Anthropic said the United States can solidify its AI lead over China for a year or two, but only if it acts quickly.
In a lengthy post Thursday, the company laid out two possible scenarios for the AI landscape in 2028. One scenario is where the US restricts China’s access to US AI computing, and the other is where the US does not.
Anthropic said China is trying to close the AI gap with the United States through lax regulations on chip exports and a distillation offensive that uses developed AI models to train small-scale “student” models.
“If the United States and its allies act now to address both issues, they could secure a 12- to 24-month lead in frontier capabilities,” the company wrote.
Anthropic added that “the window of opportunity to secure a lead does not necessarily stay open for long.”
The company said it is critical that the U.S. takes the lead in making AI safer, adding that “a close race between U.S. and Chinese AI labs could further complicate industry and government-led safety and governance efforts.”
With such intense competition, AI research institutes in both countries are feeling even more pressure to release new models faster without properly testing safety measures, it added.
Mr. Anthropic called for policy changes to tighten chip export controls, increase enforcement budgets, and thwart distillation attacks by China’s AI labs.
Referring to the Chinese Communist Party, the company wrote: “Our past successes mean that our current challenge is primarily to decide not to squander our advantages, and therefore not to facilitate the Chinese Communist Party’s catching up.”
Anthropic issued a statement in February saying that three of China’s largest AI companies, DeepSeek, MiniMax, and Moonshot AI, were “illegally” using Claude to advance their models.
The Biden administration imposed restrictions on exports of U.S. chips to China for the first time in 2022. The Trump administration further tightened those regulations, banning Nvidia and AMD from selling chips to China. But last August, the company partially reversed that decision and allowed Nvidia to sell its H200 chips if it paid a 25% levy on sales to the U.S. government.
And underground operations by Chinese attackers seeking access to U.S. chips continue. In December, the United States indicted several people for trying to smuggle Nvidia’s cutting-edge chips, which they dubbed the “SANDKYAN” chips.
Contrary to Anthropic’s claims that China is closing the AI gap, former ByteDance engineer Zhang Chi said in April that China is actually falling further behind.
Zhang, now a research scientist and assistant professor at Peking University, said in a podcast interview that China’s AI lacks high-quality data to train models, as well as access to advanced chips.
Anthropic’s post was posted on the same day that President Donald Trump met with Chinese leader Xi Jinping, making it the first visit by a U.S. president to China since Trump’s visit in 2017. President Trump has brought in U.S. business leaders such as Tesla CEO Elon Musk, Apple CEO Tim Cook, and Nvidia CEO Jensen Huang as his entourage.