Chinese AI models challenge US dominance as tech gap narrows

The U.S. is the global leader in artificial intelligence, but a new report shows China is rapidly expanding its development.

“China is moving incredibly quickly and could absolutely catch up if the United States falls off its game,” said Gregory Allen, senior advisor with the Wadhwani AI center at the Center for Strategic and International Studies.

According to Stanford’s AI index report, the U.S. had 40 notable AI models in 2024, China had 15 and Europe had 3. While the U.S. is far ahead in quantity, its advantage in quality has been closing quickly.

OPINION: AMERICA MUST WIN THE AI RACE — AND PREPARE FOR THE WORST

“I think that winning the AI race means that the U.S. will continue to be the dominant power in the world in the 21st century,” White House AI and Crypto Czar David Sacks said. “It’s true that since the launch of DeepSeek, I think we’re only maybe three to six months ahead of China at the AI model layer.”

Sacks says the AI model layer is just the surface of a larger stack of competition, but China is showing the most improvement in that layer. Models are frequently assessed in various tests, including languages, math problems, general reasoning and coding. All platforms have improved in recent years, with jumps of 18.8% in general reasoning (MMMU) and 48.9% in PhD level science (GPQA). Coding accuracy (SWE-Bench) jumped from 4.4% in 2023 to 71.7% in 2024.

David Sacks, the White House AI and Crypto Czar, looks on as President Donald Trump sign a series of executive orders in the Oval Office of the White House on January 23, 2025, in Washington, D.C. (Getty Images)

“AI is very, very good at learning how to be good at any given test,” Allen said. “The same challenge that you have in college entrance examinations—are you measuring how capable a student is in their overall intellectual potential? Are you measuring how good they are at studying for standardized testing? There’s a similar version of that phenomenon when it comes to AI.”

Experts expressed concern about how quickly Chinese models have improved in recent years. At the end of 2023, U.S. models performed better 17.5% of the time in language understanding tests (MMLU). Chinese systems narrowed that gap to 0.3% in 2024. In General Reasoning testing (MMMU), the U.S.’s 13.5% advantage fell to 8.1% in the same period. In a more advanced language test (MMLU-Pro), Chinese Platform DeepSeek-R1 posted a higher score than any American-made model.

“When it comes to where the U.S. stacks up in global competition, I think the far more relevant figure is in global adoption of the technology,” Allen said.

CHINA EXPERIMENTING WITH BRAIN-COMPUTER INTERFACES IN GLOBAL RACE FOR AI DOMINANCE: REPORT

According to search engine optimization firm Backlinko, DeepSeek is estimated to…