Aravind Srinivas just explained why China’s open-source AI may become more powerful than ever. And why Anthropic has lobbied very hard for export control. "the only reason why there is even a 12-month gap between open source and frontier models is export controls. But there is a chance that,... because of that, they now get really good at the physical layer. One advantage they (China) have is that they can actually build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem. And so, by forcing them to go out there and build all this, you are converting them into a far more potent competitor." --- From "20VC with Harry Stebbings" YouTube channel (Harry Stebbings ), link in commentshow more

Chinese Open Source models are realle awesome. It's so exciting that they almost meet the frontier models in most of the benchmarks and still the user won't have to pay for its usage. A more important angle to all this situation is that these open source models would put heavy pressure on firms like OpenAI and Anthropic to innovate faster and try to reduce costs of token usage. The AI race can soon become a cost competition as intelligence and benchmarks stop becoming the bottleneck.

Perplexity own business depends heavily how good Chinese models are. If they have to pay open Ai and anthropic for every query, they are cooked as it will be cheaper on their respective website. For Chinese models they mainly have to pay hosting cost, no model cost

China will have the last laugh on AI battle? $KWEB

💯 💯

Export controls might inadvertently accelerate China's AI infrastructure development, creating a more self-sufficient and formidable competitor in the long run.

这波反而把工程能力逼出来了

China will have models that train models that train models that train models that train models..

China lost the AI race on Saturday, when I told Anthropic they could just poison the data stolen in Chinese distillation attacks to destroy the AIs built on them. I got into the details last night. They can build on their own merits, or not at all.

everyone measures model gaps, fewer people measure infrastructure learning curves

jensen was right

Srinivas's actual point doesn't need the extra framing tacked on, restricting chip access didn't slow China down, it forced them to get fast at the one thing the US still struggles with: building power and data centers quickly. The barrier became the training ground. One thing worth flagging though: the line about Anthropic lobbying for this isn't something he said in the interview, that part's the post's own addition.

Export controls may inadvertently turn China into an even stronger AI competitor.

Open-source AI has always found ways to close the gap. The more barriers that emerge, the stronger the incentives become to build independent infrastructure and alternative ecosystems. Competition ultimately benefits innovation when knowledge continues to circulate

He is of course completely right. Another dimension is that China shares its models and can benefit from review by the top engineers in the US, the opposite is not true. They see their models are stepping stones, not as the final thing.

Srinivas is describing a trade that looks smart on a five year horizon and reckless on a fifteen year one. The control gap bought the US a temporary model quality lead while China was forced to master the one bottleneck that now constrains American AI buildout more than chip supply does, permitting, power, and construction speed. Multiple US hyperscale projects face interconnection delays exceeding four years and transformer lead times near two years, while Chinese data centers go from permit to power in a fraction of that. The export controls did not just narrow the model gap. They handed China a structural advantage in the layer that determines who can actually deploy what they build.

America could have open source AI companies too, but they allowed Scam Altman to run it instead.

That 12-month gap really highlights how policy can’t keep up with tech. If export controls loosen, does Anthropic lose its edge—or are they betting they can stay ahead on capabilities before China catches up on scale?

Export controls have been effectively circumvented without consequence for too long.

The Perplexity CEO is such an disappointment. He has same silly logic as Jensen Power is not problem, Labor is not problem, People is not problem. Everything is Free in China? So they can provide Free AI model to the world forever? Pure Communist propoganda

They can both piss off

Talent origins reveal China’s structural edge. Long-term investment in education and ecosystem is a multiplier that compounds in AI.

What’s his self interest in all this, I wonder

Labor isn't a problem? Sure.

🤣 It's not even close. Maybe some day they'll catch up.

watched open models win demos, then fail evals on boring customer edge cases

Funny that China has less than 500 data centers compared to few thousands in U.S. Argument for weak people

We need import control as well.

Nobody seems to be asking why China is only building 1/10 of the data centers the US is even though they can build them so much faster and cheaper than we can. Is it possible they know something we don’t?