ever realized.

MAD Chips

So is this chip ban the right move?

In the medium term, the impacts will be significant, particularly in terms of the stated target of these sanctions — AI. Only now is it becoming possible to manufacture intelligence, and the means to do so is incredibly processor intensive, both in terms of quality and quantity. Moreover, not only does AI figure to loom large in military applications, but is also likely to spur innovation in its own right, perhaps even in terms of figuring out how to keep pushing the frontier of chip design.

In the long run, meanwhile, the U.S. may have given up what would have been, thanks to the sheer amount of cost and learning curve distance involved, a permanent economic advantage. Absent politics there simply is no reason to compete with TSMC or ASML or any of the other specialized parts of the supply chain; it would simply be easier to buy instead of build. Now, though, it is possible to envision a future where China undercuts U.S. companies in chips just like they once did in more labor-intensive industries, even as its own AI capabilities catch up and, given China’s demonstrated willingness to use technology in deeply intrusive ways, potentially surpass the West with its concerns about privacy and property rights.

The big question that I am raising in this article is the short run: while I have spent most of the last two years cautioning Americans who thought Taiwan was Thailand to not go from 0 to 100 in terms of the China threat, this move has in fact raised my concern level significantly. I am still, on balance, skeptical about a conflict, thanks in large part to how intertwined the U.S. and Chinese economies still are: any conflict would be mutually assured economic destruction.

Chips did, until three weeks ago, fall under the same paradigm; I wrote earlier this year in Tech and War:

This point applies to semiconductors broadly: as long as China needs U.S. technology or TSMC manufacturing, it is heavily incentivized to not take action against Taiwan; when and if China develops its own technology, whether now or many years from now, that deterrence is no longer a factor. In other words, the short-term and longer-term are in opposition to the medium-term…

There is no obvious answer, and it’s worth noting that the historical pattern — i.e. the Cold War — is a complete separation of trade and technology. That is one possible path, that we may fall into by default. It’s worth remembering, though, that dividers in the street are no way to live, and while most U.S. tech companies have flexed their capabilities, the most impressive tech of all is attractive enough and irreplaceable enough that it could still create dependencies that lead to squabbles but not another war.

Those dependencies are being severed; hopefully we still find sufficient reason to go no further than squabbles.

  1. The very first Nvidia chips were manufactured by SGS-Thomson Microelectronics, but have been manufactured by mostly TSMC from the original GeForce on 

Microsoft Full Circle

In last week’s interview with Stratechery, Microsoft CEO Satya Nadella explained why the company was open to partnering with Meta for VR:

The way I come at it, Ben, is that I like to separate out, “What is the system, what are the apps”? Of course, we want to bring the two things together where we can create magic, but at the same time, I also want our application experiences in particular to be available on all platforms, that’s very central to how our strategy is.

For example, when I think about the Metaverse, the first thing I think about is it’s not going to be born in isolation from everything else that’s in our lives, which is you’re going to have a Mac or a Windows PC, you’re going to have an iOS or an Android phone, and maybe you’ll have a headset. So if that is your life, how do we bring, especially Microsoft 365, all of the relationships that are set up, the work artifacts I’ve set up all to life in that ecosystem of devices? That’s at least how I come to it and that’s where when Mark started talking to us about his next generation stuff around Quest was pretty exciting, so it made a lot of sense for us to bring — whether it’s Teams with its immersive meetings experience to Quest or whether it’s even Windows 365 streaming, and then, of course, all our management and security and even Xbox — [to Quest]; that’s what is the motivation behind it.

This seems obvious today in 2022, but it was a fairly radical point of view when Nadella took over Microsoft in 2014. Nadella’s first event in April 2014 centered on the announcement of Microsoft’s iconic Office Suite on Apple’s iPad; the apps had been developed under former CEO Steve Ballmer, but had been withheld from launch until the company had touch-centric versions ready for Windows-based touch devices. From the beginning of Stratechery I was adamant that this was a major mistake driven by Microsoft’s inability to imagine a future without Windows at the center; from 2013’s Services, Not Devices:

The truth is that Microsoft is wrapping itself around an axle of its own creation. The solution to the secular collapse of the PC market is not to seek to prop up Windows and force an integrated solution that no one is asking for; rather, the goal should be the exact opposite. Maximum effort should be focused on making Office, Server, and all the other products less subservient to Windows and more in line with consumer needs and the reality of computing in 2013.

A drawing of The Horizontal Layer of Services

The trouble for Microsoft in the devices layer is that they only know horizontal domination. When there was nothing but PC’s, the insistence on one experience no matter the hardware worked perfectly. However, a Dell and an HP are much more similar than a tablet and a web page, for example, each of which has its own input method, user expectations, and constraints. A multi-device world demands bespoke experiences, not one size fits all. Microsoft simply doesn’t seem to understand that, and the longer they seek to “horizontalize” devices the greater the write-offs will become.

However, look again at that picture: there remains a horizontal layer — services — and it’s there that Microsoft should focus its energy. For Office and Server specifically:

  • Documents remain essential and ubiquitous to all of the world outside of Silicon Valley; an independent Office division should be delivering bespoke experiences on every meaningful platform. Office 365 is a great start that would be even better with a version for iPad.
  • A great many apps are simply front-ends for web-based services; an independent Server division should be delivering best-in-class interfaces and tools for app developers on every meaningful platform.

[…]“Devices and services” is only half right; unfortunately Ballmer picked the wrong half.

This is why it was so important that Office for iPad was Nadella’s first major announcement; I wrote after the event in When CEOs Matter:

This is the power CEOs have. They cannot do all the work, and they cannot impact industry trends beyond their control. But they can choose whether or not to accept reality, and in so doing, impact the worldview of all those they lead.

Four years later Nadella’s reworking of the culture was all but complete, as I wrote in The End of Windows:

The story of Windows’ decline is relatively straightforward and a classic case of disruption…What is more interesting, though, is the story of Windows’ decline in Redmond, culminating with last week’s reorganization that, for the first time since 1980, left the company without a division devoted to personal computer operating systems (Windows was split, with the core engineering group placed under Azure, and the rest of the organization effectively under Office 365; there will still be Windows releases, but it is no longer a standalone business).

This new reality couldn’t have been clearer at last week’s Microsoft Inspire worldwide partner conference: Nadella’s keynote was all about the cloud, from Azure to Teams; Windows was demoted to one section of the company’s Surface announcements held as a precursor to the main event.

Do More With Less

This is how Nadella opened his keynote:

We’re going through a period of historic economic, societal, and technological change. But for all the uncertainty we continue to see in the world, one thing is clear: organizations in every industry are turning to you and your digital capability to help them do more with less, so that they can navigate this change and emerge stronger. You are the change agents who make doing more with less possible. Less time, less cost, less complexity, with more innovation, more agility, and more resilience. Doing more with less doesn’t mean working harder or longer — it’s not going to scale — it means applying technology to amplify what you can do and ultimately what an organization can achieve amidst today’s constraints.

Over the past few years, we have talked extensively about digital transformation. But today we need to deliver on the digital imperative for every organization. It all comes down to how we can help you do this with the Microsoft cloud. No other cloud offers the best of category products, and the best of suite solutions, and that’s what we’ll focus on at Ignite this week as we walk through the five key imperatives.

This “do more with less” message recurred throughout Nadella’s presentation. Three separate times Nadella emphasized how much customers would save by going with a Microsoft bundle, but that was only the “with less” part of the message; each pitch also explained why the Microsoft approach was also better (i.e. “do more”). Start with security:

Protecting is complex and get expensive. Every organization experiences this with so many different devices, connections to partners, and an ever shifting cloud resource deployment. The more agile you become, the more your security team struggles to manage the risk; the more connected we become, the faster a successful attacker can move laterally through the enterprise to their target. For far too long customer have been forced to adopt multiple disconnected solutions from disparate sources that don’t integrate well and leave gaps. We offer a better option: a natively integrated security solution that is supported by a vibrant partner ecosystem…you get a comprehensive solution that closes gaps and works for you at machine speed. On average, customers save more than 60% when they turn to use compared to a multi-vendor solution.

Nadella’s argument: not only can you save money, but because all of the products come from one vendor you can rest assured that they are comprehensive and are designed to work together.

Now let’s turn to data: with our Microsoft Intelligent Data Platform we provide a complete data fabric, from the operational stores to the analytics engines to data governance so that you can spend more time creating value and less time integrating and managing your data estate. Our goal is to provide you with the most comprehensive end-to-end data platforms so you don’t have to wrestle with the complexities of building and operating cloud scale data infrastructure yourself. Analytics alone on our data intelligence platform cost up to 59% less than any other cloud analytics out there.

That bit about “spend more time creating value and less time integrating and managing” is the part of Microsoft’s value proposition that Silicon Valley startups so frequently miss. Slack, perhaps most famously, was so certain its superior chat experience would beat out Teams (and it is superior), that company CEO Stewart Butterfield took out an ad in the New York Times welcoming Microsoft to the space; four years later, after Teams had over six times the daily active users (and before Slack was acquired by Salesforce), I explained in Teams OS and the Slack Social Network what Butterfield got wrong:

This is what Slack — and Silicon Valley, generally — failed to understand about Microsoft’s competitive advantage: the company doesn’t win just because it bundles, or because it has a superior ground game. By virtue of doing everything, even if mediocrely, the company is providing a whole that is greater than the sum of its parts, particularly for the non-tech workers that are in fact most of the market. Slack may have infused its chat client with love, but chatting is a means to an end, and Microsoft often seems like the only enterprise company that understands that.

That end is, to use Nadella’s words, “creating value”; “integrating and managing” is exactly what companies want to avoid.

With Microsoft 365 we provide a complete cloud-first experience that makes work better for today’s digitally connected and distributed workforce. Customers can save more than 60% compared to a patchwork of solutions. Microsoft 365 includes Teams plus the apps you always relied on — Word, Excel, Powerpoint, and Outlook — as well as new applications for creation and expression like Loop, Clipchamp, Stream, and Designer, and it’s all built on the Microsoft graph, which makes available to you the information about people, their relationships, all their work artifacts, meetings, events, documents, in one interconnected system. Thanks to the graph you can understand how work is changing and how your digitally distributed workforce is working. This is so critical, and it all comes alive in the new Microsoft 365 application.

Ah, there are the Office applications I referenced at the beginning. But notice the word that is missing: Office.

From Office to Microsoft

From The Verge:

Microsoft is making a major change to its Microsoft Office branding. After more than 30 years, Microsoft Office is being renamed “Microsoft 365” to mark the software giant’s collection of growing productivity apps. While Office apps like Excel, Outlook, Word, and PowerPoint aren’t going away, Microsoft will now mostly refer to these apps as part of Microsoft 365 instead of Microsoft Office.

Microsoft has been pushing this new branding for years, after renaming Office 365 subscriptions to Microsoft 365 two years ago, but the changes go far deeper now. “In the coming months,, the Office mobile app, and the Office app for Windows will become the Microsoft 365 app, with a new icon, a new look, and even more features,” explains a FAQ from Microsoft. That means if you use any of the dedicated Office apps, they’ll all be branded with Microsoft 365 soon, and with a new logo. The first logo and design changes will appear at in November, followed by the Office app on Windows, iOS, and Android all getting rebranded in January.

I’ll be honest: as an increasingly old man in technology the end of the “Office” name kind of bums me out. My nostalgia is satisfied, though, by a Microsoft that has truly come full circle.

The truth about Microsoft is that while Windows’ relationship with hardware has traditionally been modular (the Surface line notwithstanding), the company’s strategy has always been about integration and bundling. This is why Ballmer was so hesitant to give up on Windows as the center of the company’s go-to-market: sure, people wanted the Office applications on different devices, but it was Windows that tied Office to Outlook to Exchange to Active Directory to Windows Server and on down the line. This, by extension, is why Nadella’s willingness to embrace reality was a risk: Office on its own was a nice business, but it wasn’t the center of enterprise like Windows had been.

It turned out, though, that facing reality brought another benefit: the ability to see and grasp an opportunity when it appeared. Teams, which started development in 2015, a year after Nadella’s announcement, wouldn’t simply be a chat app: it would be the new hub around which Office orbited. Teams (and Outlook) development leader Brian MacDonald said at a press event in 2019:

One of the really key things and drivers of what we wanted to do with Teams was have that be a hub for Office 365. Before what we had done was just taken all those personal productivity workloads and then moved them to the cloud, but we wanted something that was purpose-built for the cloud that could be a hub across all of Office and frankly across the rest of what we’re doing at Microsoft. A lot of the Power BI, Power Apps, and Dynamics tools that James was building, but also third party. So we built a platform for that and the third-party platform and the first-party platform are actually the same.

If that sounds a lot like Windows — a hub that hosted not just Office, but other Microsoft applications and services, and a platform for 3rd-party developers — Nadella agrees with you. From the same event:

Sometimes I think the new OS is not going to start from the hardware, because the classic OS definition, that Tanenbaum, one of the guys who wrote the book on Operating Systems that I read when I went to school was: “It does two things, it abstracts hardware, and it creates an app model”. Right now the abstraction of hardware has to start by abstracting all of the hardware in your life, so the notion that this is one device is interesting and important, it doesn’t mean the kernel that boots your device just goes away, it still exists, but the point of real relevance I think in our lives is “hey, what’s that abstraction of all the hardware in my life that I use?” – some of it is shared, some of it is personal. And then, what’s the app model for it? How do I write an experience that transcends all of that hardware? And that’s really what our pursuit of Microsoft 365 is all about.

Office being on its own gave Teams an easy go-to-market: Microsoft just bundled it in. Today, though, it is Teams and everything built on that scaffolding that is Microsoft’s new Windows. It is the company and its operating system, not its apps, that are back at the center. In this sense, renaming Office 365 to Microsoft 365 is the most natural thing in the world: Office was a ship that set sail from the declining civilization that was Windows, with an uncertain destination. Today, though, that ship is but a footnote in Microsoft’s new empire in the cloud.

Moreover, it seems likely this empire will be more durable than the old Microsoft republic: the entire reason why Windows faltered as a strategic linchpin is that it was tied to a device — the PC — that was disrupted by a paradigm shift in hardware. Microsoft 365, on the other hand, is attached to the customer. Nadella again:

What we are trying to do [with Microsoft 365] is bring home that notion that it’s about the user, the user is going to have relationships with other users and other people, they’re going to have a bunch of artifacts, their schedules, their projects, their documents, many other things, their to-do’s, and they are going to use a variety of different devices.

This is why Microsoft, instead of being late to the iPad, is remarkably early to VR. Why not? Devices are but mere conduits to the cloud, which means that Microsoft is well-placed to navigate this new paradigm if it becomes a major platform — and to not miss a beat if it is not.1 In other words, to say that Microsoft has come full circle may be selling Nadella’s transformation short: the all-encompassing dominant Microsoft of old may be back, but in a version that is even stronger and more resilient than before.

  1. This also, it must be said, casts doubt on Meta’s determination to go in the opposite direction, and give up its position as a user-centric service to be a hardware-dependent platform 

Meta Meets Microsoft

There is an easy to way to write this Article, and a hard way.

This weekend the easy way seemed within reach: I watched Meta’s Connect Keynote (I had early access in order to prepare for an interview with Meta CEO Mark Zuckerberg and Microsoft CEO Satya Nadella) and was, like apparently much of the Internet, extremely underwhelmed. Sure, the new Quest Pro looked cool, and I was very excited about the partnership with Microsoft (more on both in a moment); the presentation, though, was cringe, and seemed to lack any compelling demos of virtual reality.

What was particularly concerning was the entire first half of the keynote, which was primarily focused on consumer applications, including Horizon Worlds; Horizon Worlds was the the app The Verge reported was so buggy that Meta employees working on it barely used it, or more worryingly, was buggy because Meta employees couldn’t be bothered to dogfood it. The concerning part from the keynote was you could see why.

That was why this Article was going to be easy: writing that Meta’s metaverse wasn’t very compelling would slot right in to most people’s mental models, prompting likes and retweets instead of skeptical emails; arguing that Meta should focus on its core business would appeal to shareholders concerned about the money and attention devoted to a vision they feared was unrealistic. Stating that Zuckerberg got it wrong would provide comfortable distance from not just an interview subject but also a company that I have defended in its ongoing dispute with Apple over privacy and advertising.

Indeed, you can sense my skepticism in the most recent episode of Sharp Tech, which was recorded after seeing the video but before trying the Quest Pro. See, that was the turning point: I was really impressed, and that makes this Article much harder to write.

Meetings in VR

I wrote about virtual reality and the Metaverse a number of times last year, including August’s Metaverses, Meta’s keynote and name-change in October, and Microsoft and the Metaverse in November. The most important post though, at least in terms of my conception of the space, was this August Update about Horizon Workrooms (not to be confused with the aforementioned Horizon Worlds):

My personal experience with Workrooms didn’t involve any dancing or fitness; it was simply a conversation with the folks that built Workrooms. The sense of presence, though, was tangible. Voices came from the right place, thanks to Workrooms’ spatial audio, and hand gestures and viewing directions really made it feel like the three of us were in the same room. What was particularly compelling was the way that Workrooms’ virtual reality space seamlessly interfaced with the real world…

I don’t want to go too far given I’ve only tried out Workrooms once, but this feels like something real. And, just as importantly, there is, thanks to COVID, a real use case. Of course companies will need to be convinced, and hardware will need to be bought, but that’s another reason why the work angle is so compelling: companies are willing to pay for tools that increase productivity to a much greater extent than consumers are.

I don’t have much of a company, but I did buy Quest 2’s for the Passport team, and we held one meeting a week in Workrooms. One in particular stands out to me: we made a major decision about the product, and my memory of that decision does not involve me sitting at my desk in Taiwan, but of being in that virtual room. The sense of place and presence was that compelling.

Then one of the developers moved house, temporarily misplaced his headset, and we haven’t used it since.

Microsoft’s Advantage

It was my experience with Workrooms that undergirded my argument that Microsoft was the best placed to succeed with virtual reality. Yes, virtual reality entails putting on a headset and leaving your current environment for a virtual one, but that is not so different from leaving your house and going to the office. Moreover, Microsoft’s shift to Teams as its de facto OS meant it was well-placed to deliver company-specific metaverses:

This integration, though, also means that Microsoft has a big head start when it comes to the Metaverse: if the initial experience of the Metaverse is as an individual self-contained metaverse with its own data and applications, then Teams is already there. In other words, not only is enterprise the most obvious channel for virtual reality from a hardware perspective, but Teams is the most obvious manifestation of virtual reality’s potential from a software perspective.

The shortcoming was hardware: Microsoft had the HoloLens, but that was an augmented reality device. Continuing from that Article:

What is not integrated is the hardware; Microsoft sells a number of third party VR headsets on said webpage, all of which have to be connected to a Windows computer. Microsoft’s success will require creating an opportunity for OEMs similar to the opportunity that was created by the PC. At the same time, this solution is also an advantageous one for the long-term Metaverse-as-Internet vision: Windows is the most open of the consumer platforms, and that applies to Microsoft’s current implementation of VR. The company would do well to hold onto this approach.

This Article seems quite prescient given the announcement that Microsoft is partnering with Meta going forward: Microsoft is bringing its Teams-based ecosystem to Quest, along with enterprise tools like Azure Active Directory and Intune device management, with Xbox Game Pass thrown in for good measure. In doing so Microsoft gets to piggy-back on Meta’s massive investments in hardware.

It’s difficult to overstate what a massive win this feels like for Microsoft: the company will have a privileged position on what is for now the most advanced headset with the most resources behind it, not because it is paying for the privilege but because it is the most obvious go-to-market for this new technology. I argued in that Article that VR adoption would probably look more like the PC than it did smartphones:

Implicit in assuming that augmented reality is more important than virtual reality is assuming that this new way of accessing the Internet will develop like mobile did. Smartphone makers like Apple, though, had a huge advantage: people already had and wanted mobile phones; selling a device that you were going to carry anyway, but which happened to be infinitely more capable for only a few hundred more dollars, was a recipe for success in the consumer market.

PCs, though, didn’t have that advantage: the vast majority of the consumer market had no knowledge of or interest in computers; rather, most people encountered computers for the first time at work. Employers bought their employees computers because computers made them more productive; then, once consumers were used to using computers at work, an ever increasing number of them wanted to buy a computer for their home as well. And, as the number of home computers increased, so did the market opportunity for developers of non-work applications like games.

I suspect that this is the path that virtual reality will take. Like PCs, the first major use case will be knowledge workers using devices bought for them by their employer, eager to increase collaboration in a remote work world, and as quality increases, offer a superior working environment. Some number of those employees will be interested in using virtual reality for non-work activities as well, increasing the market for non-work applications.

This is still my position, and my experience with the Quest Pro only confirmed it. A lot of the new functionality is very impressive: the facial expression detection really works, and the new controllers are shockingly precise (one highlight was playing Operation and feeling like you actually had a chance). The best demo of all, though, was the updated version of Workrooms; the product manager showing me the new features was actually in Austin, Texas (I was in California), but that was the only demo where I actually forgot about the people in the room with me: mentally I was truly in a virtual meeting room.

At the same time, this was a $1,500 device. On one hand, that was actually cheaper than I expected; on the other hand, it still has significant weaknesses: it’s heavy, the battery life is only an hour or two, and the resolution is still disappointingly low. I’m not sure I can justify buying it for my Passport team, even if we were still going through the hassle of pulling on a headset for one meeting a week.

Meta’s Outlook

So where does this leave Meta?

First off, while Quest Pro is a definite leap forward, we still seem a few years away from a device that is truly ready for mass consumption. I noted the big issues above: weight, battery life, and resolution.

A bigger concern I have is software: the most compelling use case is meetings, and that matters for a market — enterprise — that is not only not Meta’s primary focus but that the company is effectively outsourcing to Microsoft. This raises a further strategic concern: the inverse of Microsoft winning by virtue of Meta spending billions on hardware to make Microsoft VR software compelling is that Meta runs the risk of being the IBM to Microsoft’s DOS. Zuckerberg admitted the risks but argued the benefits outweighed them:

I think the pros of the partnership way outweigh the risks. Obviously nothing is risk-free, but at the end of the day, we also have to do our job and deliver world-class services and hardware. If we don’t do that, then obviously we will lose. I do think though that there is this alignment that we talked about before between the things that we primarily care about, which are the aspects of the platform around expression…

I think basically everyone else in the space would focus more on the the single-player experience. Our bet in this is a deep bet that the connection aspect matters more and this has been part of the experience of running the company all along, is that even just growing up we’re told, “Do your homework, then go play with your friends.” I just think at some level that’s wrong. The connection between people is the point, not the thing that you do after everything else.

I do get the vision: while meetings have obvious utility, if you go back to the broader Metaverse vision the idea of, say, watching a basketball game courtside with my friends, despite the fact we are scattered all over the world and no where near a stadium, is a really compelling one. Social experiences in VR, though, require everyone involved to have a compatible VR headset.

Social media on your computer or phone isn’t like this: one of the reasons why an ad model is so compatible with social media is because it enables the service to be free, which makes it much more plausible that your friends are on the platform. Introducing even the slightest barrier to entry — much less a several hundred dollar one — makes it much less likely that a multiplayer experience is even possible.

This by extension means the timing question is an even more daunting one for Meta in particular. I do think that VR has real utility, but it will take time for that utility to be accessible on a cost-effective basis for enterprises and individual users; meaningful social experiences will take longer yet, simply because social experiences depend on the people you want to hang out with being bought in as well.1

This then is where I stand on VR and the metaverse concept, on Meta’s one-year anniversary:

  • VR does have real utility, but I think that utility will be realized in the enterprise first, in part because the value of VR only becomes apparent when you use it, and you’re more likely to use it if your company pays for it (VR really doesn’t demo well, as yesterday’s presentation showed).
  • Microsoft is well-placed to deliver that utility on top of Meta hardware.
  • Meta is likely to be the catalyst for VR becoming a widely used technology but it is much more uncertain as to whether the company will capture sufficient value to justify its massive investment, thanks in part to its focus on social networking.

I very well might be wrong on one or all of these points, in either direction. On one hand, maybe most people will never buy into the idea of putting a headset on; on the other hand, maybe Meta will figure out a go-to-market strategy that somehow communicates VR presence in a way that gets people to buy headsets in sufficient mass to make social experiences viable.

What is clear is that Zuckerberg in particular seems more committed to VR than ever. It may be the case that he is seen as the founding father of the Metaverse, even as Meta is a potential casualty.

You can read an interview I conducted with Zuckerberg and Nadella about yesterday’s announcement here.

  1. Meta’s argument would be that these social experiences will also be accessible via your smartphone or PC, but that means giving up on presence, which is VR’s killer feature 

Nvidia In the Valley

Nvidia investors have been in the valley before:

A drop in Nvidia's stock price

This chart, though, is not from the last two years, but rather from the beginning of 2017 to the beginning of 2019; here is 2017 to today:

Nvidia's current stock price drop

Three big things happened to Nvidia’s business over the last three years that drove the price to unprecedented heights:

  • The pandemic led to an explosion in PC buying generally and gaming cards specifically, as customers had both the need for new computers and a huge increase in discretional income with nowhere to spend it beyond better game experiences.
  • Machine learning applications, which were trained on Nvidia GPUs, exploded amongst the hyperscalers.
  • The crypto bubble led to skyrocketing demand for Nvidia chips to solve Ethereum proof-of-work equations to earn — i.e. mine — Ether.

Crypto isn’t so much a valley as it is a cliff: Ethereum successfully transitioned to a proof-of-stake model, rendering entire mining operations, built with thousands of Nvidia GPUs, worthless overnight; given that Bitcoin, the other major crypto network to use proof-of-work, is almost exclusively mined on custom-designed chips, all of those old GPUs are flooding the second-hand market. This is particularly bad timing for Nvidia given that the pandemic buying spree ended just as the company’s attempt to catch up on demand for its 3000-series of chips were coming to fruition. Needless to say, too much new inventory plus too much used inventory is terrible for a company’s financial results, particularly when you’re trying to clear the channel for a new series:

Nvidia's gaming revenue drop

Nvidia CEO Jensen Huang told me in a Stratechery Interview last week that the company didn’t see this coming:

I don’t think we could have seen it. I don’t think I would’ve done anything different, but what I did learn from previous examples is that when it finally happens to you, just take the hard medicine and get it behind you…We’ve had two bad quarters and two bad quarters in the context of a company, it’s frustrating for all the investors, it’s difficult on all the employees.

We’ve been here before at Nvidia.

We just have to deal with it and not be overly emotional about it, realize how it happened, keep the company as agile as possible. But when the facts presented itself, we just made cold, hard decisions. We took care of our partners, we took care of our channel, we took care of making sure that everybody had plenty of time. By delaying Ada, we made sure that everybody had plenty of time for and we repriced all the products such that even in the context of Ada, even if Ada were available, the products that after it’s been repriced is actually a really good value. I think we took care of as many things as we could, it resulted in two fairly horrific quarters. But I think in the grand scheme of things, we’ll come right back so I think that was probably the lessons from the past.

This may be a bit generous; analysts like Tae Kim and Doug O’Laughlin forecast the stock plunge earlier this year, although that was probably already too late to avoid this perfect storm of slowing PC sales and Ethereum’s transition, given that Nvidia ordered all of those extra 3000-series GPUs in the middle of the pandemic (Huang also cited the increasing lead times for chips as a big reason why Nvidia got this so wrong).

What is more concerning for Nvidia, though, is that while its inventory and Ethereum issues are the biggest drivers of its “fairly horrific quarters”, that is not the only valley its gaming business is navigating. I’m reminded of John Bunyan’s Pilgrim’s Progress:

Now Christian had not gone far in this Valley of Humiliation before he was severely tested, for he noticed a very foul fiend coming over the field to meet him; his name was Apollyon [Destroyer].

Call Apollyon inventory issues; Christian defeated him, as Nvidia eventually will.

Now at the end of this valley there was another called the Valley of the Shadow of Death; and it was necessary for Christian to pass through it because the way to the Celestial City was in that direction. Now this valley was a very solitary and lonely place. The prophet Jeremiah describes it as, “A wilderness, a land of deserts and of pits, a land of drought and of the shadow of death, a land that no man” (except a Christian) “passes through, and where no man dwells.”

What was striking about Nvidia’s GTC keynote last week was the extent to which this allegory seems to fit Nvidia’s ambitions: the company is setting off on what appears to be a fairly solitary journey to define the future of gaming, and it’s not clear when or if the rest of the industry will come along. Moreover, the company is pursuing a similarly audacious strategy in the data center and with its metaverse ambitions as well: in all three cases the company is pursuing heights even greater than those achieved over the last two years, but the path is surprisingly uncertain.

Gaming in the Valley: Ray-Tracing and AI

The presentation of 3D games has long depended on a series of hacks, particularly in terms of lighting. First, a game determines what you actually see (i.e. there is no use rendering an object that is occluded by another); then the correct texture is applied to the object (i.e. a tree, or grass, or whatever else you might imagine). Finally light is applied based on the position of a pre-determined light source, with a shadow map on top of that. The complete scene is then translated into individual pixels and rendered onto your 2D screen; this process is known as rasterization.

Ray tracing handles light completely differently: instead of starting with a pre-determined light source and applying light and shadow maps, ray tracing starts with your eye (or more precisely, the camera through which you are viewing the scene). It then traces the line of sight to every pixel on the screen, bounces it off that pixel (based on what type of object it represents), and continues following that ray until it either hits a light source (and thus computes the lighting) or discards it. This produces phenomenally realistic lighting, particularly in terms of reflections and shadows. Look closely at these images from PC Magazine:

Let’s see how ray tracing can visually improve a game. I took the following screenshot pairs in Square Enix’s Shadow of the Tomb Raider for PC, which supports ray-traced shadows on Nvidia GeForce RTX graphics cards. Specifically, look at the shadows on the ground.

An image with rasterized lighting
Rasterized shadows
Ray-traced shadows
Ray-traced shadows

[…]The ray-traced shadows are softer and more realistic compared with the harsher rasterized versions. Their darkness varies depending on how much light an object is blocking and even within the shadow itself, while rasterization seems to give every object a hard edge. The rasterized shadows still don’t look bad, but after playing the game with ray traced shadows, it’s tough to go back.

Nvidia first announced API support for ray tracing back in 2009; however, few if any games used it because it is so computationally expensive (ray-tracing is used in movie CGI; however, those scenes can be rendered over hours or even days; games have to be rendered in real-time). That is why Nvidia introduced dedicated ray tracing hardware in its GeForce 2000-series line of cards (which were thus christened “RTX”) which came out in 2018. AMD went a different direction, adding ray-tracing capabilities to its core shader units (which also handle rasterization); this is slower than Nvidia’s pure hardware solution, but it works, and, importantly, since AMD makes graphics cards for the PS5 and Xbox, it means that ray tracing support is now industry-wide. More and more games will support ray tracing going forward, although most applications are still fairly limited because of performance concerns.

Here’s the important thing about ray tracing, though: by virtue of calculating light dynamically, instead of via light and shadow maps, it is something developers can get “for free.” A game or 3D environment that depended completely on ray tracing should be easier and cheaper to develop; more importantly, it means that environments could change in dynamic ways that the developer never anticipated, all while having more realistic lighting than the most labored-over pre-drawn environment.

This is particularly compelling in two emerging contexts: the first is in simulation games like Minecraft. With ray tracing it will be increasingly realistic to have highly detailed 3D-worlds that are constructed on the fly and lit perfectly. Future games could go further: the keynote opened with a game called RacerX where every single part of the game was fully simulated, including objects; the same sort of calculations for light were used for in-game physics as well.

The second context is a future of AI-generated content I discussed in DALL-E, the Metaverse, and Zero Marginal Cost Content. All of those textures I noted above are currently drawn by hand; as graphical capabilities — largely driven by Nvidia — have increased, so has the cost of creating new games, thanks to the need to create high resolution assets. One can imagine a future where asset creation is fully automated and done on the fly, and then lit appropriately via ray tracing.

For now, though Nvidia is already using AI to render images: the company also announced version 3 of its Deep Learning Super Sampling (DLSS) technology, which predicts and pre-renders frames, meaning they don’t have to be computed at all (previous versions of DLSS predicted and pre-rendered individual pixels). Moreover, Nvidia is, as with ray-tracing, backing up DLSS with dedicated hardware to make it much more performant. These new approaches, matched with dedicated cores on Nvidia’s GPUs, make Nvidia very well-placed for an entirely new paradigm in not just gaming but immersive 3D experiences generally (like a metaverse).

Here’s the problem, though: all of that dedicated hardware comes at a cost. Nvidia’s new GPUs are big chips — the top-of-the-line AD102, sold as the RTX 4090, is a fully integrated system-on-a-chip that measures 608.4mm2 on TSMCs N4 process;1 the top-of-the-line Navi 31 chip in AMD’s upcoming RDNA 3 graphics line, in comparison, is a chiplet design with a 308mm2 graphics chip on TSMC’s N5 process,2 plus six 37.5mm2 memory chips on TSMC’s N6 process.3 In short, Nvidia’s chip is much larger (which means much more expensive), and it’s on a slightly more modern process (which likely costs more). Dylan Patel explains the implications at SemiAnalysis:

In short, AMD saves a lot on die costs by forgoing AI and ray tracing fixed function accelerators and moving to smaller dies with advanced packaging. The advanced packaging cost is up significantly with AMD’s RDNA 3 N31 and N32 GPUs, but the small fan-out RDL packages are still very cheap relative to wafer and yield costs. Ultimately, AMD’s increased packaging costs are dwarfed by the savings they get from disaggregating memory controllers/infinity cache, utilizing cheaper N6 instead of N5, and higher yields…Nvidia likely has a worse cost structure in traditional rasterization gaming performance for the first time in nearly a decade.

This is the valley that Nvidia is entering. Gamers were immediately up-in-arms after Nvidia’s keynote because of the 4000-series’ high prices, particularly when the fine print on Nvidia’s website revealed that one of the tier-two chips Nvidia announced was much more akin to a rebranded tier-3 chip, with the suspicion being that Nvidia was playing marketing games to obscure a major price increase. Nvidia’s cards may have the best performance, and are without question the best placed for a future of ray tracing and AI-generated content, but at the cost of being the best values for games as they are played today. Reaching the heights of purely simulated virtual worlds requires making it through a generation of charging for capabilities that most gamers don’t yet care about.

AI in the Valley: Systems, not Chips

One reason to be optimistic about Nvidia’s approach in gaming is that the company made a similar bet on the future when it invented shaders; I explained shaders after last year’s GTC in a Daily Update:

Nvidia first came to prominence with the Riva and TNT line of video cards that were hard-coded to accelerate 3D libraries like Microsoft’s Direct3D:

The GeForce line, though, was fully programmable via a type of computer program called a “shader” (I explained more about shaders in this Daily Update). This meant that a GeForce card could be improved even after it was manufactured, simply by programming new shaders (perhaps to support a new version of Direct3D, for example).

[…]More importantly, shaders didn’t necessarily need to render graphics; any sort of software — ideally programs with simple calculations that could be run in parallel — could be programmed as shaders; the trick was figuring out how to write them, which is where CUDA came in. I explained in 2020’s Nvidia’s Integration Dreams:

This increased level of abstraction meant the underlying graphics processing unit could be much simpler, which meant that a graphics chip could have many more of them. The most advanced versions of Nvidia’s just-announced GeForce RTX 30 Series, for example, has an incredible 10,496 cores.

This level of scalability makes sense for video cards because graphics processing is embarrassingly parallel: a screen can be divided up into an arbitrary number of sections, and each section computed individually, all at the same time. This means that performance scales horizontally, which is to say that every additional core increases performance. It turns out, though, that graphics are not the only embarrassingly parallel problem in computing…

This is why Nvidia transformed itself from a modular component maker to an integrated maker of hardware and software; the former were its video cards, and the latter was a platform called CUDA. The CUDA platform allows programmers to access the parallel processing power of Nvidia’s video cards via a wide number of languages, without needing to understand how to program graphics.

Now the Nvidia “stack” had three levels:

The important thing to understand about CUDA, though, is that it didn’t simply enable external programmers to write programs for Nvidia chips; it enabled Nvidia itself.

Much of this happened out of desperation; Huang explained in a Stratechery interview last spring that introducing shaders, which he saw as essential for the future, almost killed the company:

The disadvantage of programmability is that it’s less efficient. As I mentioned before, a fixed function thing is just more efficient. Anything that’s programmable, anything that could do more than one thing just by definition carries a burden that is not necessary for any particular one task, and so the question is “When do we do it?” Well, there was also an inspiration at the time that everything looks like OpenGL Flight Simulator. Everything was blurry textures and trilinear mipmapped, and there was no life to anything, and we felt that if you didn’t bring life to the medium and you didn’t allow the artist to be able to create different games and different genres and tell different stories, eventually the medium would cease to exist. We were driven by simultaneously this ambition of wanting to create a more programmable palette so that the game and the artist could do something great with it. At the same time, we also were driven to not go out of business someday because it would be commoditized. So somewhere in that kind of soup, we created programmable shaders, so I think the motivation to do it was very clear. The punishment afterwards was what we didn’t expect.

What was that?

Well, the punishment is all of a sudden, all the things that we expected about programmability and the overhead of unnecessary functionality because the current games don’t need it, you created something for the future, which means that the current applications don’t benefit. Until you have new applications, your chip is just too expensive and the market is competitive.

Nvidia survived because their ability to do direct acceleration was still the best; it thrived in the long run because they took it upon themselves to build the entire CUDA infrastructure to leverage shaders. This is where that data center growth comes from; Huang explained:

On the day that you become processor company, you have to internalize that this processor architecture is brand new. There’s never been a programmable pixel shader or a programmable GPU processor and a programming model like that before, and so we internalize. You have to internalize that this is a brand new programming model and everything that’s associated with being a program processor company or a computing platform company had to be created. So we had to create a compiler team, we have to think about SDKs, we have to think about libraries, we had to reach out to developers and evangelize our architecture and help people realize the benefits of it, and if not, even come close to practically doing it ourselves by creating new libraries that make it easy for them to port their application onto our libraries and get to see the benefits of it.

The first reason to recount this story is to note the parallels between the cost of shader complexity and the cost of ray tracing and AI in terms of current games; the second is to note that Nvidia’s approach to problem-solving has always been to do everything itself. Back then that meant developing CUDA for programming those shaders; today it means building out entire systems for AI.

Huang said during last week’s keynote:

Nvidia is dedicated to advancing science and industry with accelerated computing. The days of no-work performance scaling are over. Unaccelerated software will no longer enjoy performance scaling without a disproportionate increase in costs. With nearly three decades of a singular focus, Nvidia is expert at accelerating software and scaling computer by a 1,000,000x, going well beyond Moore’s Law.

Accelerated computing is a full-stack challenge. It demands deep understanding of the problem domain, optimizing across every layer of computing, and all three chips: CPU, GPU, and DPU. Scaling across multi-GPUs on multi-nodes is a datacenter-scale challenge, and requires treating the network and storage as part of the computing fabric, and developers and customers want to run software in many places, from PCs to super-computing centers, enterprise data centers, cloud, to edge. Different applications want to run in different locations, and in different ways.

Today, we’re going to talk about accelerated computing across the stack. New chips and how they will boost performance, far beyond the number of transistors, new libraries, and how it accelerates critical workloads to science and industry, new domain-specific frameworks, to help develop performant and easily deployable software. And new platforms, to let you deploy your software securely, safely, and with order-of-magnitude gains.

In Huang’s view, simply having fast chips is no longer enough for the workloads of the future: that is why Nvidia is building out entire data centers using all of its own equipment. Here again, though, a future where every company needs accelerated computing generally, and Nvidia to build it for them specifically — Nvidia’s Celestial City — is in contrast to the present where the biggest users of Nvidia chips in the data center are hyperscalers who have their own systems already in place.

A company like Meta, for example, doesn’t need Nvidia’s networking; they invented their own. What they do need are a lot of massively parallelizable chips to train their machine learning algorithms on, which means they have to pay Nvidia and their high margins. Small wonder that Meta, like Google before them, is building its own chip.

This is the course that all of the biggest companies will likely follow: they don’t need an Nvidia system, they need a chip that works in their system for their needs. That is why Nvidia is so invested in the democratization of AI and accelerated computing: the long term key to scale will be in building systems for everyone but the largest players. The trick to making it through the valley will be in seeing that ecosystem develop before Nvidia’s current big customers stop buying Nvidia’s expensive chips. Huang once saw that 3D accelerators would be commoditized and took a leap with shaders; one gets the sense he has the same fear with chips and is thus leaping into systems.

Metaverse in the Valley: Omniverse Nucleus

In the interview last spring I asked Huang if Nvidia would ever build a cloud service;

If we ever do services, we will run it all over the world on the GPUs that are in everybody’s clouds, in addition to building something ourselves, if we have to. One of the rules of our company is to not squander the resources of our company to do something that already exists. If something already exists, for example, an x86 CPU, we’ll just use it. If something already exists, we’ll partner with them, because let’s not squander our rare resources on that. And so if something already exists in the cloud, we just absolutely use that or let them do it, which is even better. However, if there’s something that makes sense for us to do and it doesn’t make for them to do, we even approach them to do it, other people don’t want to do it then we might decide to do it. We try to be very selective about the things that we do, we’re quite determined not to do things that other people do.

It turns out there was something no one else wanted to do, and that was create a universal database for 3D objects for use in what Nvidia is calling the Omniverse. These objects could be highly detailed millimeter-precise objects for use in manufacturing or supply chains, or they could be fantastical objects and buildings generated for virtual worlds; in Huang’s vision they would be available to anyone building on Omniverse Nucleus.

Here the Celestial City is a world of 3D experiences used across industry and entertainment — an Omniverse of metaverses, if you will, all connected to Nvidia’s cloud — and it’s ambitious enough to make Mark Zuckerberg blush! This valley, by the same token, seems even longer and darker: not only do all of these assets and 3D experiences need to be created, but entire markets need to be convinced of their utility and necessity. Building a cloud for a world that doesn’t yet exist is to reach for heights still out of sight.

There certainly is no questioning Huang and Nvidia’s ambition, although some may quibble with the wisdom of navigating three valleys all at once; it’s perhaps appropriate that the stock is in a valley itself, above and beyond that perfect storm in gaming.

What is worth considering, though, is that the number one reason why Nvidia customers — both in the consumer market and the enterprise one — get frustrated with the company is price: Nvidia GPUs are expensive, and the company’s margins — other than the last couple of quarters — are very high. Pricing power in Nvidia’s case, though, is directly downstream from Nvidia’s own innovations, both in terms of sheer performance in established workloads, and also in its investment in the CUDA ecosystem creating the tools for entirely new ones.

In other words, Nvidia has earned the right to be hated by taking the exact sort of risks in the past it is embarking on now. Suppose, for example, the expectation for all games in the future is not just ray tracing but full-on simulation of all particles: Nvidia’s investment in hardware will mean it dominates the era just as it did the rasterized one. Similarly, if AI applications become democratized and accessible to all enterprises, not just the hyperscalers, then it is Nvidia who will be positioned to pick up the entirety of the long tail. And, if we get to a world of metaverses, then Nvidia’s head start on not just infrastructure but on the essential library of objects necessary to make that world real (objects that will be lit by ray-tracing in AI-generated spaces, of course), will make it the most essential infrastructure in the space.

These bets may not all pay off; I do, though, appreciate the audacity of the vision, and won’t begrudge the future margins that may result in the Celestial City if Nvidia makes it through the valley.

  1. TSMC’s 3rd generation 5nm process 

  2. TSMC’s 1st generation 5nm process 

  3. TSMC’s 3rd generation 7nm process 

Sharp Tech and Stratechery Plus

I am excited to announce both a new podcast and a substantial expansion in the value of a Stratechery subscription. We’ll start with the podcast:

Sharp Tech with Ben Thompson

Sharp Tech with Ben Thompson is a new podcast from Andrew Sharp and myself about how technology works, and the ways it is impacting the world. We will publish one free episode weekly, and there are already six episodes in the catalog:

In addition, there will be a weekly subscriber-only episode that will be built on listener questions and feedback; the first paid episode dropped yesterday. You can get Sharp Tech for Apple Podcasts, Overcast, or the podcast player of your choice by loging in at the Sharp Tech website, or search for it in Spotify.

Here is the good news: Sharp Tech Premium is included with a Stratechery subscription.

That leads to today’s second announcement: the Stratechery Daily Update subscription is transforming into Stratechery Plus:

Stratechery Plus

Stratechery Plus is the same $12/month or $120/year price as the Stratechery Update, but it is now expanded to include not just the Stratechery Update and Stratechery Interviews but also Dithering and Sharp Tech.

Stratechery Plus includes the Stratechery Update, Stratechery Interviews, Sharp Tech, and Dithering

The Stratechery Update consists of substantial analysis of the news of the day delivered via three weekly emails or podcasts (including free bi-weekly Stratechery Articles). If you enjoy Stratechery Articles you will love the Stratechery Update.

Stratechery Interviews include interviews with leading public CEOs like Mark Zuckerberg, Jensen Huang, and Satya Nadella; the Founder Series with private company founders like Parker Conrad, Laura Behrens Wu, and Shishir Mehrotra; and discussions with fellow analysts like Eric Seufert, Matthew Ball, and Bill Bishop.

Dithering is a twice-weekly podcast from Daring Fireball’s John Gruber and myself: 15 minutes an episode, not a minute less, not a minute more. Dithering, which costs $5/month, was previously available as a $3 add-on for Stratechery subscribers; now it is available to all Stratechery subscribers. You can get Dithering for Apple Podcasts, Overcast, or the podcast player of your choice by logging in at the Dithering website.

This is, I hope, only the beginning for Stratechery Plus. Right now the content is obviously very Ben-centric, but my hope is to expand the offering over time. For now, I am delighted to be doing my part to make Stratechery more valuable than ever.

The AI Unbundling

My first job was as a paper boy:

Paper boy, by Midjourney

The job was remarkably analog: a bundle of newspapers would be dropped off at my house, I would wrap them in rubber-bands (or plastic bags if it were raining), load them up in a canvas sack, and set off on my bike; once a month my parents would drive me around to collect payment. Little did I appreciate just how essential my role writ large was to the profitability of newspapers generally.

Newspapers liked to think that they made money because people relied on them for news, furnished by their fearless reporters and hard-working editors; not only did people pay newspapers directly, but advertisers were also delighted to pay for the privilege of having their products placed next to the journalists’ peerless prose. The Internet revealed the fatal flaw in this worldview: what newspapers provided was distribution thanks to infrastructure like printing presses and yours truly.

Printing press, by Midjourney

Once the Internet reduced distribution costs to zero, three truths emerged: first, that “news”, once published, retained no economic value. Second, newspapers no longer had geographic monopolies, but were instead in competition with every publication across the globe. Third, advertisers didn’t care about content, but rather about reaching customers.

I illustrated these three truths in 2015’s Popping the Publishing Bubble:

Popping the Publishing Bundle

Editorial and ads used to be a bundle; next, the Internet unbundled editorial and ads, and provided countless options for both; the final step was ads moving to platforms that gave direct access to users, leaving newspapers with massive reach and no way to monetize it.

The Idea Propagation Value Chain

As much as newspapers may rue the Internet, their own business model — and my paper delivery job — were based on an invention that I believe is the only rival for the Internet’s ultimate impact: the printing press. Those two inventions, though, are only two pieces of the idea propagation value chain. That value chain has five parts:

The five parts of the idea propagation value chain: creation, substantiation, duplication, distribution, consumption

The evolution of human communication has been about removing whatever bottleneck is in this value chain. Before humans could write, information could only be conveyed orally; that meant that the creation, vocalization, delivery, and consumption of an idea were all one-and-the-same. Writing, though, unbundled consumption, increasing the number of people who could consume an idea.

Writing unbundled consumption from the rest of the value chain

Now the new bottleneck was duplication: to reach more people whatever was written had to be painstakingly duplicated by hand, which dramatically limited what ideas were recorded and preserved. The printing press removed this bottleneck, dramatically increasing the number of ideas that could be economically distributed:

The printing press unbundled duplication and distribution from creation

The new bottleneck was distribution, which is to say this was the new place to make money; thus the aforementioned profitability of newspapers. That bottleneck, though, was removed by the Internet, which made distribution free and available to anyone.

The Internet unbundled distribution from duplication

What remains is one final bundle: the creation and substantiation of an idea. To use myself as an example, I have plenty of ideas, and thanks to the Internet, the ability to distribute them around the globe; however, I still need to write them down, just as an artist needs to create an image, or a musician needs to write a song. What is becoming increasingly clear, though, is that this too is a bottleneck that is on the verge of being removed.

A flood emerging from a door ajar, by Midjourney

This image, like the first two in this Article, was created by AI (Midjourney, specifically). It is, like those two images, not quite right: I wanted “A door that is slightly open with light flooding through the crack”, but I ended up with a door with a crack of light down the middle and a literal flood of water; my boy on a bicycle, meanwhile, is missing several limbs, and his bike doesn’t have a handlebar, while the intricacies of the printing press make no sense at all.

They do, though, convey the idea I was going for: a boy delivering newspapers, printing presses as infrastructure, and the sense of being overwhelmed by the other side of an opening door — and they were all free.1 To put in terms of this Article, I had the idea, but AI substantiated it for me — the last bottleneck in the idea propagation value chain is being removed.

AI Democratization

What is notable about all of these AI applications it that they go back to language itself; Roon writes with regards to large language models (LLMs) on the Scale blog:

In a previous iteration of the machine learning paradigm, researchers were obsessed with cleaning their datasets and ensuring that every data point seen by their models is pristine, gold-standard, and does not disturb the fragile learning process of billions of parameters finding their home in model space. Many began to realize that data scale trumps most other priorities in the deep learning world; utilizing general methods that allow models to scale in tandem with the complexity of the data is a superior approach. Now, in the era of LLMs, researchers tend to dump whole mountains of barely filtered, mostly unedited scrapes of the Internet into the eager maw of a hungry model.

Roon’s focus is on text as the universal input, and connective tissue.2 Note how this insight fits into the overall development of communication: oral communication was a prerequisite to writing and reading; widespread literacy was a prerequisite to anyone being able to publish on the Internet; the resultant flood of text and images enabled by zero marginal distribution is the prerequisite for models that unbundle the creation of an idea and its substantiation.

This, by extension, hints at an even more surprising takeaway: the widespread assumption — including by yours truly — that AI is fundamentally centralizing may be mistaken. If not just data but clean data was presumed to be a prerequisite, then it seemed obvious that massively centralized platforms with the resources to both harvest and clean data — Google, Facebook, etc. — would have a big advantage. This, I would admit, was also a conclusion I was particularly susceptible to, given my focus on Aggregation Theory and its description of how the Internet, contrary to initial assumptions, leads to centralization.

The initial roll-out of large language models seemed to confirm this point of view: the two most prominent large language models have come from OpenAI and Google; while both describe how their text (GPT and GLaM, respectively) and image (DALL-E and Imagen, respectively) generation models work, you either access them through OpenAI’s controlled API, or in the case of Google don’t access them at all. But then came this summer’s unveiling of the aforementioned Midjourney, which is free to anyone via its Discord bot. An even bigger surprise was the release of Stable Diffusion, which is not only free, but also open source — and the resultant models can be run on your own computer.

There is, as you might expect, a difference in quality; Dall-E, for example, had the most realistic “newspaper delivery boy throwing a newspaper”:

Newspaper boys, by Dall-E

Stable Diffusion was on the other end of the spectrum:

Newspaper delivery boys, by Stable Diffusion

What is important to note, though, is the direction of each project’s path, not where they are in the journey. To the extent that large language models (and I should note that while I’m focusing on image generation, there are a whole host of companies working on text output as well) are dependent not on carefully curated data, but rather on the Internet itself, is the extent to which AI will be democratized, for better or worse.

The Impact on Creators

The worse is easy to envision; Charlie Warzel issued a mea culpa for using an AI image as an illustration in a post about Alex Jones:

I told Bors that what I felt worst about was how mindless my decision to use Midjourney ultimately had been. I was caught up in my own work and life responsibilities and trying to get my newsletter published in a timely fashion. I went to Getty and saw the same handful of photos of Alex Jones, a man who I know enjoys when his photo is plastered everywhere. I didn’t want to use the same photos again, nor did I want to use his exact likeness at all. I also, selfishly, wanted the piece to look different from the 30 pieces that had been published that day about Alex Jones and the Sandy Hook defamation trial. All of that subconsciously overrode all the complicated ethical issues around AI art that I was well apprised of.

What worries me about my scenario is that Midjourney was so easy to use, so readily accessible, and it solved a problem (abstracting Jones’ image in a visually appealing way), that I didn’t have much time or incentive to pause and think it through. I can easily see others falling into this like I did.

For these reasons, I don’t think I’ll be using Midjourney or any similar tool to illustrate my newsletter going forward (an exception would be if I were writing about the technology at a later date and wanted to show examples). Even though the job wouldn’t go to a different, deserving, human artist, I think the optics are shitty, and I do worry about having any role in helping to set any kind of precedent in this direction. Like others, I also have questions about the corpus used to train these art tools and the possibility that they are using a great deal of art from both big-name and lesser-known artists without any compensation or disclosure to those artists. (I reached out to Midjourney to ask some clarifying questions as to how they choose the corpus of data to train the tool, and they didn’t respond.)

I get Warzel’s point, and desire to show solidarity to artists worried about the impact of AI-generated art on their livelihoods. They are, it seems to me, right to worry: I opened this Article discussing the demise of newspapers which, once the connection between duplication and distribution was severed, quickly saw their business models fall apart. If the connection between idea creation and idea substantiation is being severed, it seems reasonable to assume all attendant business models might suffer the same fate.

There are, though, two rejoinders: the first is that abundance has its own reward. I am uniquely biased in this regard, seeing as how I make my living on the Internet as a publisher effectively competing with the New York Times, but I would argue that not just the quantity but, in absolute terms, the quality of content available to every single person in the world is dramatically higher than it was before the distribution bottleneck was removed. It seems obvious that removing the substantiation bottleneck from ideas will result in more good ones as well (along with, by definition, an even greater increase in not so good ones).

The analogy to publishing also point to what will be the long-term trend for any profession affected by these models: relatively undifferentiated creators who depended on the structural bundling of idea creation and substantiation will be reduced to competing with zero marginal cost creators for attention generated and directed from Aggregators; highly differentiated creators, though, who can sustainably deliver both creation and substantiation on their own will be even more valuable. Social media, for example, has been a tremendous boon to differentiated publishers: it gives readers a megaphone to tell everyone how great said publisher is. These AI tools will have a similar effect on highly differentiated creators, who will leverage text-based iteration to make themselves more productive and original than ever before.

The second rejoinder is perhaps more grim: this is going to happen regardless. Warzel may be willing to overlook the obvious improvement in not just convenience but also, for his purposes, quality proffered by his use of Midjourney, but few if any will make the same choice. AI-generated images will, per the image above, soon be a flood, just as publishing on the Internet quickly overwhelmed the old newspaper business model.

Moreover, just as native Internet content is user-generated content, the iterative and collaborative nature of AI-generated content — both in the sense of being a by-product of content already created, and also the fact that every output can be further iterated upon by others — will prove to be much more interesting and scalable than what professional organizations can produce. TikTok, which pulls content from across its network to keep users hooked, is the apotheosis of user-generated content; Metaverses may be the apotheosis of AI-generated content.

I wrote a follow-up to this Article in this Daily Update.

  1. Beyond the $600 annual fee I paid to Midjourney to have access to the fully rights-unencumbered Corporate plan 

  2. In the case of these image applications, noise is added to an known image and then the model is trained on backing out the image from pure noise; the resultant model can then be applied to any arbitrary text applied to pure noise, based on further training of matched text and images 

The Services iPhone

Back when Stratechery first launched there was no bigger day than iPhone announcement day, and there was arguably no bigger year than 2013, when I was just getting started. That was the year Apple was, for the first time, coming out with a non-top-of-the-line iPhone, what ultimately became known as the iPhone 5C. We would find out later that the iPhone 5C was a bit of a one-off: the iPhone 5 with its chamfered edges was simply too expensive to make (and the edges chipped horribly), making it unsuitable for Apple’s “our cheaper phones are our older phones with a lower price” strategy; Apple would one day come out with the iPhone SE to address the lower end, but the real expansion came when the iPhone X was priced at $300 more than the iPhone 8 — up-market, not down.

Still, it’s interesting to look back at how much time I spent on that product launch:

  • Before the event I wrote Thinking about iPhone Pricing, where I guessed that the 5C would cost $450 (but made the case $550 might make more sense).
  • After the event I wrote Two Minutes, Fifty-six Seconds, which referred to how long into the keynote it took me to realize that $550 would be the price; it was clear that Apple was focused on differentiation, justifying its prices, not lowering them.
  • A week later I wrote The $550 iPhone 5C Makes Perfect Sense, which was mostly about subsidies and how they might have impacted Apple’s pricing decisions.
  • A couple of days after that I wrote what I consider one of the seminal Articles of early Stratechery — the ideas in What Clayton Christensen Got Wrong are one of the reasons I wanted to start the site. This was in response to critics who were sure Apple was setting themselves up for disruption with their pricing strategy; my argument was that differentiation in the user experience derived from integration was a sustainable moat that justified higher pricing.
  • Finally, a week after that I wrote Obsoletive, which made the case that thinking about the iPhone as a disruptive product was mistaken; rather, what made it so compelling — including its high price — was the fact it obsoleted so many other products in our lives.

(Oh, and for what it’s worth, I asked a month later: So the 5S is (allegedly) killing the 5C. Why is this bad news?. If customers were actually heavily favoring the more expensive 5S then this actually made all of the points I had written over the previous month.)

I was originally inspired to look up this history as part of an introduction explaining why iPhone events no longer seem Article worthy, but are worth saving an Update slot for; what struck me while reading through these old pieces though, is not only the degree to which they show Apple’s consistency, but also how much the company has changed — and that is Article worthy.

Apple’s Increasing ARPU

CEO Tim Cook is fond of citing Apple’s ability to integrate hardware and software; over the last few years he has taken care to add “and services” as well. What was interesting about his opening in yesterday’s keynote, though, was that he has now moved up to a higher level of abstraction: devices.

Products that are intuitive and easy-to-use, that have a unique integration of hardware and software, and that are incredibly personal. Today we’re here to talk about three products that have become essential in our lives: iPhone, AirPods, and Apple Watch. They’re always with you, whenever and wherever you need them, and are designed to work seamlessly together. On their own, each is industry-leading. Together, they provide a magical experience.

This is an expression of a strategy that became clear several years ago; I wrote in Apple’s Middle Age:

Apple’s growth is almost entirely inwardly-focused when it comes to its user-base: higher prices, more services, and more devices.

Very few people just buy an iPhone: they upgrade to a higher-priced model, they spend money in the App Store and on subscriptions, and they buy an Apple Watch and AirPods that work seamlessly with their phone. The end result is that Apple isn’t making $550 per customer, to go back to the iPhone 5C, or $650 in the case of the 5S: they’re making upwards of $2000 — $1,000+ for a top-of-the-line iPhone, $400+ for a Watch, $200+ for AirPods, and all of that App Store revenue (and this doesn’t even include what is likely a thriving accessories business, AppleCare, or the Google search deal).

It is, to be frank, justified: all of these devices really do work well together, and iPhones remain top-of-class. And, for all of the kvetching about the App Store, and Apple’s arguably anticompetitive actions to maintain its control, it is true that the concept was revolutionary.

The Service Narrative

It was in January 2016 that Apple first articulated the so-called “Services Narrative”; after a quarter in which the company’s new iPhone 6S posted relatively disappointing sales, CFO Luca Maestri made the case on the earnings call that it was a mistake to think of Apple as a hardware company, subject to the vagaries of consumer demand. I wrote at the time (forgive the long excerpt, but it’s directly applicable to the point):

Cook and CFO Luca Maestri went to a lot of effort to sell the narrative that Apple is becoming a services company, and frankly, I think they kind of overdid it.

Specifically, Apple created a new way of evaluating their services called “Installed Base Related Purchases.” This is basically Apple’s services revenue plus the amount they pay to developers from the App Store and to most digital content owners on the iTunes Store. Said payouts don’t appear on Apple’s balance sheet, nor should they: Apple isn’t some sort of middle man for Candy Crush Saga, buying it wholesale and then selling it at a profit. Rather, they are facilitating a transaction between a content creator and a consumer, and taking a 30% tax.

Moreover, one of the benefits of being recognized as a services company is that your revenue is valued more highly with the presumption that it is higher margin; by adding in the 70% Apple pays out they are certainly able to crow about a higher revenue number, but they are dramatically reducing their associated margin by doing so. It’s silly.

To be sure, Apple’s services revenue numbers are impressive (although it should be noted that services revenue on a per-active-user basis actually decreased year-over-year). But it is very clear that the company remains a differentiated hardware company, as evidenced by everyone’s favorite question:

In the past, Apple’s been very known in always having a premium product. With the slowdown in the macro FX and also GDP revision, is Apple’s strategy go-to-market still always at premium product, or is there a need to go to more also a middle market or lower price point to attract more customers?

Ah, it’s the ol’ “Will Apple make a cheaper iPhone?” query. This one, though, was smarter than it appears, thanks to the next sentence:

Just because it seems like growing that installed base and services, as you pointed out, really economically could really help out Apple in the long-term.

As I’ve written innumerable times, services (horizontal) and hardware (vertical) companies have very different strategic priorities: the former ought to maximize their addressable market (by, say, making a cheaper iPhone), while the latter ought to maximize their differentiation. And, Cook’s answer made clear what Apple’s focus remains:

Our strategy is always to make the best products…We have the premium part of our line is the 6s and the 6s Plus. We also have a mid-price point, with the iPhone 6 and the iPhone 6 Plus. And we continue to offer the iPhone 5s in the market and it continues to do quite well. And so we offer all of those and I don’t see us deviating from that approach.

To be clear, I think this is the exact right approach for Apple, and as I noted above, I think these results show that the strategy continues to work. But let’s be honest: that means Apple is not a services company; they have a nice services revenue stream, but the company is rightly judged now and for the foreseeable future on the performance of its hardware.

I revisited this take a few years later, in an Update entitled Apple the Services Company (for Real!), where I noted that Apple’s estimated $44/user per year in services revenue still paled in comparison to Google or Facebook, but was meaningful all the same, particularly once you realized that Apple’s users had to spend a whole lot of money to even enter their ecosystem. I concluded:

Of course only looking at services revenue is not quite right either: that number does not include the cost of the iPhone itself, the price of entry to Apple’s ecosystem (although that payment may not necessarily flow to Apple, as is the case with phones handed down or resold). Apple, though, by de-emphasizing unit sales and focusing on the installed base, is making clear that the number that matters is average revenue per installed device; in other words, to the extent a company is what it measures — or at least reports — Apple is now a services company for real.

This was perhaps a bit generous, given the strategic tension I noted in the earlier excerpt: Apple may have been shifting its metrics to ones that reflected a focus on services, but at the end of the day, the company was still charging a pretty high price for entry.

The Services Company

I thought the products Apple introduced yesterday were pretty impressive:

  • The Apple Watch Ultra looks like the SUV of watches: it will be sold as a tool for extreme athletes, and mostly bought because it is so clearly new and different, clearly distinguishable from all of the other Apple Watches on the market.
  • AirPods Pro was a killer product in its first iteration; the 2nd generation looks set to address the 1st generation’s few flaws while delivering the sort of improvements we might expect from every tech product.
  • The iPhone Pro is increasingly differentiated from the iPhone, not just in terms of having a faster processor — a first — but also with software-driven differentiation like the new Dynamic Island functionality.

The most surprising announcement of all, though, were the prices. Everything stayed the same! This was not what I, or close followers of Apple like John Gruber, expected at all. After all, Apple’s strategy the past several years seemed to be focused on wringing more revenue out of existing customers. More importantly, the last year has seen a big increase in inflation:

U.S. inflation over the last five years

What this means is that in real terms Apple’s products actually got cheaper. Apple did, to be sure, raise prices around the world, but this is better explained by the fact the company runs on the dollar, which is the strongest in years; to put it another way, those foreign prices are derived from the U.S. price, and that price stayed the same, which means the price is lower.

This doesn’t make much sense for the product company Apple has always been thought to be, and doesn’t fully align with the approach I laid out in Apple’s Middle Age. It does, though, make all kinds of sense for a services company, which is focused first-and-foremost on increasing its install base. Indeed, this is the missing piece from that Update I wrote about Apple’s changing metrics. To measure its business based on users, not products, was to measure like a services company; to lower the prices of the products that lead to services revenue is to price like one.

This is, in a weird way, a relief: it has been disconcerting for people who think of Apple as a product company to see the company fight so fiercely for its App Store model, and to see the way it is willing to approach if not cross the line of anticompetitive behavior when it comes to App Tracking Transparency and its clear ambitions in the advertising space. To declare that the company is now clearly driven by Services doesn’t refute these narratives; rather, it at least justifies them, because they are exactly what a Services company ought to do. Here’s hoping that the products that made the company great don’t suffer from what is, at this point, a clear shift in strategy.

I wrote a follow-up to this Article in this Daily Update.

Rights, Laws, and Google

The first and most important takeaway from Kashmir Hill’s excellent article in the New York Times about Mark, the man flagged by Google as a purveyor of Child Sexual Abuse Material (CSAM) for taking pictures of his son’s penis and sending them to their family doctor, and who subsequently lost nearly every aspect of his digital life when Google deleted his account, are the tremendous trade-offs entailed in the indiscriminate scanning of users’ cloud data.

On one hand, it seems like an incredible violation of privacy to have a private corporation effectively looking through every photo you upload, particularly when those uploads happen as part of the expected way in which your smartphone operates (users technically agree to this scanning, but as part of an endless End User License Agreement that is both ridiculously long and, more pertinently, inescapable if you want to use your phone as it was intended). Moreover, Google doesn’t simply scan for CSAM photos that are already known to exist via the PhotoDNA database of photos of exploited children; the company also leverages machine learning to look for new CSAM that hasn’t yet been identified as such.

On the other hand, as horrific as the material in the PhotoDNA database is, much of it has been floating around the Internet for years, which is to say the abuse depicted happened long ago; Google’s approach has the potential to discover abuse as it is happening, making it possible for the authorities to intercede and rescue the child in question. Hill’s story noted that in 2021 the CyberTipline at the National Center for Missing and Exploited Children, the only entity legally allowed to hold CSAM (NCMEC also manages the PhotoDNA database), “alerted authorities to ‘over 4,260 potential new child victims'”. We don’t know how many of those children were subsequently rescued, but a question worth posing to anyone unilaterally opposed to Google’s approach is how big that number would have to be to have made it worthwhile?

But, to return to the original hand, one of those 4,260 potential new child victims was Mark’s son (and another was taken by Cassio, a second father found by Hill caught in the same predicament, for the same reasons): the question for those applauding Google’s approach is how big the number of false positives would have to be to shut the whole thing down?

It was the exploration of these trade-offs that was at the heart of the Update I wrote about Hill’s story last week; as I noted there are no easy answers:

Nearly every aspect of this story is incredibly complex, and I understand and respect arguments on both sides: should there be scanning of cloud-related content? Should machine learning be leveraged to find new photos? Is it reasonable to obliterate someone’s digital life — except for what you give the police — given the possibility that they may be committing horrific crimes? These are incredibly difficult questions, particularly in the absence of data, because the trade-offs are so massive.

However, it seemed to me that one aspect of the case was very clear:

There is, though, one part of the story that is black-and-white. Google is unquestionably wrong to not restore the accounts in question. In fact, I am stunned by the company’s approach in these cases. Even if you grant the arguments that this awesome exercise of surveillance is warranted, given the trade-offs in question, that makes it all the more essential that the utmost care be taken in case the process gets it wrong. Google ought to be terrified it has this power, and be on the highest alert for false positives; instead the company has gone in the opposite direction, setting itself as judge, jury, and executioner, even when the people we have collectively entrusted to lock up criminals ascertain there was no crime. It is beyond arrogant, and gives me great unease about the company generally, and its long-term investments in AI in particular.

Not that it matters, one may argue: Google can do what they want, because they are a private company. That is an argument that may ring familar.

Tech and Liberty

In 2019 I discussed the distinction between public and private restrictions on speech in Tech and Liberty:

Alexander Hamilton was against the Bill of Rights, particularly the First Amendment. This famous xkcd comic explains why:

Free Speech by xkcd

According to Randall Munroe, the author, the “Right to Free Speech” is granted by the First Amendment, which was precisely the outcome Hamilton feared in Federalist No. 84:

I go further, and affirm that bills of rights, in the sense and to the extent in which they are contended for, are not only unnecessary in the proposed Constitution, but would even be dangerous. They would contain various exceptions to powers not granted; and, on this very account, would afford a colorable pretext to claim more than were granted. For why declare that things shall not be done which there is no power to do? Why, for instance, should it be said that the liberty of the press shall not be restrained, when no power is given by which restrictions may be imposed? I will not contend that such a provision would confer a regulating power; but it is evident that it would furnish, to men disposed to usurp, a plausible pretense for claiming that power. They might urge with a semblance of reason, that the Constitution ought not to be charged with the absurdity of providing against the abuse of an authority which was not given, and that the provision against restraining the liberty of the press afforded a clear implication, that a power to prescribe proper regulations concerning it was intended to be vested in the national government. This may serve as a specimen of the numerous handles which would be given to the doctrine of constructive powers, by the indulgence of an injudicious zeal for bills of rights.

Hamilton’s argument is that because the U.S. Constitution was created not as a shield from tyrannical kings and princes, but rather by independent states, all essential liberties were secured by the preamble (emphasis original):

WE, THE PEOPLE of the United States, to secure the blessings of liberty to ourselves and our posterity, do ORDAIN and ESTABLISH this Constitution for the United States of America.

Hamilton added:

Here, in strictness, the people surrender nothing; and as they retain every thing they have no need of particular reservations.

Munroe, though, assumes the opposite: liberty, in this case the freedom of speech, is an artifact of law, only stretching as far as government action, and no further. Pat Kerr, who wrote a critique of this comic on Medium in 2016, argued that this was the exact wrong way to think about free speech:

Coherent definitions of free speech are actually rather hard to come by, but I would personally suggest that it’s something along the lines of “the ability to voluntarily express (and receive) opinions without suffering excessive penalties for doing so”. This is a liberal principle of tolerance towards others. It’s not an absolute, it isn’t comprehensive, it isn’t rigorously defined, and it isn’t a law.

What it is is a culture.

The context of that 2019 Article was the differing decisions between Facebook and Twitter in terms of allowing political ads on their platforms; over the ensuing three years the willingness and length to which these and other large tech platforms have been willing to go to police speech has expanded dramatically, even as the certainty that private censorship is ‘good actually’ has become conventional wisdom. I found this paragraph in a New York Times article about Elon Musk’s attempts to buy Twitter striking:

The plan jibes with Mr. Musk’s, Mr. Dorsey’s and Mr. Agrawal’s beliefs in unfettered free speech. Mr. Musk has criticized Twitter for moderating its platform too restrictively and has said more speech should be allowed. Mr. Dorsey, too, grappled with the decision to boot former President Donald J. Trump off the service last year, saying he did not “celebrate or feel pride” in the move. Mr. Agrawal has said that public conversation provides an inherent good for society. Their positions have increasingly become outliers in a global debate over free speech online, as more people have questioned whether too much free speech has enabled the spread of misinformation and divisive content.

In other words, the culture has changed; the law persists, but it does not and, according to the New York Times, ought not apply to private companies.


The Google case is not about the First Amendment, either legally or culturally. The First Amendment is not absolute, and CSAM is an obvious example. In 1957’s Roth v. United States the Supreme Court held that obscene speech was not protected by the First Amendment; Justice William Brennan Jr. wrote:

All ideas having even the slightest redeeming social importance — unorthodox ideas, controversial ideas, even ideas hateful to the prevailing climate of opinion — have the full protection of the guaranties, unless excludable because they encroach upon the limited area of more important interests. But implicit in the history of the First Amendment is the rejection of obscenity as utterly without redeeming social importance. This rejection for that reason is mirrored in the universal judgment that obscenity should be restrained, reflected in the international agreement of over 50 nations, in the obscenity laws of all of the 48 States, and in the 20 obscenity laws enacted by the Congress from 1842 to 1956.

This reasoning is a reminder that laws ultimately stem from culture; still, the law being the law, definitions were needed, which the Supreme Court provided in 1973’s Miller v. California. Obscene works (1) appeal to the prurient interest in sex, (2) portrays in a patently offensive way sexual conduct specifically defined by a relevant law and (3) lack serious literary, artistic, political, or scientific value. The Supreme Court went further in terms of CSAM in 1982’s New York v. Ferber, holding that the harm inflicted on children is sufficient reason to make all forms of CSAM illegal, above and beyond the standards set forth by Miller. Justice Byron White wrote:

Recognizing and classifying child pornography as a category of material outside the protection of the First Amendment is not incompatible with our earlier decisions. “The question whether speech is, or is not, protected by the First Amendment often depends on the content of the speech”…

The test for child pornography is separate from the obscenity standard enunciated in Miller, but may be compared to it for the purpose of clarity. The Miller formulation is adjusted in the following respects: a trier of fact need not find that the material appeals to the prurient interest of the average person; it is not required that sexual conduct portrayed be done so in a patently offensive manner; and the material at issue need not be considered as a whole. We note that the distribution of descriptions or other depictions of sexual conduct, not otherwise obscene, which do not involve live performance or photographic or other visual reproduction of live performances, retains First Amendment protection. As with obscenity laws, criminal responsibility may not be imposed without some element of scienter on the part of the defendant.

“Scienter”, the “knowledge of the nature of one’s act”, is what ties this judicial history back to the original discussion of Google’s actions against Mark. As Hill explained in the New York Times:

I have seen the photos that Mark took of his son. The decision to flag them was understandable: They are explicit photos of a child’s genitalia. But the context matters: They were taken by a parent worried about a sick child.

The problem in this case comes from who is determining scienter.

Google and the Bill of Rights

Quite clearly Mark did not intend for the pictures he took for his son’s telemedicine to be used for pornographic purposes. The San Francisco Police Department, which had been notified by Google after a human reviewer confirmed the machine learning-driven discovery of Mark’s photos of his son, agreed. From Hill’s story:

In December 2021, Mark received a manila envelope in the mail from the San Francisco Police Department. It contained a letter informing him that he had been investigated as well as copies of the search warrants served on Google and his internet service provider. An investigator, whose contact information was provided, had asked for everything in Mark’s Google account: his internet searches, his location history, his messages and any document, photo and video he’d stored with the company.

The search, related to “child exploitation videos,” had taken place in February, within a week of his taking the photos of his son. Mark called the investigator, Nicholas Hillard, who said the case was closed. Mr. Hillard had tried to get in touch with Mark but his phone number and email address hadn’t worked. “I determined that the incident did not meet the elements of a crime and that no crime occurred,” Mr. Hillard wrote in his report. The police had access to all the information Google had on Mark and decided it did not constitute child abuse or exploitation.

Mark asked if Mr. Hillard could tell Google that he was innocent so he could get his account back. “You have to talk to Google,” Mr. Hillard said, according to Mark. “There’s nothing I can do.” Mark appealed his case to Google again, providing the police report, but to no avail…A Google spokeswoman said the company stands by its decisions, even though law enforcement cleared the two men.

In short, the questions about Google’s behavior are not about free speech; they do, though, touch on other Amendments in the Bill of Rights. For example:

  • The Fourth Amendment bars “unreasonable searches and seizures”; while you can make the case that search warrants were justified once the photos in question were discovered, said photos were only discovered because Mark’s photo library was indiscriminately searched in the first place.
  • The Fifth Amendment says no person shall be deprived of life, liberty, or property, without due process of law; Mark lost all of his data, email account, phone number, and everything else Google touched forever with no due process at all.
  • The Sixth Amendment is about the rights to a trial; Mark was not accused of any crime in the real world, but when it came to his digital life Google was, as I noted, “judge, jury, and executioner” (the Seventh Amendment is, relatedly, about the right to a jury trial for all controversies exceeding $20).

Again, Google is not covered by the Bill of Rights; all of these Amendments, just like the First, only apply to the government. The reason why this case is useful, though, is it is a reminder that specific legal definitions are distinct from questions of right or wrong.

Working backwards, Google isn’t legally compelled to give Mark a hearing about his digital life (Sixth Amendment); they are wrong not to. Google isn’t legally compelled to give Mark due process before permanently deleting his digital life (Fifth Amendment); they are wrong not to. Google isn’t legally compelled to not search all of the photographs uploaded to Google (by default, if you click through all of the EULA’s); they are…well, this is where it gets complicated.

I started out this Article discussing the impossible trade-offs presented by questions of CSAM. People can and do make the case that to not search for this vileness, particularly if there is a chance that it can lead to the rescue of an abused child, is its own wrong. Resolving this trade-off in this way, though — that is, to violate the spirit and culture of the Fourth Amendment — makes it all the more essential to honor the spirit and culture of the Fifth and Sixth.

Paper Barriers

James Madison answered Hamilton’s objections in a speech to Congress introducing the Bill of Rights. What is interesting is that while Hamilton took it as a given that people would know and value their rights, Madison assumed the culture would run in the opposite direction, making an articulation of those rights important not just to restrain the government, but to remind the majority to not trample the rights of the minority:

But I confess that I do conceive, that in a Government modified like this of the United States, the great danger lies rather in the abuse of the community than in the Legislative body. The prescriptions in favor of liberty ought to be levelled against that quarter where the greatest danger lies, namely, that which possesses the highest prerogative of power. But this is not found in either the Executive or Legislative departments of Government, but in the body of the people, operating by the majority against the minority.

It may be thought that all paper barriers against the power of the community are too weak to be worthy of attention. I am sensible they are not so strong as to satisfy gentlemen of every description who have seen and examined thoroughly the texture of such a defence; yet, as they have a tendency to impress some degree of respect for them, to establish the public opinion in their favor, and rouse the attention of the whole community, it may be one means to control the majority from those acts to which they might be otherwise inclined.

This Article is a manifestation of Madison’s hope. Start with the reality that it seems quaint in retrospect to think that any of the Bill of Rights would be preserved absent the force of law. This is one of the great lessons of the Internet and the rise of Aggregators: when suppressing speech entailed physically disrupting printing presses or arresting pamphleteers, then restricting government, which retains a monopoly on real world violence, was sufficient to preserve speech. Along the same lines, there was no need to demand due process or a restriction on search and seizure on any entity but the government, because only the government could take your property or send you to jail.

Aggregators, though, make private action much more possible and powerful than ever before: yes, if you are kicked off of Twitter or Facebook, you can still say whatever you want on a street corner; similarly, if you lose all of your data and phone and email, you are still not in prison — and thank goodness that is the case! At the same time, it seems silly to argue that getting banned from a social media platform isn’t an infringement on individual free speech rights, even if it is the corporations’ own free speech rights that enable them to do just that legally, just as it is silly to argue that losing your entire digital life without recourse isn’t a loss of property without due process. The big Internet companies are manifesting Madison’s fears of the majority operating against the minority, and there is nothing the Bill of Rights can do about it.

What remains are those paper barriers, and what respect they might still engender, if it is possible to “rouse the attention of the whole community.” Rights are larger than laws, and Google has violated the former, even if they are not bound by the latter. The company ought not only change its policy with regards to Mark and Cassio, but fundamentally re-evaluate the balance it has struck between its unprecedented power over people’s lives and the processes it has in place to ensure that power is not abused. If it doesn’t, the people ought to, with what power they still conserve, do it for them.

I wrote a follow-up to this Article in this Daily Update.

Instagram, TikTok, and the Three Trends

Back in 2010, during my first year of Business School, I helped give a presentation entitled “Twitter 101”:

The introductory slide from a Twitter 101 presentation in business school

My section was “The Twitter Value Proposition”, and after admitting that yes, you can find out what people are eating for lunch on Twitter, I stated “The truth is you can find anything you want on Twitter, and that’s a good thing.” The Twitter value proposition was that you could “See exactly what you need to see, in real-time, in one place, and nothing more”; I illustrated this by showing people how they could unfollow me:

A slide noting that Twitter is what you make of it

The point was that Twitter required active management of your feed, but if you put in the effort, you could get something uniquely interesting to you that was incredibly valuable.

Most of the audience didn’t take me up on it.

Facebook Versus Instagram

If there is one axiom that governs the consumer Internet — consumer anything, really — it is that convenience matters more than anything. That was the problem with Twitter: it just wasn’t convenient for nearly enough people to figure out how to follow the right people. It was Facebook, which digitized offline relationships, that dominated the social media space.

Facebook’s social graph was the ultimate growth hack: from the moment you created an account Facebook worked assiduously to connect you with everyone you knew or wish you knew from high school, college, your hometown, workplace, you name an offline network and Facebook digitized it. Of course this meant that there were far too many updates and photos to keep track of, so Facebook ranked them, and presented them in a feed that you could scroll endlessly.

Users, famously, hated the News Feed when it was first launched: Facebook had protesters outside their doors in Palo Alto when it was introduced, and far more online; most were, ironically enough, organized on Facebook. CEO Mark Zuckerberg penned an apology:

We really messed this one up. When we launched News Feed and Mini-Feed we were trying to provide you with a stream of information about your social world. Instead, we did a bad job of explaining what the new features were and an even worse job of giving you control of them. I’d like to try to correct those errors now…

The errors to be corrected were better controls over what might be shared; Facebook did not give the users what they claimed to want, which was abolishing the News Feed completely. That’s because the company correctly intuited a significant gap between its users stated preference — no News Feed — and their revealed preference, which was that they liked News Feed quite a bit. The next fifteen years would prove the company right.

It was hard to not think of that non-apology apology while watching Adam Mosseri’s Instagram update three weeks ago; Mosseri was clear that videos were going to be an ever great part of the Instagram experience, along with recommended posts. Zuckerberg reiterated the point on Facebook’s earnings call, noting that recommended posts in both Facebook and Instagram would continue to increase. A day later Mosseri told Casey Newton on Platformer that Instagram would scale back recommended posts, but was clear that the pullback was temporary:

“When you discover something in your feed that you didn’t follow before, there should be a high bar — it should just be great,” Mosseri said. “You should be delighted to see it. And I don’t think that’s happening enough right now. So I think we need to take a step back, in terms of the percentage of feed that are recommendations, get better at ranking and recommendations, and then — if and when we do — we can start to grow again.” (“I’m confident we will,” he added.)

Michael Mignano calls this recommendation media in an article entitled The End of Social Media:

In recommendation media, content is not distributed to networks of connected people as the primary means of distribution. Instead, the main mechanism for the distribution of content is through opaque, platform-defined algorithms that favor maximum attention and engagement from consumers. The exact type of attention these recommendations seek is always defined by the platform and often tailored specifically to the user who is consuming content. For example, if the platform determines that someone loves movies, that person will likely see a lot of movie related content because that’s what captures that person’s attention best. This means platforms can also decide what consumers won’t see, such as problematic or polarizing content.

It’s ultimately up to the platform to decide what type of content gets recommended, not the social graph of the person producing the content. In contrast to social media, recommendation media is not a competition based on popularity; instead, it is a competition based on the absolute best content. Through this lens, it’s no wonder why Kylie Jenner opposes this change; her more than 360 million followers are simply worth less in a version of media dominated by algorithms and not followers.

Sam Lessin, a former Facebook executive, traced this evolution from the analog days to what is next in a Twitter screenshot entitled “The Digital Media ‘Attention’ Food Chain in Progress”:

Lessin’s five steps:

  1. The Pre-Internet ‘People Magazine’ Era
  2. Content from ‘your friends’ kills People Magazine
  3. Kardashians/Professional ‘friends’ kill real friends
  4. Algorithmic everyone kills Kardashians
  5. Next is pure-AI content which beats ‘algorithmic everyone’

This is a meta observation and, to make a cheap play on words, the first reason why it made sense for Facebook to change its name: Facebook the app is eternally stuck on Step 2 in terms of entertainment (the app has evolved to become much more of a utility, with a focus on groups, marketplace, etc.). It’s Instagram that is barreling forward. I wrote last summer about Instagram’s Evolution:

The reality, though, is that this is what Instagram is best at. When Mosseri said that Instagram was no longer a photo-sharing app — particularly a “square photo-sharing app” — he was not making a forward-looking pronouncement, but simply stating what has been true for many years now. More broadly, Instagram from the very beginning — including under former CEO Kevin Systrom — has been marked first and foremost by evolution.

To put this in Lessin’s framework, Instagram started out as a utility for adding filters to photos put on other social networks, then it developed into a social network in its own right. What always made Instagram different than Facebook, though, is the fact that its content was default-public; this gave the space for the rise of brands, meme and highlight accounts, and the Instagram influencer. Sure, some number of people continued to use Instagram primarily as a social network, but Meta, more than anyone, had an understanding of how Instagram usage had evolved over time.

Kylie Jenner and Kim Kardashian asking Instagram to be Instagram

In other words, when Kylie Jenner posts a petition demanding that Meta “Make Instagram Instagram again”, the honest answer is that changing Instagram is the most Instagram-like behavior possible.

Three Trends

Still, it’s understandable why Instagram did back off, at least for now: the company is attempting to navigate three distinct trends, all at the same time.

The first trend is the shift towards ever more immersive mediums. Facebook, for example, started with text but exploded with the addition of photos. Instagram started with photos and expanded into video. Gaming was the first to make this progression, and is well into the 3D era. The next step is full immersion — virtual reality — and while the format has yet to penetrate the mainstream this progression in mediums is perhaps the most obvious reason to be bullish about the possibility.

The trend in mediums online

The second trend is the increase in artificial intelligence. I’m using the term colloquially to refer to the overall trend of computers getting smarter and more useful, even if those smarts are a function of simple algorithms, machine learning, or, perhaps someday, something approaching general intelligence. To go back to Facebook, the original site didn’t have any smarts at all: it was just a collection of profile pages. Twitter came along and had the timeline, but the only smarts there was the ability to read a time stamp: all of the content was presented in chronological order. What made Facebook’s News Feed work was the application of ranking: from the very beginning the company tried to present users the content from their network that it thought you might be most interested in, mostly using simple signals and weights. Over time this ranking algorithm has evolved into a machine-learning driven model that is constantly iterating based on every click and linger, but on the limited set of content constrained by who you follow. Recommendations is the step beyond ranking: now the pool is not who you follow but all of the content on the entire network; it is a computation challenge that is many orders of magnitude beyond mere ranking (and AI-created content another massive step-change beyond that).

The trend in AI and content online

The third trend is the change in interaction models from user-directed to computer-controlled. The first version of Facebook relied on users clicking on links to visit different profiles; the News Feed changed the interaction model to scrolling. Stories reduced that to tapping, and Reels/TikTok is about swiping. YouTube has gone further than anyone here: Autoplay simply plays the next video without any interaction required at all.

The trend in UI online

One of the reasons Instagram got itself in trouble over the last few months is by introducing changes along all of these vectors at the same time. The company introduced more video into the feed (Trend 1), increased the percentage of recommended posts (Trend 2), and rolled out a new version of the app that was effectively a re-skinned TikTok to a limited set of users (Trend 3). It stands to reason that the company would have been better off doing one at a time.

That, though, would only be a temporary solution: it seems likely that all of these trends are inextricably intertwined.

Medium, Computing, and Interaction Models

Start with medium: text is easy, which is why it was the original medium of the Internet; effectively anyone can create it. The first implication is that there is far more text on the Internet than anything else; it also follows that the amount of high quality text is correspondingly high as well (a small fraction of a large number is still very large). The second implication has to do with AI: it is easier to process and glean insight from text. Text, meanwhile, takes focus and the application of an acquired skill for humans to interpret, not dissimilar to the deliberate movement of a mouse to interact with a link.

Photos used to be more difficult: digital cameras came along around the same time as the web, but even then you needed to have a dedicated device, move those photos to your computer, then upload them to a network. What is striking about the impact of smartphones is that not only did they make the device used to take pictures the same device used to upload and consume them, but they actually made it easier to take a picture than to write text. Still, it took time for AI to catch up: at first photos were ranked using the metadata surrounding them; only over the last few years has it become possible for services to understand what the photo actually is. The most reliable indicator of quality — beyond a like — remains the photo that you stop at while scrolling.

The ease of making a video followed a similar path to photos, but more extreme: making and uploading your own videos before the smartphone was even more difficult than photos; today the mechanics are just as easy, and it’s arguably even easier to make something interesting, given the amount of information conveyed by a video relative to photos, much less a text. Still, videos require more of a commitment than text or photos, because consuming them takes time; this is where the user interaction layer really matters. Lessin again, in another Twitter screenshot:

I saw someone recently complaining that Facebook was recommending to them…a very crass but probably pretty hilarious video. Their indignant response [was that] “the ranking must be broken.” Here is the thing: the ranking probably isn’t broken. He probably would love that video, but the fact that in order to engage with it he would have to go proactively click makes him feel bad. He doesn’t want to see himself as the type of person that clicks on things like that, even if he would enjoy it.

This is the brilliance of Tiktok and Facebook/Instagram’s challenge: TikTok’s interface eliminates the key problem of what people want to view themselves as wanting to follow/see versus what they actually want to see…it isn’t really about some big algorithm upgrade, it is about relesing emotional inner tension for people who show up to be entertained.

This is the same tension between stated and revealed preference that Facebook encountered so many years ago, and its exactly why I fully expect the company to, after this pullback, continue to push forward with all three of the Instagram changes it is exploring.

Instagram’s Risk

Still, there is considerably more risk this time around: when Facebook pushed forward with the News Feed it was the young upstart moving aside incumbents like MySpace; it’s not as if its userbase was going to go backwards. This case is the opposite: Instagram is clearly aping TikTok, which is the young upstart in the space. It’s possible its users decide that if they must experience TikTok, they might as well go for the genuine thing.

This also highlights why TikTok is a much more serious challenge than Snapchat was: in that case Instagram’s network was the sword used to cut Snapchat off at the knees. I wrote in The Audacity of Copying Well:

For all of Snapchat’s explosive growth, Instagram is still more than double the size, with far more penetration across multiple demographics and international users. Rather than launch a “Stories” app without the network that is the most fundamental feature of any app built on sharing, Facebook is leveraging one of their most valuable assets: Instagram’s 500 million users…Instagram and Facebook are smart enough to know that Instagram Stories are not going to displace Snapchat’s place in its users lives. What Instagram Stories can do, though, is remove the motivation for the hundreds of millions of users on Instagram to even give Snapchat a shot.

Instagram has no such power over TikTok, beyond inertia; in fact, the competitive situation is the opposite: if the goal is not to rank content from your network, but to recommend videos from the best creators anywhere, then it follows that TikTok is in the stronger relative position. Indeed, this is why Mosseri spent so much time talking about “small creators” with Newton:

I think one of the most important things is that we help new talent find an audience. I care a lot about large creators; I would like to do better than we have historically by smaller creators. I think we’ve done pretty well by large creators overall — I’m sure some people will disagree, but in general, that’s what the data suggests. I don’t think we’ve done nearly as well helping new talent break. And I think that’s super important. If we want to be a place where people push culture forward, to help realize the promise of the internet, which was to push power into the hands of more people, I think that we need to get better at that.

There is the old Internet AMA question as to whether you would rather fight a horse-sized duck or 100 duck-sized horses. The analogy here is that in a world of ranking a horse-sized duck that everyone follows is valuable; in a world of recommendations 100 duck-sized horses are much more valuable, and Instagram is willing to sacrifice the former for the latter.

Meta’s Reward

The payoff, though, will not be “power” for these small creators: the implication of entertainment being dictated by recommendations and AI instead of reputation and ranking is that all of the power accrues to the platform doing the recommending. Indeed, this is where the potential reward comes in: this power isn’t only based on the sort of Aggregator dynamics underpinning dominant platforms today, but also absolutely massive amounts of investment in the computing necessary to power the AI that makes all of this possible.

In fact, you can make the case that if Meta survives the TikTok challenge, it will be on its way to the sort of moat enjoyed by the likes of Apple, Amazon, Google, and Microsoft, all of which have real world aspects to their differentiation. There is lots of talk about the $10 billion the company is spending on the Metaverse, but that is R&D; the more important number for this moat is the $30 billion this year in capital expditures, most of which is going to servers for AI. That AI is doing recommendations now, but Meta’s moat will only deepen if Lessin is right about a future where creators can be taken out of the equation entirely, in favor of artificially-generated content.

What is noteworty is that AI content will be an essential part of any sort of Metaverse future; I wrote earlier this year in DALL-E, the Metaverse, and Zero Marginal Content:

What is fascinating about DALL-E is that it points to a future where these three trends can be combined. DALL-E, at the end of the day, is ultimately a product of human-generated content, just like its GPT-3 cousin. The latter, of course, is about text, while DALL-E is about images. Notice, though, that progression from text to images; it follows that machine learning-generated video is next. This will likely take several years, of course; video is a much more difficult problem, and responsive 3D environments more difficult yet, but this is a path the industry has trod before:

  • Game developers pushed the limits on text, then images, then video, then 3D
  • Social media drives content creation costs to zero first on text, then images, then video
  • Machine learning models can now create text and images for zero marginal cost

In the very long run this points to a metaverse vision that is much less deterministic than your typical video game, yet much richer than what is generated on social media. Imagine environments that are not drawn by artists but rather created by AI: this not only increases the possibilities, but crucially, decreases the costs.

These AI challenges, I would add, apply to monetization as well: one of the outcomes of Apple’s App Tracking Transparency changes is that advertising needs to shift from a deterministic model to a probabilistic one; the companies with the most data and the greatest amount of computing resources are going to make that shift more quickly and effectively, and I expect Meta to be top of the list.

None of this matters, though, without engagement. Instagram is following the medium trend to video, and Meta’s resources give it the long-term advantage in AI; the big question is which service users choose to interact with. To put it another way, Facebook’s next two decades are coming into sharper focus than ever; it is how well it navigates the TikTok minefield over the next two years that will determine if that long-term vision becomes a reality.

I wrote a follow-up to this Article in this Daily Update.

Political Chips

Last Friday AMD surpassed Intel in market capitalization:

Intel vs AMD market caps

This was the second time in history this happened — the first was earlier this year — and it may stick this time; AMD, in stark contrast to Intel, had stellar quarterly results. Both stocks are down in the face of a PC slump, but that is much worse news for Intel, given that they make worse chips.

It’s also not a fair comparison: AMD, thirteen years on from its spinout of Global Foundries, only designs chips; Intel both designs and manufactures them. It’s when you include AMD’s current manufacturing partner, TSMC, that Intel’s relative decline becomes particularly apparent:

Intel, AMD, and TSMC market caps

Of course an Intel partisan might argue that this comparison is unfair as well, because TSMC manufactures chips for a whole host of companies beyond AMD. That, though, is precisely Intel’s problem.

Intel’s Stumble

The late Clay Christensen, in his 2004 book Seeing What’s Next, predicted trouble for Intel:

Intel’s well-honed processes — which are almost unassailable competitive strengths in fights for undershot customers hungering for performance increases — might inhibit its ability to fight for customers clamoring for customized products. Its exacting manufacturing process could hamper its ability to deliver customized products. Its sales force could have difficulty adapting to a very different sales cycle. It would have to radically alter its marketing process. The VCE model predicts that operating “fast fabs” will be an attractively profitable point in the value chain in the future. The good news for IDMs such as IBM and Intel is that they own fabs. The bad news is that their fabs aren’t fast. Entrants without legacy processes could quite conceivably develop better proprietary processes that can rapidly deliver custom processors.

This sounds an awful lot like what happened over the ensuing years: one of TSMC’s big advantages is its customer service. Given the fact that the company was built as a pure play foundry it has developed processes and off-the-shelf building blocks that make it easy for partners to build custom chips. This was tremendously valuable, even if the resultant chips were slower than Intel’s.

What Christensen didn’t foresee was that Intel would lose the performance crown; rather, he assumed that performance would cease to be an important differentiator:

If history is any guide, motivated innovators will continue to do the seemingly impossible and find unanticipated ways to extend the life of Moore’s Law. Although there is much consternation that at some point Moore’s Law will run into intractable physical limits, the only thing we can predict for certain is that innovators will be motivated to figure out solutions.

But this does not address whether meeting Moore’s Law will continue to be paramount to success. Everyone always hopes for the emergence of new, unimagined applications. But the weight of history suggests the unimagined often remains just that; ultimately ever more demanding applications will stop appearing or will emerge much more slowly than anticipated. But even if new, high-end applications emerge, rocketing toward the technological frontier almost always leaves customers behind. And it is in those overshot tiers that disruptions take root.

How can we tell if customers are overshot? One signal is customers not using all of a product’s functionality. Can we see this? There are ever-growing populations of users who couldn’t care less about increases in processing power. The vast majority of consumers use their computers for word processing and e-mail. For this majority, high-end microprocessors such as Intel’s Itanium and Pentium 4 and AMD’s Athlon are clearly overkill. Windows XP runs just fine on a Pentium III microprocessor, which is roughly half as fast as the Pentium 4. This is a sign that customers may be overshot.

Obviously Christensen was wrong about a Pentium III being good enough, and not just because web pages suck; rather, the infinite malleability of software really has made it possible to not just create new kinds of applications but to also substantially rework previous analog solutions. Moreover, the need for more performance is actually accelerating with the rise of machine-learning based artificial intelligence.

Intel, despite being a chip manufacturer, understood the importance of software better than anyone. I explained in a Daily Update earlier this year about how Pat Gelsinger, then a graduate student at Stanford, convinced Intel to stick with a CISC architecture design because that gave the company a software advantage; from an oral history at the Computer Museum:

Gelsinger: We had a mutual friend that found out that we had Mr. CISC working as a student of Mr. RISC, the commercial versus the university, the old versus the new, teacher versus student. We had public debates of John and Pat. And Bear Stearns had a big investor conference, a couple thousand people in the audience, and there was a public debate of RISC versus CISC at the time, of John versus Pat.

And I start laying out the dogma of instruction set compatibility, architectural coherence, how software always becomes the determinant of any computer architecture being developed. “Software follows instruction set. Instruction set follows Moore’s Law. And unless you’re 10X better and John, you’re not 10X better, you’re lucky if you’re 2X better, Moore’s Law will just swamp you over time because architectural compatibility becomes so dominant in the adoption of any new computer platform.” And this is when x86– there was no server x86. There’s no clouds at this point in time. And John and I got into this big public debate and it was so popular.

Brock: So the claim wasn’t that the CISC could beat the RISC or keep up to what exactly but the other overwhelming factors would make it the winner in the end.

Gelsinger: Exactly. The argument was based on three fundamental tenets. One is that the gap was dramatically overstated and it wasn’t an asymptotic gap. There was a complexity gap associated with it but you’re going to make it leap up and that the CISC architecture could continue to benefit from Moore’s Law. And that Moore’s Law would continue to carry that forward based on simple ones, number of transistors to attack the CISC problems, frequency of transistors. You’ve got performance for free. And if that gap was in a reasonable frame, you know, if it’s less than 2x, hey, in a Moore’s Law’s term that’s less than a process generation. And the process generation is two years long. So how long does it take you to develop new software, porting operating systems, creating optimized compilers? If it’s less than five years you’re doing extraordinary in building new software systems. So if that gap is less than five years I’m going to crush you John because you cannot possibly establish a new architectural framework for which I’m not going to beat you just based on Moore’s Law, and the natural aggregation of the computer architecture benefits that I can bring in a compatible machine. And, of course, I was right and he was wrong.

That last sentence needs a caveat: Gelsinger was right when it came to computers and servers, but not smartphones. There performance wasn’t free, because manufacturers had to be cognizant of power consumption. More than cognizant, in fact — power usage was the overriding concern. Tony Fadell, who created the iPod and led the development of the first three generations of the iPhone, told me in an interview earlier this year:

You have to have that point of view of that every nanocoulomb is sacred and compatibility doesn’t matter, we’re going to use the best bits, but we’re not going to make sure it has to be the same look and feel. It doesn’t have to have the same principles that is designed for a laptop or a standalone desktop computer, and then bring those down to something that’s smaller form factor, and works within a certain envelope. You have to rethink all the principles. You might use the bits around, and put them together in different ways and use them differently. That’s okay. But your top concept has to be very, very different about what you’re building, why you’re building it, what you’re solving, and the needs of that new environment, which is mobile, and mobile at least for a day or longer for that battery life.

The key phrase there is “compatibility doesn’t matter”; Gelsinger’s argument for CISC over RISC rested on the idea that by the time you remade all of the software created for CISC, Intel would have long since overcome the performance delta between different architectures via its superior manufacturing, which would allow compatibility to trump the competition. Smartphones, though, provided a reason to build up the software layer from scratch, with efficiency, not performance, as the paramount goal.1

All of this still fit in Christensen’s paradigm, I would note: foundries like TSMC and Samsung could accommodate new chip designs that prioritized efficiency over performance, just as Christensen predicted. What he didn’t foresee in 2004 was just how large the smartphone market would be. While there are a host of reasons why TSMC took the performance crown from Intel over the last five years, a major factor is scale: TSMC was making so many chips that it had the money and motivation to invest in Moore’s Law.

The most important decision was shifting to extreme ultraviolet lithography at a time when Intel thought it was much too expensive and difficult to implement; TSMC, backed by Apple’s commitment to buy the best chips it could make, committed to EUV in 2014, and delivered the first EUV-derived chips in 2019 for the iPhone.

Those EUV machines are made by one company — ASML. They’re worth more than Intel too (and Intel is a customer):

Intel, AMD, TSMC, and ASML market caps

The Dutch company, to an even greater degree than TSMC, is the only lithography maker that can afford to invest in the absolute cutting edge.

From Technology to Economics

In 2021’s Internet 3.0 and the Beginning of (Tech) History, I posited that the first era of the Internet was defined by technology, i.e. figuring out what was possible. Much of this technology, including standards like TCP/IP, DNS, HTTP, etc. was developed decades ago; this era culminated in the dot com bubble.

The second era of the Internet was about economics, specifically the unprecedented scale possible in a world of zero distribution costs.

Unlike the assumptions that undergird Internet 1.0, it turned out that the Internet does not disperse economic power but in fact centralizes it. This is what undergirds Aggregation Theory: when services compete without the constraints of geography or marginal costs, dominance is achieved by controlling demand, not supply, and winners take most.

Aggregators like Google and Facebook weren’t the only winners though; the smartphone market was so large that it could sustain a duopoly of two platforms with multi-sided networks of developers, users, and OEMs (in the case of Android; Apple was both OEM and platform provider for iOS). Meanwhile, public cloud providers could provide back-end servers for companies of all types, with scale economics that not only lowered costs and increased flexibility, but which also justified far more investments in R&D that were immediately deployable by said companies.

Chip manufacturing obviously has marginal costs, but the fixed costs are so much larger that the economics are not that dissimilar to software (indeed, this is why the venture capital industry, which originated to support semiconductor startups, so seamlessly transitioned to software); today TSMC et al invest billions of dollars into a single fab that generates millions of chips for decades.

That increase in scale is why a modular value chain ultimately outcompeted Intel’s integrated approach, and it’s why TSMC’s position seems so impregnable: sure, a chip designer like MediaTek might announce a partnership with Intel to maybe produce some lower-end chips at some point in the future, but there is a reason it is not a firm commitment and not for the leading edge. TSMC, for at least the next several years, will make the best chips, and because of that will have the most money to invest in what comes next.

Scale, though, is not the end of the story. Again from Internet 3.0 and the Beginning of (Tech) History:

This is why I suspect that Internet 2.0, despite its economic logic predicated on the technology undergirding the Internet, is not the end-state…After decades of developing the Internet and realizing its economic potential, the entire world is waking up to the reality that the Internet is not simply a new medium, but a new maker of reality…

To the extent the Internet is as meaningful a shift [as the printing press] — and I think it is! — is inversely correlated to how far along we are in the transformation that will follow — which is to say we have only gotten started. And, after last week, the world is awake to the stakes; politics — not economics — will decide, and be decided by, the Internet.

Time will tell if my contention that an increasing number of nations will push back against American Internet hegemony by developing their own less efficient but independent technological capabilities is correct; one could absolutely make the case that the U.S.’s head start is so overwhelming that attempts to undo Silicon Valley centralization won’t pan out anywhere other than China, where U.S. Internet companies have been blocked for a generation.

Chips, though, are very much entering the political era.

Politics and the End-State

Taiwan President Tsai Ing-wen shared, as one does, some pictures from lunch on social media:

Taiwan President Tsai Ing-wen's Facebook post featuring TSMC founder Morris Chang

The man with glasses and the red tie in the first picture is Morris Chang, the founder of TSMC; behind him is Mark Liu, TSMC’s chairman. They were the first guests listed in President Tsai’s write-up of the lunch with House Speaker Nancy Pelosi, which begins:


Taiwan and the United States not only share the values ​​of democracy, freedom and human rights, but also continue to work together on economic development and democratic supply chains.

That sentence captures why Taiwan looms so large, not only on the occasion of Pelosi’s visit, but to world events for years to come. Yes, the United States supports Taiwan because of democracy, freedom and human rights; the biggest reason why that support may one day entail aircraft carriers is because of chips and TSMC. I wrote two years ago in Chips and Geopolitics:

The international status of Taiwan is, as they say, complicated. So, for that matter, are U.S.-China relations. These two things can and do overlap to make entirely new, even more complicated complications.

Geography is much more straightforward:

A map of the Pacific

Taiwan, you will note, is just off the coast of China. South Korea, home to Samsung, which also makes the highest end chips, although mostly for its own use, is just as close. The United States, meanwhile, is on the other side of the Pacific Ocean. There are advanced foundries in Oregon, New Mexico, and Arizona, but they are operated by Intel, and Intel makes chips for its own integrated use cases only.

The reason this matters is because chips matter for many use cases outside of PCs and servers — Intel’s focus — which is to say that TSMC matters. Nearly every piece of equipment these days, military or otherwise, has a processor inside. Some of these don’t require particularly high performance, and can be manufactured by fabs built years ago all over the U.S. and across the world; others, though, require the most advanced processes, which means they must be manufactured in Taiwan by TSMC.

This is a big problem if you are a U.S. military planner. Your job is not to figure out if there will ever be a war between the U.S. and China, but to plan for an eventuality you hope never occurs. And in that planning the fact that TSMC’s foundries — and Samsung’s — are within easy reach of Chinese missiles is a major issue.

China, meanwhile, is investing heavily in catching up, although Semiconductor Manufacturing International Corporation (SMIC), its Shanghai-based champion, only just started manufacturing on a 14nm process, years after TSMC, Samsung, and Intel. In the long run, though, the U.S. faced a scenario where China had its own chip supplier, even as it threatened the U.S.’s chip supply chain.

This reality is why I ultimately came down in support of the CHIPS Act, which passed Congress last week. I wrote in a Daily Update:

This is why Intel’s shift to being not simply an integrated device manufacturer but also a foundry is important: yes, it’s the right thing to do for Intel’s business, but it’s also good for the West if Intel can pull it off. That, by extension, is why I’m fine with the CHIPS bill favoring Intel…AMD, Qualcomm, Nvidia, et al, are doing just fine under the current system; they are drivers and beneficiaries of TSMC’s dominance in particular. The system is working! Which, to the point above, is precisely why Intel being helped disproportionately is in fact not a flaw but a feature: the goal should be to counteract the fundamental forces pushing manufacturing to geopolitically risky regions, and Intel is the only real conduit available to do that.

Time will tell if the CHIPS Act achieves its intended goals; the final version did, as I hoped, explicitly limit investment by recipients in China, which is already leading chip makers to rethink their investments. That this is warping the chip market is, in fact, the point: the structure of technology drives inexorably towards the most economically efficient outcomes, but the ultimate end state will increasingly be a matter of politics.

I wrote a follow-up to this Article in this Daily Update.

  1. As an example of how efficiency trumped performance, the first iPhone’s processor was actually underclocked — better battery life was more of a priority than faster performance.