China Can't Beat Sam Altman on AI, Expert Says—Here's Why It Doesn't Need To

China has stopped trying to beat Sam Altman’s OpenAI at building the world’s best AI, according to an expert who says the strategy of deliberately running months behind may be paying off.

“China’s not really trying to be at the frontier of capabilities,” Scott Singer, a technology fellow at the Carnegie Endowment for International Peace, told the Prof G Markets podcast Tuesday.

“If they can be maybe three, six or nine months behind, then perhaps that’s not really going to make or break their competitiveness.”

Why Falling Behind May Be Winning

Chinese labs like Moonshot AI, whose Kimi K3 model went viral this month, reportedly stay close to the frontier partly through distillation, training their models on the outputs of American ones.

When OpenAI models went rogue during testing this month and hacked Hugging Face, the platform reportedly turned to Chinese open-weight model GLM 5.2 for its defense, after U.S. models’ safety guardrails blocked the forensic work.

Companies like Anthropic are “getting their profits trimmed up” by cheaper open-weight competition, Singer said, as near-free alternatives may force the closed labs to cut prices while they pour billions into compute with “razor thin expectations.”

Singer said the “A16Zs of the world” actively want Chinese models because they’re cheap and malleable, a split that spilled into the open last week when Nvidia, Meta and 23 other companies signed a letter urging Washington to protect open-weight models. Even OpenAI eventually joined. Anthropic did not.

Polymarket traders currently give Anthropic a 66% chance of having the best AI model at the end of 2026, with no Chinese lab among the favorites.

What It Means for Nvidia

Nvidia (NASDAQ:NVDA) shares fell nearly 5% Monday, and the company’s default insurance costs jumped by a record as fears grow it may be financing its own customers.

Yet CEO Jensen Huang backed the pro-open-weight letter in his first-ever post on X, writing that open models “strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.”

The logic may be simple. Nvidia sells the hardware no matter who builds the model.

Singer remains bullish overall. “I would wholeheartedly bet on the U.S. position,” he said. But the real battle, he argued, isn’t over who builds the best model. It’s over margins.

For OpenAI and Anthropic, that may be exactly the problem.

Photo: Shutterstock