RIP American AI Industry

You had a good run

May 30, 2025

∙ Paid

There are two unspoken rules in international geopolitics that hegemonic powers must follow. First: when an adversary pushes, you push; when an adversary pulls, you pull. Second: no single power should dominate the global chessboard so completely that it can do as it pleases. It is ironic, then, that the US has spent the past 35 years exploiting the absence of the second rule but hasn’t had time to learn the first.

Yesterday, I woke up to this:

“Critical fields” means essentially anything that matters in geopolitics. Like AI.

Well, let me tell you, Mr. Rubio, that given that China is already pulling ahead in strategic areas—and continuously pulls top talent from its STEM-focused universities—you shouldn't respond by pushing out talent you devoted time and resources to educate to work in those very critical fields in China.

Two experts on Chinese-US relations that I respect claim this policy may not be the win Rubio thinks it is but instead a disaster in the making.

One, Arnaud Bertrand, is a French entrepreneur who leans pro-China. The other, Kaiser Kuo, is a Chinese-American rock star who strives to “foster mutual understanding” from his podcast, Sinica. It is interesting that despite their disparate origins and unrelated backgrounds they both agree this is a terrible decision for the US.

Here’s Bertrand:

Here’s Kuo, who called this decision “an own-goal of historic proportions”:

As the child of immigrants who came to the U.S. for graduate (Dad) and undergraduate study (Mom), and as someone who has spent decades trying to foster mutual understanding between Americans and Chinese, I’m struggling to find words that adequately capture just how self-defeating — and frankly stupid — this is. It is hard to think of a better example of an own-goal in the annals of American soft power. Somehow, we just keep doing it.
This will be framed, no doubt, as a national security measure. But no serious person believes this is about genuine risk mitigation. It’s about spectacle. It’s a gesture meant to placate a political base that’s been primed to see every Chinese student as a CCP sleeper agent, and every American university as a hotbed of un-American radicalism. It’s part of a larger campaign that, rather than diagnosing the real ailments of higher education or geopolitical rivalry, seeks only to punish and purge.

The Republican Party is going to hand the Communist Party the win.

A bunch of AI researchers/analysts I follow have also reacted to the news:

If their first impressions are accurate, this is really, really bad. It gets worse once you realize that the American Asian demographic (among them, a majority of Indian and Chinese) tends to top the rankings in STEM competitions. But perhaps the worst part is that the fraction of Chinese top talent working in AI was 38% in 2022 (probably larger by now, although not much longer).

That’s more than the US Americans themselves (37%).

Here’s what Matt Sheehan, a fellow at the Carnegie Endowment for International Peace, told the New York Times about these numbers in 2024:

The data shows just how critical Chinese-born researchers are to the United States for A.I. competitiveness. . . . We’re the world leader in A.I. because we continue to attract and retain talent from all over the world, but especially China.

That the US government wants to improve the education system and the conditions of its native-born citizens is good, but perhaps shooting itself in the one foot that’s keeping it barely standing above its geopolitical enemy is not the best idea.

You guys have a bleak future ahead if this comes to pass. Where do you think those 38% top-tier Chinese-American AI researchers will go next if you push them out of America?

To put it plainly, the only reason China has lagged behind the US in AI for over a decade is that there was a huge incentive for students to graduate and work overseas despite China’s attempts to retain them or bring them back. That’s over.

Next time, don’t push if China is pulling.

REMINDER: The current Three-Year Birthday Offer—20% off forever—runs from May 30th to July 1st. Lock in your annual subscription now for $80/year.

Starting July 1st, The Algorithmic Bridge will move to $120/year. Existing paid subs, including those of you who redeem this offer, will retain their rates indefinitely.

If you’ve been thinking about upgrading, now is the time.

Get 20% off forever

Besides the growing institutional attractiveness of top Chinese universities and the CCP’s long-term strategy to lead in all important fields, the tech sector is also doing a great job of catching up.

In the software front, we have DeepSeek, that young AI startup you should be all familiar with by now. They just gave its reasoning model, R1, a “minor version upgrade” that sent it above Grok, Llama, and Claude over to the podium with o3 and Gemini:

That’s what DeepSeek calls a minor upgrade. In case you’re not worried yet, let me remind you that next-gen models DeepSeek-V4 and DeepSeek-R2 are incoming (it was rumored they would be launched in May, but given that it’s May 30th and we don’t have them yet, that was probably wrong).

So China is rightfully optimistic they will win, and the US keeps validating that optimism. As Google's former CEO, Eric Schmidt, and China analyst Selina Xu wrote in a guest essay for the New York Times in early May, that’s a recipe for victory:

. . . China is at parity or pulling ahead of the United States in a variety of technologies, notably at the A.I. frontier. And it has developed a real edge in how it disseminates, commercializes and manufactures tech. History has shown us that those who adopt and diffuse a technology the fastest win. . . .
In a dozen years, China has gone from a copycat nation to a juggernaut with world-class products that have at times leapfrogged those in the West. . . .
We should learn from what China has done well. The United States needs to openly share more of its A.I. technologies and research, innovate even faster and double down on diffusing A.I. throughout the economy.

Indeed, inventing an incredible technology is cool and the promise of a sustained edge, but only if diffusion and adoption follow. If China, despite the delay in getting into the AI bandwagon, is marching forward faster when it comes to deployment and applicability, then the US’s theoretical advantage won’t materialize.

As I wrote in February, it’s key to notice the sharp attitude contrast between the US (and the EU, although it’s pointless to mention us), with its excessive caution and reactionary outlook at the governmental level—do White House officials even know how to use ChatGPT?—and China:

The Chinese government has taken a clear stance on DeepSeek: Officials must learn to use it—as a means to master AI and large language models (LLMs)—and integrate it into governance. Nationwide deployment. It’s being applied across the board to reduce workloads, enhance decision-making, and accelerate tasks wherever it makes sense. Unprecedented appreciation for this technology at the bureaucratic level. . . .
In times of competition—some would say war, yet to be seen if a cold or hot one—knowing how to march in formation efficiently and with the right attitude matters more than knowing how to vote or debate.

Moreover, the new DeepSeek-R1 has cemented China’s open-source dominance over US companies, which insist on keeping an ironclad opacity over their research.

Fearful that their competitors will steal their secrets, they refrain from sharing them, which may serve them individually in the short term, but guarantees a collective defeat in the long term, further helping China's efforts. Working together beats working alone every time.

In the hardware front, the same story is playing out.

Here’s today’s Financial Times headline:

Arnaud Bertrand is at it again, poking fun at how the export controls seem to have backfired (I’m not sure I agree but can’t deny his take is amusing):

All the Chinese tech giants are currently in the process of switching from Nvidia to homegrown AI chips, forced to do so by US export controls.
Hard to overstate how ironic it is: the entire point of the export controls was to prevent exactly this, the whole of China - the world's largest market for chips by far - adopting Chinese chips at the expense of US firms.

Bloomberg reports that Nvidia CEO Jensen Huang considers Huawei, his competition, to be “quite formidable” now:

“The Chinese competitors have evolved,” he said Wednesday in an interview with Bloomberg Television. Huawei Technologies Co., a Chinese tech company blacklisted by the US government, has become “quite formidable,” he said. . . .
Huang cautioned that the gap between US products and Chinese alternatives is decreasing. Huawei’s latest AI chip is similar to the performance of Nvidia’s own H200 — a component that was state-of-the-art until its replacement in recent months.

All of that is in line with what leading semiconductor firm Semianalysis wrote in an April post:

Huawei is making waves with its new AI accelerator and rack scale architecture. Meet China’s newest and most powerful Chinese domestic solution, the CloudMatrix 384 built using the Ascend 910C. This solution competes directly with the GB200 NVL72, and in some metrics is more advanced than Nvidia’s rack scale solution. The engineering advantage is at the system level, not just at the chip level, with innovation at the networking, optics, and software layers.

That means that even if Huawei is not at Nvidia’s level on a chip vs. chip basis, it’s “a generation ahead of Nvidia and AMD’s current products on the market” in a rack vs. rack basis. (H200s, which Huang referred to in the excerpt above are not the newest GPU generation; GB200s are.) China is not only better at applying and deploying AI innovations but also at scaling them up.

This news about Huawei's chip performance is not really about a Chinese hardware company competing with Nvidia, but about China’s efforts to develop the entire AI vertical at home, both software (models, data, algorithms) and hardware (chips, fabs).

Let me add to what I said before: There are actually two reasons why China has lagged behind the US in AI. The first one is a lack of national talent, but they solved that. The second is that they didn’t have state-of-the-art chips, lacked an in-house TSMC-level fab, and didn’t have cutting-edge extreme ultraviolet (EUV) lithography. They’re now solving all three fronts in parallel.

So far, I’ve only mentioned the chips. What about fabs and EUV lithography technology?

Critically, Semianalysis notes that Huawei’s Ascend chips “can be fabricated at SMIC [China’s analogue to Taiwan’s TSMC],” even if its main production remains at TSMC, which means China is closing the gap to be competitive also at the fabrication level. (Chip companies like Nvidia design the chips but it’s fabs like SMIC and TSMC that manufacture them.)

And, to top it off, the South China Morning Post reported in January that Chinese scientists have developed an alternative method for EUV lithography, which would allow China to eventually compete with Dutch firm ASML, the sole producer and seller of the machines that can provide this technology at the required quality.

So, to sum up, China is now competitive with the US in AI research, development, and deployment, and is improving rapidly in hardware (chips, fabs, EUV lithography). And, at the same time, the US government wants to push out all the Chinese citizens who work in strategic areas, fulfilling China’s long-term plan to bring them back.

I guess the only reasonable conclusion we can draw is that people are correct when they joke that Xi Jinping does nothing and wins, or that the 21st century is painting itself red. If you need me, I'll be learning Chinese.

What this means going forward: 8 Implications

I want to end with a short extra section for paid subscribers with eight implications on what it means in the context of the US–China AI race to push out a third of your top AI researchers, to train your rival’s elite for free, to lose open-source leadership, to give up a critical edge in hardware development, and to pretend you're still ahead just because you were once. The momentum is shifting. Here's how it might unfold.