Last week, the energy infrastructure company Hut 8 announced a 15-year, $7 billion deal with Fluidstack to develop a data center for Anthropic, which would be capable of deploying between 245 and 2,296 megawatts of utility capacity. The lease will be guaranteed by Google, which has backstopped Fluidstack leases before and is also an investor in Anthropic. The project, which will be financed by J.P. Morgan and Goldman Sachs, represents another massive shot of debt injected into the A.I. infrastructure market.

Runner-up: Speaking of Google, Alphabet announced an agreement on Monday to acquire Intersect, a data center infrastructure company, for $4.75 billion in cash. Google will also assume Intersect’s debts as part of the deal. Congrats to everyone on the cap table.

And now for the main event…

For the past few years, Silicon Valley and Washington have been largely aligned around the notion that America must win the “A.I. arms race” with China. Fears of a second Sputnik moment deepened in 2017, when China’s State Council published a seminal document detailing Beijing’s plans to achieve global A.I. dominance by 2030. In January, the panic spread to Wall Street after Chinese startup DeepSeek launched a free, open-source L.L.M. whose rapid and seemingly cheap development cost U.S. tech stocks about $1 trillion in the course of a single day. The incoming Trump administration quickly adopted the same geopolitical narrative as the largest A.I. developers: A growth-at-all-costs strategy wasn’t just good for Wall Street—it was also a national security imperative.

Of course, there are plenty of reasons to believe that this framing is self-serving. OpenAI, Google, Nvidia, Anthropic, Microsoft, and Amazon have all massively benefited from the president’s laissez-faire regulatory posture. And, perhaps most importantly, there’s not even a consensus definition for artificial general intelligence or superintelligence—a hypothetical goal without any clear finish line. This contest, in other words, might be reasonably described as a race to nowhere.

Regardless, anxiety about Chinese dominance has become the top political justification for the massive investments and lax regulations fueling the current A.I. build-out. And while there’s no question that Beijing also believes it’s engaged in a technological cold war, it’s pursuing a very different development strategy. So I called up Matt Sheehan, a senior fellow at the Carnegie Endowment for International Peace and a leading expert on China’s A.I. ecosystem, to discuss the country’s approach to the technology: what’s real, what’s overblown, and what we tend to get wildly off the mark. The following has been lightly edited for clarity.

Ian Krietzberg: In the West, A.I. development has become a major economic force. V.C. money has been rolling in at higher and higher levels for years, and public market investors have yet to balk at steadily higher valuations because they’re terrified of missing out on A.I.-related gains. I’m curious, what does the development dynamic look like in China, and what do people get wrong about it?

Matt Sheehan: There’s this idea that China has infinite resources—that they can just plow infinite money into A.I., throw infinite people at it. I lived in China from 2010 to 2016, when money was just sloshing around everywhere. But that’s not the case anymore. They’re very cash-strapped at a bunch of levels, and the V.C. ecosystem has basically collapsed. Since 2020, levels of investment have dropped by around 70 percent. And the number one problem for companies over there is that they don’t have money and can’t make money.

Chinese A.I. companies have a very hard time generating revenue because, generally speaking, Chinese people don’t pay for subscriptions. And Chinese companies don’t buy enterprise software. They don’t have the recurring annual revenue pouring in. So a lot of times, they have to go overseas to make any money at all.

Then what’s the rationale in China for pushing and advancing A.I.? In the U.S., the motivations are clear: an advancement in capabilities hopefully leads to greater adoption, which means more revenue.

It’s pretty diverse across the ecosystem. Some startups are founded by people who are A.G.I.-pilled: The founder of DeepSeek believes in A.G.I., and that they need to push to get there; the founder of another leading startup is similar. [DeepSeek] is probably the closest they have to an OpenAI or Anthropic, at least in terms of mentality. It’s able to survive and thrive because it grew out of an algorithmic-trading company, so they can just print and spend money.

Across the big platform companies, though, there’s a variety of strategies. Alibaba is essentially trying to build A.I. to empower their clouds and attract people to their cloud services. Tencent, which creates WeChat and gaming stuff, wants to make money through in-app advertising. So everyone has slightly different interests.

In terms of where they are compared with the U.S., they’ve mostly caught up. There’s still a performance gap, but their advantage is computational efficiency, which is pretty helpful if your model is cloud services. But if you sat them down, gave them truth serum, and asked, Do you really feel like you need to be at the frontier? Most of them would say, Not exactly.

In the U.S., of course, the China threat is held up as a justification for all the policy frameworks coming from the Trump administration. Do you think these guys actually buy their own narrative?

I talked to Ilya Sutskever at OpenAI in 2019 or something like that, and I talked to people at Anthropic very early on. So it’s been funny to watch their thinking on China evolve, either opportunistically or genuinely. A lot of people will say, All this safety stuff is basically just hype material to drive valuation. I don’t really think that, because I’ve known some of these people since long before there was the chance at these huge valuations, and they were already talking about that before they even started these companies. But some of them have decided that beating China is the best way to curry favor with the Trump administration to get resources, etcetera.

Dario Amodei has been one of the more aggressive ones saying we need export controls because we need to be ahead of China. I can’t open up his brain and say this for sure, but I think that’s relatively genuine. People at Anthropic tend to have very short timelines for A.G.I., and they tend to see two bad paths if China is strong in A.I.: There’s the outcome where China just surges ahead, and they either use it to dominate in America, or they just do it unsafely, and that’s disastrous for whatever reason. And there’s the outcome where we’re neck and neck, we’re racing to A.G.I., and so we start cutting corners on safety, ourselves.

Do you think China’s work in A.I. represents a legitimate threat to national security?

I personally don’t want Xi Jinping to control the most powerful A.I. model in the world. I’m close to China in many ways, and probably see it more positively than a lot of people, but he’s not a good guy. So I’d be worried about a world in which China leads in an unsafe way. I’m also worried about a world where the two countries are neck and neck—in which they both assume the other side is cutting corners, and so they cut corners, and things get dangerous.

I also think there’s a very good chance that, no matter what we do, China is going to be right there with us or maybe even ahead. And how they choose to regulate and govern A.I. is going to really matter in terms of international, global catastrophic risk. So we really need to have a grounded understanding of what they’re doing, and maybe even talk to each other about [these risks].

In the U.S., we’ve seen a very pro-corporate push from the White House to reduce regulation while speeding up development and adoption. How does the private-public dynamic in China differ from the U.S.?

How tightly the government wants to control companies varies over time. From 2020 to 2022, they were super harsh in a variety of ways. The last couple of years they’ve been like, Okay, maybe we need to back off. We’re going to work with them on this. It also varies by company. Huawei is far, far more integrated with the government than even DeepSeek, which was actually a very private company that became the poster child on its own.

Generally, it’s less tightly integrated than people imagine. We think they’re working hand in glove, but the government is actually quite suspicious of private enterprises in China. At the end of the day, though, if they want to seize control, if they want to punish somebody, they will. They just don’t, really, with frontier technology [companies], because they know they actually can’t run it and need the private sector.

What do people get wrong about how China is approaching A.I. regulation?

There’s an idea that China doesn’t have A.I. regulations. That’s just categorically wrong. They have at least five binding ones, and they actually affect the way companies operate: heavy compliance requirements, mandatory reporting, mandatory testing. All of it [helps] the C.C.P.’s goal, which is information control. There’s also an idea that, Okay, they have A.I. regulations, but it’s an authoritarian state. So, it’s just whatever Xi Jinping thinks. Actually, Xi Jinping doesn’t have a granular handle on ensuring that models don’t stray from the party line, or on how datasets should be treated or how outputs should be labeled.

China also has a super sophisticated A.I. policy community. In many ways, it mirrors the U.S. community, except it’s like a fun-house mirror because the outputs have to work their way through an authoritarian bureaucracy. But they rely heavily on academics and scientists. Companies lobby. Media and public outcry forces regulatory interventions. This landscape makes it possible to better predict where China is going on A.I. regulation, because you can actually read about and even talk to A.I. policy people over there.

Finally, I would never say we’re directly influencing Chinese A.I. policy in terms of dictating to them, but they are hugely interested in what’s going on in American A.I. policy. They know a ton and follow our A.I. policy extremely closely. They’re always taking ideas and working them into their system.

That’s all for today. Hope you all enjoy this short holiday week. I’ll see you next Tuesday.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006