TSMC in Arizona, Arizona Challenges, Realities and Motivations

Good morning,

As if you don’t have enough opportunities to hear me on a podcast, I appeared on the latest episode of Acquired to discuss the history and strategy of Stratechery.

On to the update:

TSMC in Arizona

From the Wall Street Journal:

President Biden visited a Taiwan Semiconductor Manufacturing Co. site Tuesday in Arizona, where the company plans to build a second factory and increase its investment to $40 billion. TSMC, the world’s largest contract chip maker, is stepping up its plans for U.S. manufacturing. That investment follows the passage of a law earlier this year designed to boost U.S. semiconductor production, reflecting Washington’s concerns about reliance on Asia for critical chip making. “This is going to be an incredible asset to the state of Arizona,” Mr. Biden said during remarks following a tour of the site. “American manufacturing is back.”

Under a previously disclosed $12 billion investment plan, TSMC is already building its first factory in Arizona.In advance of Mr. Biden’s visit to the plant on Tuesday, the White House released details of a second factory planned for the site, saying construction would start in the coming year and production would begin in 2026. The second factory, first reported by The Wall Street Journal, will make chips with 3-nanometer technology, equal to the tiniest and fastest chips available today, and TSMC’s total investment in Arizona will expand to $40 billion, the White House said. It didn’t say how many years the investment would cover.

The White House said Mr. Biden met with a group of top executives following his Arizona remarks, including Morris Chang and Mark Liu of TSMC, Tim Cook of Apple Inc., Jensen Huang of Nvidia Corp., Sanjay Mehrotra of Micron Technology Inc., Lisa Su of Advanced Micro Devices Inc. and Peter Wennink of ASML Holding NV. Mr. Cook tweeted that Apple would be the Arizona site’s largest customer…The White House said the first TSMC factory in Arizona, which originally was supposed to produce 5-nanometer chips, would also produce 4-nanometer chips. That factory is expected to start mass production in 2024.

Here is Cook’s tweet:

I don’t want to be a wet blanket here, but, well, I’m going to be a bit of a wet blanket.

Arizona Challenges

Start with the announced switch to 4-nanometer chips: as I’ve noted previously, TSMC’s 4-nanometer chips, branded N4, are just an evolved version of the company’s 5-nanometer process, which was first introduced in 2020 with the iPhone 12; calling it 4-nanometer is pure marketing. Of course these size descriptors have been pure marketing for literally decades, which Intel in particular will be happy to tell you: Intel’s 10-nanometer chips, for example, have a transistor gate pitch of 54 nanometers, while TSMC’s 7-nanometer chips have a transistor gate pitch of 55 nanometers, despite having a name that suggests they are a generation ahead. In the case of 4-nanometer, it has a transistor gate pitch of 51 nanometers, the exact same as the company’s previous two iterations of 5-nanometer chips — it’s the same generation.

In short, if that first fab comes online in 2024 (which is no sure thing — more on that in a moment), it will be, in the most generous interpretation, 2 years behind 4-nanometer production in Taiwan, and if we’re being honest, 4 years behind the 5-nanometer generation that said factory will be producing. What that means in terms of products is that at some point well after launch iPhone 15 (which will presumably have the iPhone 14 Pro’s A16 processor, which is built on TSMC’s 4-nanometer process) may have some chips made in America (but probably only some, given the fab’s small capacity), but the A17 processor in the iPhone 15 Pro, which will be built on TSMC’s new 3-nanometer process which just started production, will be made in Taiwan. That, by extension, means that we will have to wait for the iPhone 18 or so to get chips from the 2nd fab announced this week.

What is somewhat ironic about this 2~4 year delay in leading edge nodes is that it roughly mirrors the delay that Washington D.C. has traditionally put on foreign chip manufacturing, particularly in China. It’s a testament to how the U.S. has fallen behind on the leading edge that being granted two to four-year-old technology on U.S. soil calls for all of this pomp and circumstance.

That pomp and circumstance is, though, clearly the biggest driver here: notice how all of these announcements are coming from the White House, not TSMC; TSMC, if they had their druthers, would certainly not be building these fabs at all. From another article in the Wall Street Journal earlier this week:

The $12 billion Arizona semiconductor plant under construction that President Biden is visiting Tuesday represents U.S. hopes for a renaissance in manufacturing, but the Taiwanese company building it says it won’t be easy. High costs, lack of trained personnel and unexpected construction snags are among the issues cited by Taiwan Semiconductor Manufacturing Co. as it rushes to get the north Phoenix factory ready to start production in December 2023. “A range of construction costs and project uncertainty in Phoenix makes building the same advanced logic wafer fab in Taiwan considerably less capital intensive,” TSMC said in a letter last month to the Commerce Department. “The real barrier” to setting up manufacturing in the U.S. “is comparative cost to build and operate,” it said.

TSMC executives have said it isn’t easy to re-create in America the manufacturing ecosystem they have built over decades in Taiwan, drawing on local engineering talent and a network of suppliers including many in East Asia. Mr. Chang said the cost of making chips in Arizona may be at least 50% higher than in Taiwan.

TSMC has shipped as much equipment as possible from Taiwan, including clean-room equipment and chip-making tools, because American suppliers cost more or aren’t available, people familiar with the project said. Another challenge is personnel. TSMC has invested more in recruiting after struggling to find new engineering graduates in the U.S., said people familiar with its efforts. Engineers hired in the U.S. are sent to Taiwan for a year or a year and a half of training, they said. The company said TSMC Arizona currently has more than 1,000 employees and other personnel, and the number is expected to grow to 2,000 next year. The company is also sending Taiwan-based engineers to Arizona to augment staffing, luring them with doubled salaries and other benefits, according to people familiar with the matter.

From what I have heard the culture clash between US-sourced employees and Taiwanese ones is massive; TSMC has the advantage of being a very prestigious and well-paying job in Taiwan, which helps compensate for the intense work hours and always being on call. It’s not a culture that most U.S. employees are used to, which in part undergirds the point that the reason the U.S. has fallen behind on chip manufacturing does, in many respects, rest more on the “manufacturing” part than the “chip” part (after all, Intel has moved a lot of manufacturing out of the U.S. as well, particularly highly labor-intensive steps like testing and packaging). I wrote in a Daily Update earlier this year:

There is something more fundamental going on. Specifically, it is striking that so many of these issues look similar to other industries that have long since outsourced manufacturing to other countries. The repeated pattern is that high margin knowledge work, from design to marketing to software, becomes the focus of companies’ operations in the U.S.; low margin manufacturing, with its attendant pollution, workplace dangers, and lower wages ends up abroad. Some of this is endemic to the U.S. dollar being the world’s reserve currency, which gives other countries an inherent cost advantage; some of it is that those knowledge workers not only don’t want to work in manufacturing, they don’t want to live near it either. To put it another way, the U.S. has reverse-colonized the rest of the world as its industrial base, and spends liberally on its military to protect it; the rest of the world pays the bill by buying U.S. treasuries with their trade surpluses.

And yet, here we are: TSMC is investing more in a location it doesn’t want to, to build chips that will be more expensive than those built in Taiwan, and everyone is celebrating.

Realities and Motivations

I already wrote back in May about why TSMC would not want to build in the U.S. (or rather, I extensively quoted retired TSMC founder Morris Chang, who didn’t hold back about what he perceived as being a tremendous waste of money for both the U.S. and TSMC). In short, chips from these fabs will not be competitive on a unit cost basis with chips from Taiwan, and they won’t provide any real insurance for a Chinese invasion of Taiwan either, because these foundries will not be able to function without a real-time connection to TSMC in Taiwan (and the entire global supply chain for semiconductors).

It’s not just costs and interconnectedness that is a challenge for TSMC, though: remember, TSMC is a foundry, which means it produces all kinds of chips for all kinds of companies, which means there is a premium on flexibility; this is a big reason why TSMC benefits from everything being in Taiwan, because all of its engineering is in one place, and can easily adapt assembly lines for new orders and new equipment. A useful contrast here is Intel: the reason why the American chip maker could build fabs around the world is because it was only manufacturing huge volumes of its own designs, and it was doing so on fabs that were perfect replications of each other; in other words, Intel could be more distributed because it was less flexible (the need to be more flexible is one of the many challenges Intel faces in its attempted shift to being a foundry for external customers).

What is interesting about this Arizona project is that I wouldn’t be surprised if it represents TSMC adopting a bit of the Intel playbook: Apple is TSMC’s largest customer, and Nvidia and AMD are almost certainly in the top 5. Moreover, their combined volume far exceeds what TSMC is building in Arizona, which may be a feature, not a bug: I suspect these Arizona foundries are being built specifically for those three companies, and while I doubt that TSMC is going to the same extremes as Intel’s Copy Exactly! approach, my assumption is that these fabs are going to be designed for mass production of a limited number of chips, not the flexibility that is at the core of the company’s Taiwan operations.

In other words, not only will these fabs be uncompetitive economically (although this strategy would be the best way to wring as much efficiencies as possible out of the project), and not only will they be much less of a hedge against Chinese invasion than you might think, but they will also likely not provide the flexible foundry capacity that is the biggest hole in the U.S.’s domestic capabilities (beyond the trailing edge capacity that is the U.S.’s biggest shortcoming, and which is uniquely well-suited to government subsidies that are unfortunately mostly going to the leading edge).

What is more important to the powers-that-be, though, is that President Biden gets to make a big speech about a big dollar figure and Tim Cook gets to talk about “Made in America” without having to discuss the company’s continued dependence on China. TSMC will have to bear the cost, but if that is the price of shoring up U.S. support for Taiwan, well, that is the best possible insurance policy the company could buy for its operations that truly matter, which are intrinsically tied to Taiwan.

This Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.

The Stratechery 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 subscriber, and have a great day!