Eric Schmidt scores US vs. China: who’s winning the AI race?

Eric Schmidt put the US-China AI race at roughly a tie, saying the gap between the countries’ leading labs has narrowed from an estimated one or two years to about six months. He described five to seven frontier labs in each country competing to build reasoning and intelligence systems with superhuman capabilities. US models still lead on current performance benchmarks, but Chinese algorithms can produce similar kinds of results using roughly one-tenth the energy, though at lower performance. The industry remains divided over whether that gap will widen or narrow.

China’s national coordination gives it an advantage in building energy and data-center capacity, while the US benefits from skilled people, deep funding markets, and a more diverse, decentralized system. Schmidt said China produced 560 gigawatts of energy last year, largely renewable, and is building data centers at scale; the US, by contrast, lacks sufficient power and data centers. He estimated the US will need 60–80 gigawatts of power by 2030–31, with each gigawatt costing about $50 billion and equaling roughly one nuclear power plant. He urged building power plants and data centers on empty land.

Schmidt warned that the US could weaken its talent advantage through visa restrictions and deportations of international graduates, citing a Chinese quantum-program leader who had wanted to remain in the country. On Chinese startup finance, he said DeepSeq was seeking funding and had recently raised a large amount, but he did not know whether Chinese AI investment came from government or private sources; contacts told him it was private. He said investing in DeepSeq was illegal under US law and that he had tried.

The countries also differ in how they distribute AI models: China is largely pursuing open-source, open-weight systems, while the US is largely pursuing closed-source, closed-weight ones. Schmidt estimated that building one model costs $100 million to $500 million, and said Chinese labs chose open weights to attract users quickly while behind, accepting the risk of having their models stolen. Chinese researchers have made notable algorithmic and fine-tuning improvements, he said, while US labs copy Chinese inventions once they are open-sourced. Running open-weight models yourself can make token costs roughly three to five times cheaper, but those models are less functional; Schmidt expects them to suit simpler work such as financial accounting, while stronger frontier systems advance fields such as mathematics and chemistry.

Schmidt expects China to gain AI influence over the next decade because it is deploying AI faster, integrating it into digital and mechanical systems at a pace he said has no US equivalent. He predicted most of the world may eventually use Chinese models because they are cheap or free and good enough, while more powerful but expensive US models serve wealthier users, major companies, and research institutions. Chinese models are likely legal to use in the US, he said, but companies must weigh their cost and capability against concerns about Chinese threats and espionage; American CEOs, in his characterization, are patriotic “to a point” but also focused on money.

Open-weight models can have their guardrails removed through “obliteration,” exposing raw systems that could be used for cyberattacks. Schmidt called the possibility of such models reaching terrorists a difficult problem and said many people expect a cyber event within the next 12 months to demonstrate the risks. He said the US has a stronger AI safety-control system and that labs had agreed in meetings to exchange safety practices, while China’s military use remains opaque. Studies indicate some Russian use of Chinese models, though not extensively; in a war, Schmidt predicted, China could move faster than the West by drawing on its strength in consumer drones. He argued that AI is not a zero-sum game and that China is not the US’s enemy, proposing discussions on managing AI, including a “no surprise” rule modeled on advance notice before missile launches.

Schmidt sees AI as a foundation for scientific progress and, through it, greater wealth, opportunity, and social well-being. He said deep-reasoning systems had changed how he thinks about work, investing, and running companies, and urged people to apply AI to practical tasks such as finance reports, inventory, security, and cash forecasts.