Good morning,
I hope my American readers had a pleasant Labor Day holiday, and that my international readers enjoyed a day with us Americans mostly off-line!
On to the update:
AI Chips and China
A long-time reader remarked to me last week about the amount of time that has been spent on chips over the last couple of years — including an update last week about Nvidia — but it’s a topic that is difficult to escape right now. Look no further than this seismic announcement from last week; from the Wall Street Journal:
U.S. restrictions on sales of Nvidia Corp.’s high-end processors to China throw a wrench in Beijing’s ambitions to lead in artificial intelligence, as Chinese officials accused the U.S. of monopolizing advanced technologies. The curbs cut off China’s biggest tech companies from some of the world’s most advanced chips. Nvidia’s affected customers include Alibaba Group Holding Ltd., the internet giant that operates China’s largest cloud service business, and Tencent Holdings Ltd., the gaming and social media behemoth. Both sell cloud services powered by Nvidia chips that are capable of crunching huge amounts of data for advanced applications from autonomous factories to video processing. “This is a big step by the U.S. because it is targeting high performance processors that are mainly used for commercial applications,” said Handel Jones, chief executive of consulting firm International Business Strategies Inc.
Nvidia, the world’s largest chip maker by market value, said Wednesday that new U.S. rules barring the sale without a license of the advanced chips to Chinese customers would cost it $400 million in sales. It said it may have to transition some of its operations out of China. Nvidia shares fell more than 11% midday Thursday. Other chip makers also retreated. Shares in Advanced Micro Devices Inc., which said it was also affected by the license requirement though didn’t expect a material impact, were down more than 6%.
The U.S. Commerce Department, which handles export restrictions, has declined to comment on changes to its policy but said the action was aimed at preventing China from acquiring American technology to advance its military. China’s Ministry of Commerce on Thursday said the U.S. move would damage the interests of both Chinese and American companies. The U.S. should treat enterprises of all countries fairly, it said.
There is a lot to unpack here, but first, a rant: China complaining that the U.S. should treat enterprises of all countries fairly is, while not surprising, one of the most absurd statements I have ever seen. Tech is the most obvious example: U.S. Internet companies have been blocked from China for going on a generation now, even as Chinese-owned companies like TikTok are allowed to take massive chunks out of the valuation and competitive position of U.S. companies like Meta. This is, to be sure, only the tip of the spear: I could talk about forced joint ventures in China that are glorified IP theft, a systemic refusal to abide by obligations under China’s WTO membership like equal access for banks and payment companies, or the fact that Chinese companies get to raise money in the U.S. with little or no oversight. But I’ll just stick to tech: China blatantly discriminates against U.S. companies, and even if you want to make the case this is smart — there is a case to be made! — it is, and always will be, infuriating to read cynical statements like this.
That noted, there is a reason, and a benefit, to the U.S. undertaking this one-way relationship with China. I wrote earlier this year in Tech and War:
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:
- The short-term upside of relaxing sanctions against China in semiconductors in exchange for supporting sanctions against Russia is a potentially earlier end to the conflict in Ukraine.
- The medium-term risk of giving China access to Western technology is that China develops more advanced products that could be used by its military.
- The long-term risk of cutting China off is the development of an alternative to the West that is completely unconstrained by sanctions, public or private.
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.
The downsides to China being the manufacturing base of the world are well-known at this point; what is less understood is that opportunity cost is real, and has major implications. Specifically, in the period where China’s economy was relatively open (I think it is increasingly appropriate to refer to that in the past tense) it made sense for Chinese entrepreneurs — at least those competing in the global market without the benefit of the Chinese firewall — to default towards labor-intensive parts of globalized supply chains. At the beginning this was because Chinese labor was cheap, but by the time Chinese labor became expensive — which has been the case for a good while now, which is why industries like textiles have long since moved on to lower-cost countries — the built-up network and expertise was compelling enough to keep those industries in China.
At the same time, this meant that capital and IP-intensive industries — like chips, to take an obvious example — were the paths not chosen. After all, you could simply buy cutting edge capital-intensive tech from foreign suppliers; to put it another way, in an open market, to the extent it made sense for China companies to focus on labor-intensive parts of the value chain, it made sense for other countries, from the U.S. to Taiwan to Japan to Germany, to focus on capital-intensive parts, and for Chinese companies to buy capital-intensive products from foreign companies: comparative advantage for the win!
All of this worked like a charm as long as the U.S. and China were closely aligned; it is much more problematic as each increasingly views the other as the enemy. To put it another way, if you think the U.S. regrets giving up its manufacturing base to China, China to an even greater extent regrets giving up its capital-driven manufacturing capability to the U.S.
Geopolitical Considerations
This gets at the big strategic question I was driving at in Tech and War: is it better strategically to keep China dependent on U.S. tech, or to cut the country off now, thus driving them to develop their own alternatives? This, by extension, rests on an evaluation of how far off those alternatives are: as I was among the first to report in that Article, Semiconductor Manufacturing International Corporation (SMIC) was already capable of building 7nm chips, but it’s important to note that that capability is almost completely dependent on foreign suppliers; almost all of them are U.S. companies or, in the case of ASML, so dependent on U.S. IP that we control their exports. Chip manufacturing isn’t the only U.S. expertise that China depended on, though: chip design matters as well, which is where Nvidia and AMD come in (more on the Nvidia implications in a moment).
To go back to the Wall Street Journal excerpt above, I think it is silly to state that the sort of AI capabilities unlocked by graphics chips — which is to say machine learning models, the development of which are well-suited to the massive parallelization enabled by shaders, an Nvidia invention — is all about commercial applications, and not military ones. If the U.S. has chosen the path of limiting China instead of co-opting the country then cutting off Nvidia and AMD chips is an obvious move.
That, by extension, is the biggest takeaway from this news: as late as six months ago, when I wrote Tech and War, the U.S.’s approach to China was, despite the moves to cut-off Huawei, still up in the air. Now the direction is clear: the U.S. is not going to nourish Chinese dependency on U.S. technology, but rather seek to cut the country off completely. This will, to be sure, limit Chinese capabilities for some amount of time (the duration of which is a matter of debate); it also will be a clear signal to China that it needs to develop alternatives. The race now is to see who can better survive the potential elimination of the flashpoint at the center, which is TSMC and Taiwan.
Nvidia Considerations
The big issue for Nvidia is that the company, to a greater extent than maybe anyone other than Apple, has bet on the globalized world remaining as it was. Nvidia sells massive amounts of product into China, across its entire line: not only do Chinese cloud providers buy Nvidia’s datacenter products, but its electronics manufactures buy Nvidia chips for the graphics cards and computers they build. That’s even before you get to the consumer market, and remember that China is the largest gaming market in the world. Yes, a lot of this is mobile, but Nvidia has exposure all over the place.
It’s not just the end-market, though. Nvidia also has significant development capacity in China. This part of the business has been spared, for now; from CNBC:
Nvidia on Thursday said the U.S. government will allow it to continue developing its H100 artificial intelligence chip in China. It’s a win for the company after it warned Wednesday that new export restrictions could hamper its operations in the country.
Nvidia said in an SEC filing Wednesday that the U.S. government is restricting sales of high-performance AI chips for servers, the A100 and H100, to China and Russia. Sales of both chips are still restricted in those markets, though it can still develop the H100 in China. Nvidia expects a $400 million hit to revenue in the current quarter from new export restrictions…
The H100 is Nvidia’s upcoming enterprise AI chip that was previously expected to ship by the end of the year. Part of its development takes place in China. The A100 is an older model that has been shipping for three years. They are both graphics processors that can be used for supercomputing and artificial intelligence.
Nvidia’s annual report cites the risk of geopolitics as far as the sale of its products are concerned:
Geopolitical tensions and conflicts worldwide, including but not limited to Taiwan, China, Hong Kong, Israel and Korea where the manufacture of our product components and final assembly of our products are concentrated, may result in changing regulatory requirements, trade policies, export controls, import duties and economic disruptions that could impact our operating strategies, product demand, access to global markets, hiring, and profitability. The increasing focus on the strategic importance of AI technologies may result in additional regulatory restrictions that target products
What is unmentioned is this development capacity. However, this strikes me as a significant risk for both Nvidia and the U.S.: for now Nvidia’s R&D in China has been spared, but how long can that endure, given that from a U.S. perspective, the long-term risk — given the change in posture noted in the last item — is not the sale of advanced chips to China but rather the risk of China gaining the capability of building their own? That that is the question at hand is, in the end, why this story is so significant.
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!