Chinese models are the latest boogeyman for the AI trade, and investing veteran Ed Yardeni see problems ahead for US regulators trying to contain them.
The summer has been marked by the rise of Chinese AI models like as Moonshot AI’s Kimi K3, putting Silicon Valley on high alert for new competition. Yet, as Yardeni noted, Washington, DC, might struggle to keep a lid on models from China’s tech companies.
“Short of a ban on foreign open-weight models, Washington is using the tools it can to prevent distillation and data theft,” he wrote. “That mostly involves security advisories, limits on frontier US closed-weight models, and congressional pressure on corporate adopters. But so far, those actions have been toothless given the rapid adoption of Chinese open-weight models.”
US chip stocks fell on the news that the Chinese AI startup had released Kimi K3, calling to mind the DeepSeek selloff of January 2025. However, DeekSeek hasn’t proven to be the threat that investors saw it as at the time, and the Kimi K3 scare has also mostly subsided in recent weeks.
However Yardeni notes that doesn’t mean the rise of newer models isn’t a risk for US AI interests.
“Adoption is moving faster than policy,” he said of China’s AI tech.
In Yardeni’s view, that could pose a threat to US AI dominance if regulators can’t react quickly enough.
He cited Anthropic suspending access to its Fable 5 model after a dispute with the White House as a recent stumble in Washington’s approach.
While Anthropic has since restored access, Yardeni said that the momentary pause gave Chinese models an opportunity to fill the void left in the market. He added that one of Washington’s greatest challenges may be its inability to govern open- and closed-weight models the same way.
Anthropic has recently drawn criticism for executives’ refusal to embrace open-source AI. It remains the only AI leader to not sign the “Open Weights and American AI Leadership” letter supporting the development of open weight-AI technology.
This debate underscores the dynamic that Yardeni identified as posing a threat to Washington’s quest to regulate these Chinese AI models.
“Such restriction on open-weight models is ultimately impossible because there’s no domestic cloud ‘on/off’ switch once the models’ files are privately downloaded,” he stated. “And open-weight models that start out hosted can quickly pass through Washington’s window of control.”
