Anthropic urges the U.S. to move quickly to establish a one- to two-year AI advantage over China.
The firm highlights U.S. superiority in compute resources and pushes for policies to seal gaps allowing chips to reach China.
Anthropic emphasizes that the U.S. must "avoid squandering our advantage."

Anthropic asserts that the United States can solidify a one- to two-year lead in AI development against China - but swift action is essential.

In an extensive Thursday blog post, the company outlined two contrasting visions for the AI field by 2028: a scenario where the U.S. curbs China's access to American AI computing power, and one where no such barriers exist.

Anthropic explained that China is narrowing the AI divide with the U.S. via lax chip export regulations and distillation techniques, where advanced AI models help create more compact "student" versions.

"If the U.S. and its allies address both challenges promptly, it may be feasible to secure a 12-24 month advantage in cutting-edge AI capabilities," the company stated.

Anthropic stressed that "the window of opportunity to lock in that lead will not necessarily remain open for long."

Such a U.S. lead is vital for ensuring AI safety, according to the firm, which noted that a close contest between U.S. and Chinese AI developers could hinder safety protocols and governance initiatives from industry and governments alike.

Intense rivalry would push labs in both nations to rush model releases without adequate safety evaluations, it continued.

The company advocated for tougher chip export rules, increased funding for enforcement, and measures to block distillation efforts by Chinese AI firms.

"Our prior achievements mean our current goal is mainly to prevent wasting our edge: to refrain from helping the CCP catch up," Anthropic wrote, alluding to China's Communist Party.

Back in February, Anthropic accused three leading Chinese AI outfits - DeepSeek, MiniMax, and Moonshot AI - of improperly leveraging its Claude model to advance their own systems.

The Biden administration initiated U.S. chip export restrictions to China in 2022. The Trump administration expanded these by barring Nvidia and AMD from direct sales there, though it eased up last August by permitting Nvidia's H200 chips with a 25% sales tax paid to the U.S. government.

Meanwhile, illicit efforts by Chinese entities to obtain U.S. chips continue underground. Last December, U.S. authorities indicted individuals for smuggling Nvidia's top-tier chips after disguising them as "SANDKYAN" variants.

Challenging Anthropic's view on China closing the gap, former ByteDance engineer Zhang Chi argued in April that Chinese AI is lagging even more.

Now a research scientist and assistant professor at Peking University, Zhang shared in a podcast that China's AI efforts suffer from scarce high-quality training data and limited access to elite chips.

Anthropic released its analysis the same day President Donald Trump held talks with Chinese President Xi Jinping - the first such U.S. presidential visit to China since Trump's 2017 journey. Trump was accompanied by top executives like Tesla's Elon Musk, Apple's Tim Cook, and Nvidia's Jensen Huang.