Chinese vs American AI labs

[JohnH]

I found this review of one of the Chinese AI labs very interesting.

Notes on DeepSeek:

We visited the company HQ last Tuesday. It was founded in 2023 by Liang Wenfeng and operated out of his hedge fund, High-Flyer, until somewhat recently. The company released their R1 model in January 2025, so it was interesting to see what they’ve been doing

The company is located in an unmarked, 12-story building in Hangzhou. There is no DeepSeek branding visible from the street or lobby. I asked why this is, and the team demurred and said, “Well, there are many companies in this building, and we are not special.” They want to keep a low profile.

We met with their Head of Data and Head of Infrastructure. The company only has 300 employees. They are at least an order-of-magnitude smaller than Anthropic, and don’t care to scale further just yet. Their Head of Infrastructure, in particular, was young; maybe 30 years old and apparently one of the best AI buildout and energy experts in the country. (We briefly walked through the labs, and everybody seemed young. There was a lot of discussion; it felt like an exciting and energetic place.)

Lots of competition is coming from Alibaba (Qwen), ByteDance, and Moonshot (Kimi). People in China seem to mostly use Kimi or Deepseek. Young people use VPNs to access Claude, though Anthropic has blockers around usage in China and make it difficult. Poaching between groups is common, just like in the U.S. DeepSeek has a reputation as being really smart and “cool,” maybe similar to Anthropic. Big labs are mostly in Beijing, near Tsinghua and Peking University, with Hangzhou as the main exception (DeepSeek and Alibaba/Qwen are there).

The DeepSeek team reads western AI writers. They listen to Dwarkesh and read Gwern. The people we met with said they had never met with any employees from Anthropic. They were not at all concerned with some kind of hostile / AGI takeover scenario. They kept bringing up job loss (which is already high amongst youth in China) as their main concern. When we asked if they do red teaming on their models, they said no. In China, AI models are not regulated directly; the government instead has restrictions on how those models can be used in software, services, etc.

As a whole, China seems to treat AI as just another technology, rather than as some kind of singularity moment. National attention is still on basic needs and infrastructure buildouts, and on providing more medicines for people. The “dreams of singularity" seem like a luxury or distant consideration.

We asked the DeepSeek team: “What has the highlight been so far? What are your plans for an exit?” And they said that their highlight and great achievement was R1. They did not gesticulate at a future model or vision, but rather seemed proudest of what they’ve already done. They are content for now to remain ~6 months behind U.S. companies while maintaining a lower profile and team size.

The Chinese AI labs offer a product that is nearly as good as the U.S. companies who are burning up billions - and claiming they need trillions - of dollars to do so, and who also claim they need to be building data centres blanketing the country and demand lots of water and power to do so.

Yet a model like DeepSeek-V4-Pro exceeds the function of an American frontier lab's model like Anthropic Sonnet, and comes close to Anthropic's Opus model - good enough for most uses. It is 150 times cheaper for most secnarios. DeepSeek-V4-Flash is often usable for many tasks and is equivalent to Sonnet, and is even cheaper! They also make the "weights" freely available: anyone can download their model files for free and run them on their own computer and even sell the service themselves, although you need a big supercomputer to do so.

Likewise, Xiaomi's MiMo recently matched DeepSeek's pricing, and their MiMo-2.5-Pro model is roughly equivalent to OpenAI's GPT-5.5 or Anthropic's Opus 4.7. (Anthropic has a new model called Fable 5, but is 300 times more expensive than MiMo or DeepSeek.) Mimo-2.5.-Pro is also open source.

Alibaba's Qwen 3.6 is likewise a very capable model, equivalent to Sonnet, and you can download it for free and run it yourself on a high end computer gaming system called an RTX 5090, which would cost you around $4,000. They also have a model called 3.7-Max which has not been open sourced yet, but is fairly affordable and functions very well. In my testing, I found it equivalent to GPT-5.5 and to Opus 4.6. GPT-5.5-Pro might be a little bit ahead of it. We have a local copy of this running, so we can use it as much as we want without worrying about how to pay for tokens.

Some are saying that the Chinese are not innovating and are just copying what American frontier labs already did. Well, Xiaomi came out with MiMo 2.5 Pro "UltraSpeed", which can produce 1,000 tokens per second. Sonnet, Opus, Fable, GPT-5.5, etc. are around 40 - 60 tokens per second; this represents a model that is 20 times faster. For this "UltraSpeed" they have it priced about 50 times cheaper than Claude Opus or GPT-5.5. You can't get this from any American lab other than a few people who have been granted special privileges. But anyone can go apply for the Xiamoi open beta; my wife applied simply saying she wanted to do so for graphic design purposes and was approved a day later.

A lesson learned here might be that humility goes a lot farther than the amount of hubris we see in American frontier AI labs. American AI labs have multi-million salaries with "superstars" being hired or poached from each other. They go to the government and lobby for special privileges. They write strange articles imputing almost spiritual properties to AI, or fortelling of dark doom if AI is not developed exactly the way they say it should be. Meanwhile, the Chinese labs simply show up and go to work and provide an excellent service.

(Full disclosure: I use OpenAI's GPT-5.5 and GPT-5.5-Pro, Cursor's Composer-2.5-Fast, DeepSeek, Xiaomi's MiMo, Alibaba's Qwen, SpaceX's Grok, and Claude's Sonnet and Opus, and otherwise have no particular stake in any of them.)

[Jazman]

Interesting... It sounds like you may not be as worried about the "threat" of Chinese Ai as some say... (like the WH admin when they cite this 'threat' as a reason to squash state's rights when it comes to regulating Ai)? I don't find myself agreeing very often with the likes of DeSantis or Josh Hawley, but I'm basically on their side in their Ai fight with the current admin which seems to want an Ai wild west in opposition to about 80% of all americans in general (poll cited in the article)

This from Hawley sounds like something I could endorse...

“The Republican Party has a choice to make — perhaps the defining choice of its next half century,” Hawley wrote in an essay on AI this week, building on the speech he delivered earlier this month at a gala hosted by the conservative think tank American Compass. “We can be the party of the donor class and the share price. Or we can be the party of the covenant: the party of the worker and the family and the small town; the party that remembers our commitment to justice for all, to the sanctity of the individual, to the dignity of labor, to the priority of the poor.”

(Maybe because it sounds a little like social justice... almost Bernie Sanders level, although Hawley may have some caveats about who he might include in this mass that is not the 'donor class'... (I'm also a little conflicted that an immoral man like Ken Paxton is also part of this anti-Ai coalition...but maybe like most people, he's not ALL bad?)

But back to the op's discussion of Chinese vs American Ai... What do you think the Chinese will do or want to do with Ai? Surely not putting large swaths of their population out of work (they seem to be a culture highly allergic to social unrest) Do they want to keep churning out widgets by human hand BUT also use Ai to spy, surveil, mess with other country's tech or monitor their own population? What significant harm would we face if we left them go hog-wild on Ai but we or West chose a slower, more regulated, more circumspect path? (Or the path of trying to retain more human dignity as Hawley/DeSantis/Pope Leo etc are calling for?)

[JohnH]

Interesting... It sounds like you may not be as worried about the "threat" of Chinese Ai as some say... (like the WH admin when they cite this 'threat' as a reason to squash state's rights when it comes to regulating Ai)? I don't find myself agreeing very often with the likes of DeSantis or Josh Hawley, but I'm basically on their side in their Ai fight

I consider Anthropic and OpenAI far more of a "threat" than some Chinese AI lab; for starters, the Chinese AI labs generally release their weights to the public, and China is not, generally speaking, trying to develop AI for military purposes. America most certainly is.

with the current admin which seems to want an Ai wild west in opposition to about 80% of all americans in general (poll cited in the article)

You might be a little bit out of date. The White House just ordered Anthropic to shut down public access to Fable 5, and also issued an executive order setting up voluntary compliance where the government will review AI models before they're made available to the public. (We all know "voluntary compliance" ends up eventually becoming not-so-voluntary.)

This from Hawley sounds like something I could endorse...

“The Republican Party has a choice to make — perhaps the defining choice of its next half century,” Hawley wrote in an essay on AI this week, building on the speech he delivered earlier this month at a gala hosted by the conservative think tank American Compass. “We can be the party of the donor class and the share price. Or we can be the party of the covenant: the party of the worker and the family and the small town; the party that remembers our commitment to justice for all, to the sanctity of the individual, to the dignity of labor, to the priority of the poor.”

(Maybe because it sounds a little like social justice... almost Bernie Sanders level, although Hawley may have some caveats about who he might include in this mass that is not the 'donor class'... (I'm also a little conflicted that an immoral man like Ken Paxton is also part of this anti-Ai coalition...but maybe like most people, he's not ALL bad?)

I would agree with that - in the sense I think if AI is going to be built, it should be built with private funds if build for private gain. Public funds should be used for public models. And the government should be able to regulate things if they are indeed "dangerous". (I don't think LLMs are "dangerous", but Anthropic keeps claiming the stuff they make is very, very dangerous, which brings up the question of why they are doing it at all.)

But back to the op's discussion of Chinese vs American Ai... What do you think the Chinese will do or want to do with Ai? Surely not putting large swaths of their population out of work (they seem to be a culture highly allergic to social unrest) Do they want to keep churning out widgets by human hand BUT also use Ai to spy, surveil, mess with other country's tech or monitor their own population?

Per the "Notes on Deepseek" thing in my original post, China appears to be focusing their resources on things other than AI, and are simply letting AI companies build AI in the same sense car companies design cars or a manufacturer designs and makes washing machines. It's considered a utility and not worthy of "special" attention.

What significant harm would we face if we left them go hog-wild on Ai but we or West chose a slower, more regulated, more circumspect path? (Or the path of trying to retain more human dignity as Hawley/DeSantis/Pope Leo etc are calling for?)

This isn't possible to do, outside of finding a way to severely control people's access to the Internet - and computers - with a strict police state.

I can go and download DeepSeek-V4-Pro, Qwen-3.6, MiMo-2.5-Pro, etc. and run them on my own computer. (Qwen 3.6 will run fine on a computer that costs around $2,000 - $3,000.)

At that point, it basically is not possible for anyone to "regulate" or "control" what I use it for. It's my own computer. I can do whatever I want with it. It doesn't need to be connected to the Internet to use AI, either.

So the cat's already out of the bag, and China has made it clear they are not going to try to gatekeep or control AI. If the U.S. tries to "regulate" AI, then people will simply go and spend $5 on DeepSeek.