“Silicon Valley and corporate America are increasingly turning to cheaper, open-source artificial intelligence models built in China.
Is Silicon Valley giving Chinese A.I. a boost?
China is still weighing heavily on the Trump administration’s approach to artificial intelligence, including a clampdown on Anthropic’s top models.
But corporate America isn’t showing as much reluctance about using Chinese A.I. — especially as newer models are quickly closing the gap with their American rivals.
Microsoft may make DeepSeek available for its Copilot Cowork product, which would mean adopting one of the most disruptive Chinese models for a product used by potentially millions of enterprise users.
The tech giant recently told Axios that DeepSeek’s V4 model would be meant as a lower-cost alternative to Anthropic or OpenAI models. It was also fine-tuning the model — possible because DeepSeek’s products are open-source — to try to reduce any biases inherent in the base model.
Developers are already embracing Chinese A.I. Six of the 10 most popular models available on OpenRouter, an A.I. model marketplace, are Chinese, including those from DeepSeek, Tencent and Xiaomi.
Alex Atallah, OpenRouter’s C.E.O., previously told DealBook that cost concerns were driving the greater adoption of Chinese models, many of which are open-source and so free to use.
Bigger companies had been using Chinese models as well: Cursor, the popular coding tool, built its Composer 2 model on Kimi K2.5, which was made by the Chinese lab Moonshot AI.
(Cursor appears to be moving away from Kimi now.)
Chinese models are getting more powerful, too:
Moonshot AI this month announced the latest version of Kimi, citing benchmarks showing that it’s competitive with top offerings by Anthropic and OpenAI.
Silicon Valley has been abuzz in recent days about GLM-5.2, a model by the Chinese lab Z.ai that Z.ai says is as powerful as high-level Anthropic and OpenAI models. Guillermo Rauch, the C.E.O. of the American A.I. developer tool maker Vercel, said he was “genuinely impressed, almost shocked” by its abilities. “This changes things,” he added.
And they’re doing so at a faster rate. Dario Amodei, Anthropic’s C.E.O., had warned this spring that Chinese rivals were about “six to 12 months” behind frontier-level U.S. models.
Chinese developers appear to be taking that as a challenge. In response to a prediction by Elon Musk that Chinese models could become as powerful as Anthropic’s Fable model, Jie Tang, a founder of Z.ai, wrote on X, “won’t take that long.”
How will the U.S. respond? In the spring, the Trump administration said it would crack down on Chinese companies reportedly using American A.I. models to train their own, a process known as distillation.
But further efforts to limit the practice haven’t been announced. Indeed, the administration reportedly held off on blacklisting DeepSeek as a national security risk, according to Reuters, citing unnamed sources.
In other A.I. news: Amazon’s movie division is abandoning plans to make a movie about Sam Altman of OpenAI and directed by Luca Guadagnino. (Amazon has invested in OpenAI and struck a cloud computing deal.)” [1]
1. Is China Closing the A.I. Gap Faster Than Expected?: DealBook Newsletter. Andrew Ross Sorkin; Warner, Bernhard; Kessler, Sarah; Michael J. de la Merced; Gallogly, Niko; et al. New York Times (Online) New York Times Company. Jun 22, 2026.
Komentarų nėra:
Rašyti komentarą