Apple App Store Follow-up, AWS ARM Servers, Intel’s Choice

Good morning,

There are five days left to order Stratechery t-shirts, polos, and hoodies; you can order those, along with coffee mugs, pins, stickers, and an Aggregation Theory print here.

On to the update:

Apple App Store Follow-up

I wanted to focus on three specific areas of follow-up to yesterday’s Antitrust, the App Store, and Apple: the customer relationship, why the iOS App Store is unique, and market definition.

The Customer Relationship: One thing that complicates Apple’s description of its role in the App Store as a facilitator for developers transacting with customers is the fact that Apple owns the customer relationship. There is very little data passed to developers; basically, they get money, and customers get the app. Apple gets everything else, which certainly suggests that consumers are Apple’s customers, not the developers (which would suggest that consumers do have standing to sue Apple for antitrust violations).

Apple versus Google/Steam etc.: Multiple folks pointed out that Google and Valve charge 30% for the Play Store and Steam respectively, so isn’t that rent-seeking as well? In fact, both are examples of how Apple should operate:

  • Google charges 30% for purchases in the Play Store, including in-app purchases using the Play Store API. However, apps can load a webview to their own storefront. For example, here is the Kindle app on Android (forgive all the kids books 😂):

    The Kindle app on Android

    Given this, Google’s 30% fee is perfectly legitimate. Lots of apps still use it, because the user-experience, especially for in-app purchases, is far superior; games, in particular, don’t want to have any friction in the purchase of consumables. To put it another way, Google is competing with other payment methods by innovating, not by making rules.

    [Correction: Google in fact requires games to use Google Play’s payment system]

  • Steam exists on open platforms, primarily Windows. That means that applications could choose to go around Steam and use whatever payment they choose. The equivalent on iOS would be to allow the side-loading of applications; personally, I think Apple is perfectly justified in not allowing that for security reasons, but said restriction makes it that much more pressing that alternative means of transactions be allowed in apps.

The closest analogy to the App Store’s limitations are consoles: console makers restrict what games can be placed on their platforms, and control all in-app monetization; still, all of those purchases are about playing video games. Apple, on the other hand, is trying to get a cut of purchases that have nothing to do with the underlying hardware.

The Digital Goods Market versus The Smartphone Market

Perhaps the most common defense to the idea that Apple is behaving monopolistically is that the iPhone has less than 15 percent market share worldwide. I think this misses the mark in several important ways though:

  • First, the antitrust case in question is in the U.S., where Apple has nearly 50% market share.
  • Second, contrary to popular opinion, a company doesn’t have to have a monopoly to be found guilty of antitrust violations. Look no further than Apple’s iBook conviction for conspiring to raise prices.
  • Third, and most importantly, I don’t believe the relevant market is smartphones, but rather digital goods and services.

Note that I say “digital goods and services”, not apps; I’m not a huge fan of Apple taking 30% of the cost of an app, but I can at least accept the argument that Apple is operating a digital storefront for apps and, like any storefront, charges a markup. And, while I don’t think it’s that relevant given the difference in marginal costs between shrink-wrapped software and purely digital apps, it is nevertheless true that Apple’s markup is a lot less than the retailers of old. Moreover, Apple does provide some measure of utility for that 30%, including credit card fees, sales tax reconciliation, etc., and, if you’re lucky, free marketing in the App Store.

What I do believe is illegal is Apple’s insistence — based on nothing other than rules enforced by their mandatory app approval process — that no company can sell digital goods in their app that have nothing to do with smartphones or their functionality without giving Apple 30%. I know “rent-seeking” is a loaded term, but there frankly is no better way to describe this practice. Just think how absurd it is that you can open the Amazon app and buy a physical good but can’t do the same and buy a digital good; Apple is contributing nothing to the digital good in question, which has nothing to do with iOS’s unique capabilities, they are merely taking 30% because they can.

The reality is that smartphones are the most important computing platforms on earth, the infrastructure upon which sit businesses of all types. Business have no choice but to have apps on iOS, and insisting that customers should choose their smartphone platform based on blatantly anticompetitive rules is like suggesting the proper response to, say, a hospital monopoly in one city is for citizens to move to another. In this the App Store is also different from platforms like Uber or Amazon or any other number of digital marketplaces: Uber has competitors, Amazon has competitors, the App Store doesn’t have any.

Again, allowing apps to link out to web pages for the purchase of digital goods would not at all penalize Apple for innovating: as I noted above purchases made using built-in APIs will always be a better user-experience, and many developers will appreciate that the lack of friction more than makes up for the 30% they pay for Apple. Which is great! That means Apple is earning App Store revenue by competing through innovation, not by making arbitrary rules.

AWS ARM Servers

From Bloomberg:

Amazon.com Inc. has taken a big step toward reducing reliance on Intel Corp. for a critical component of its cloud-computing service. The largest cloud company unveiled its own server processors late Monday and said the Graviton chips will support new versions of its main EC2 cloud-computing service. Until now, Amazon — and other big cloud operators — had almost exclusively used Intel Xeon chips.

Other attempts have so far failed to loosen Intel’s mighty grip on the server chip market. But Amazon is using a discounting strategy that has helped it win customers time and time again. The company said the Graviton-backed cloud service is available a “significantly lower cost” than existing offerings run on Intel processors. It’s the second time this month that Amazon has taken a swing at Intel’s server chip business. The cloud provider said on Nov. 6 that it’s offering services based on computers that use Advanced Micro Devices Inc. processors. That marked a breakthrough for AMD’s efforts to compete against Intel…

Intel processors run more than 98 percent of the world’s servers, and owners of massive data centers such as Amazon, Microsoft Corp. and Google have become some of its biggest customers. While these internet giants have driven down the price of most components by doing a lot of their own engineering, Intel’s Xeon chips have resisted that pressure. The average selling price of these processors has risen over time, something that almost never happens in the electronic industry.

This news has been a long time coming. Four years ago Qualcomm announced plans to design ARM server chips; at the time I was a bit skeptical: Intel was catching up in performance/watt, customization was compelling but it was unclear how much it would matter given that the trend in data centers was towards abstracting computing power, and Intel had plenty of room to win a price war, painful though it would be, particularly given its performance advantage.

I wasn’t alone in my skepticism; from Bloomberg that same month:

Amazon.com Inc., which operates some of the world’s largest data centers, said makers of chips that use ARM Holdings Plc’s technology aren’t keeping up with Intel Corp.’s pace of innovation. As a result, Amazon isn’t ready to start using alternatives to Intel’s chips in its servers, according to James Hamilton, a vice president for Amazon Web Services, which provides computing power and storage over the Internet to other companies.

“It’s just not quite moving fast enough,” Hamilton said in an interview, referring to the pace of development of ARM-chip technology. He spoke after a presentation at Amazon’s annual Web-services conference in Las Vegas.

Three things have changed in the intervening four years. First, as I discussed this summer in Intel and the Danger of Integration, Intel has hit a wall with 10 nanometers, while ARM performance continues to improve thanks to advances both in architecture and fabrication. Suddenly it is Intel that is “not quite moving fast enough.”

Second, in 2015 Amazon acquired chip designer Annapurna Labs. Amazon by virtue of actually running large data centers, is much better placed than Qualcomm to take advantage of the customizability of ARM chips. Specifically, they can customize their chips for their specific data center needs, in coordination with the underlying operating system; that is more difficult for a vendor like Qualcomm. From Wired:

Amazon says it took on the tricky business of chip design to better integrate the software and hardware inside its giant data centers, allowing it to offer new, cheaper services. That’s not possible with off-the-shelf chips from Intel or AMD, which are designed to satisfy many different customers, says Matt Garman, a vice president involved with Amazon’s project. “We can do some very specific things that make sure the processor runs very efficiently in our environment,” he says. “That leads to lower costs.” Garman says customers with early access to Graviton-powered servers cut their bills for some services almost in half.

This makes sense in theory: ARM, relative to Intel’s integration, is modular, not just in terms of the separation of design and fabrication, but in the design itself. That modularity makes it possible to integrate elsewhere in the value chain; in the case of Apple, the integration is between the design of their A-series chips and iOS, and in the case of Amazon, the integration is between their Graviton chips and their datacenter software.

Third, AWS is really big now. Hamilton explained why this matters on his personal blog:

This is an exciting day and one I’ve been looking forward to for more than a decade. As many of you know, the gestation time for a new innovation at AWS can incredibly short. Some of our most important services went from good ideas to successful, heavily-used services in only months. But, custom silicon is different. The time from idea to a packaged processor is necessarily longer than software-only solutions. And, while the famed Amazon two-pizza team can often deliver an excellent service in months, a custom processor takes far longer to get right and engineering costs will be tens of millions and could easily escalate into the hundreds of millions of dollars. These are big investments that require high volumes to be economically justified. Today I’m super happy to be able to talk about the AWS Graviton Processor that powers the Amazon EC2 A1 Instance family.

AWS can justify the up-front investment necessary to realize ARM’s advantages because the public cloud market is so large; in contrast, none of the companies running on top of AWS could ever justify that much capital expense, and, arguably, neither could Amazon.com by itself.

Intel’s Choice

To be clear, Intel is not (yet) doomed: for most applications, Intel-based servers will be the better choice. ARM may generally be better on a performance-per-watt basis, which is of peak importance for battery-powered smartphones, but Intel still has far better top-end performance (performance-per-watt still matters in data centers, because of the cost of powering and cooling servers, but it is not the single most important factor as it is in smartphones). Moreover, the vast majority of software would have to at a minimum be recompiled to run on ARM; that seems unlikely for most already-deployed applications.

Still, this certainly isn’t good news for Intel: this is almost certainly the first Graviton chip on a long roadmap that drives inexorably towards better performance at a faster rate than Intel, and critically, at a far lower price. Intel Xeon chips cost thousands of dollars (because you are paying not just for the chip but Intel’s design); ARM chips cost tens of dollars to fabricate for the companies that have spent the capital to make their own designs. The better Graviton chips get the less sustainable those Xeon prices, particularly at the lower-end, will be.

The ultimate takeaway is that, in the very long run, margin compression for Intel seems unavoidable; the question going forward is whether the company will finally embrace that reality (and move towards contract manufacturing a la TSMC), or wait until it is forced on them.


The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.

Thanks for being a supporter, and have a great day!