If the U.S.-China artificial-intelligence rivalry were a football game, America would be leading at halftime—but it just made a risky trade.
President Trump said this month that he would allow Nvidia to sell an older but still powerful chip to China, easing restrictions that had kept top American technology away from Beijing.
Politicians from both parties said Trump handed a rival a star player. Supporters called it a strategic long-term play.
But what is the score right now and how does this shift the momentum?
To be clear, this isn’t a game. There’s no fourth-quarter clock, and both sides can claim victory. The stakes are economic and military supremacy, not a trophy. But to demystify the conflict, we asked experts to translate the AI contest into touchdowns and field goals.
The consensus at halftime: U.S.A. 24, China 18.
Our experts disagreed on how close the game is. “Chip War” author Chris Miller called it a comfortable 24-12 lead, saying only U.S. companies are turning AI into money.
Others saw a nail-biter. Deepika Giri, head of AI research at IDC in Singapore, scored it 21-19. America dominates chips, she said, but China is surging with chatbots.
The experts weighed everything from power grids to software to arrive at their scores. Below, we focus on two key factors: chips and chatbots.
Entering the second half, the U.S. holds the lead. But the underdog has the momentum—and after Trump’s trade, a new quarterback.
Chips: the quarterbacks
To understand Trump’s bet on chip exports, consider the 1993 San Francisco 49ers.
The team had the game’s best quarterback, Steve Young, and the legend Young replaced, Joe Montana. Figuring Montana wasn’t worth much on the bench, the 49ers traded him to the Kansas City Chiefs for a draft pick who could complement Young.
The aging veteran in this case is Nvidia’s H200 chip, which Trump recently cleared for export. Released in mid-2024, it is a generation behind the company’s leading-edge Blackwell chip.
A report by one of our experts, Saif Khan of the Institute for Progress think tank, found the H200 to be 16% more cost efficient and 32% more powerful than the best model from Huawei, China’s chip champion.
The gap is probably wider, his group said, because Huawei’s chips are glitchy and scarce, owing to manufacturing bottlenecks.
Meanwhile, the U.S. has the supply chain to crank out superior chips. Khan’s team estimates that next year, the U.S. can produce semiconductors with the aggregate power of 6.9 million top-end Blackwells—more than 40 times China’s capacity.
But allow H200 sales to China, and the U.S. computing-power lead shrinks to less than sevenfold, it said.
National-security hawks warn that an aging Montana is still better than any quarterback China has produced.
Nvidia executives view the field differently. A spokesman for the company, which backed the H200 sales approval, said American customers will continue to receive far more computing and AI capabilities than anyone else.
Company leaders fear that blocking sales forces China to build around Huawei, creating a parallel AI ecosystem the U.S. can’t control. Better to keep China using Nvidia, they reason, so that if a Chinese startup such as DeepSeek makes a breakthrough, it relies on tech that U.S. companies can easily replicate.
They also consider the H200 near-obsolete, packing by some metrics just a quarter of the Blackwell’s power. Why not sell it to fuel the next seismic leap?
As for the 49ers, the gamble worked. Montana had two decent seasons in Kansas City. Young won a Super Bowl.
Chatbots: the receivers
If chips are the quarterbacks, chatbots are the receivers who turn a well-crafted spiral into a touchdown catch.
Chatbots are already doing real work, and both countries’ ambitions go further. U.S. developers say the ultimate goal is artificial general intelligence, systems that may eventually outfox humans by thinking independently. The more top-tier chatbots a country has, the better its odds of getting there first.
The LMArena leaderboard, a global ranking of chatbots, suggests America is running up the score. At one point this month, variants from just four U.S. companies—Google, xAI, Anthropic and OpenAI—swept the top 20 spots.
But the rest of the top 30 is crowded with Chinese heavyweights, including Alibaba, Baidu and rookie sensation DeepSeek.
DeepSeek triggered a $1 trillion market selloff earlier this year by building a world-class chatbot on second-rate Nvidia chips. It was like a receiver snagging a wobbly pass from a backup quarterback, breaking three tackles and sprinting 80 yards for a touchdown.
While U.S. labs are secretive, Chinese companies publish their research to prove they can compete despite inferior hardware, said Barrett Woodside, an AI entrepreneur in San Francisco who previously worked at Nvidia.
“The question is, when the Chinese companies get more hardware and better hardware, do they sprint away from the U.S.?” he said.
In other words, Chinese receivers have been training harder to compensate for a weaker quarterback. What could they achieve with 37-year-old Joe Montana?
Write to Stu Woo at Stu.Woo@wsj.com and Ming Li at ming.li@wsj.com