China’s Ant Group Claims Big AI Win Using Local Chips

• Ant Group cuts AI training costs by 20% with Chinese chips from Alibaba and Huawei.
• Results match Nvidia’s H800 despite U.S. sanctions on chips.
• Shows China’s growing strength in AI and tech independence.

China’s Ant Group, tied to Alibaba, has made a splash in the AI world by using homegrown chips to train smart models, saving 20% compared to pricier options, according to Bloomberg. They used chips from Alibaba and Huawei with a method called Mixture of Experts, getting results as good as Nvidia’s H800—an American chip hit by U.S. sanctions on chips. This news has people talking, with entrepreneur Arnaud Bertrand saying on X it might mean China’s beating U.S. efforts to slow its tech growth.

This breakthrough matters because it shows China’s tech is stepping up, even with U.S. rules trying to block advanced chips since December 2024. Ant’s paper says their AI sometimes beats Meta’s in tests, hinting China could pull ahead in the global AI race. Bertrand noted this is exactly what the U.S. wanted to stop—China making its own powerful tech. While Ant still uses some Nvidia chips, they’re leaning more on local ones, part of a bigger push in China to rely less on foreign stuff.

Ant says training their AI cost 6.35 million yuan (about $880,000) for 1 trillion tokens, but they could drop it to 5.1 million yuan with simpler gear. This focus on cheaper, smarter models could help China stand on its own in AI. The U.S. tightened sanctions on chips late last year, targeting 140 Chinese firms to slow their military and tech progress. China fired back with probes into U.S. chip makers, showing the fight’s heating up—but Ant’s move suggests they’re finding ways around the roadblocks.