There’s no winner nor the finish line in the US-China AI race, an expert says

AI expert Alvin W. Graylin says that even if the US or China achieves artificial general intelligence (AGI), it wouldn’t use it offensively, as it would abstain from a nuclear attack.

I spoke to Graylin, a professor of AI policy at the University of Washington and a senior fellow at the Asia Society Policy Institute’s Center for China Analysis, during the Web Summit in Lisbon, one of the world's largest tech events focused on AI and global tech shifts.

The interview took place days after US intelligence agencies concluded that Nvidia CEO Jensen Huang was targeted by China’s espionage campaign through Alice Huang, with whom he had an extramarital affair, raising major questions about the future of AI.

Graylin says the spying allegations will lead the US to impose stricter export controls and intensified geopolitical competition.

“It will have a negative impact on the world, as the US and China need to work together, particularly on AI safety,” he tells Cybernews.

China is making its own advanced chips

The Donald Trump administration greenlit the sale of Nvidia’s advanced H200 chips to China in 2026, as Beijing currently lags behind the US in terms of computing capabilities.

Alice Huang played an important role in facilitating chip sales and allegedly helped smuggle 240,000 banned chips into China, according to The Information.

Graylin says that Chinese companies are improving rapidly, attracting more talent and dedicating more resources, so China may be able to create both AI chip capacity and capability in 4 to 5 years.

Meanwhile, China’s energy capacity – higher than anywhere in the world – may also play a role in catching up with the US.

The country aims to expand its computing capacity more than sixfold by 2030 and to supply the expanding network of data centers with cheaper power from renewable energy sources.

“Once they do get compute capabilities, the advantage the US has right now may diminish, making it down to more of a race over innovation versus a race of the combination of innovation and compute,” Graylin tells Cybernews.

So far, the gap between Chinese and American AI models may be closer to 1 year, he adds, because the US labs decided to keep more capable models to themselves.

Meanwhile, Chinese companies are under pressure to release the models relatively quickly to demonstrate that they can keep up with the American rivals.

Nevertheless, the release of Kimi K3, an open-weight model by China’s Moonshot AI in July 2026, sparked anxiety across Silicon Valley, as the model is considered to match the capabilities of leading US models at much lower cost.

AGI wouldn’t be used offensively

While Huang’s suspected spying case once again shed light on the AI race between the US and China, Graylin says that AI is a global public good and shouldn’t be something to be competed for.

There is no winner in this race. There is no finish line. It’s a continuously improving thing,

Graylin says.

In the US, leaders of AI labs and the government repeatedly emphasized the need for AGI – a loosely defined type of AI that matches human intelligence – before China does.

Americans fear that Beijing could weaponize the most advanced AI against the US, including in the military field. Critics, however, say tech companies use the narratives of AI doomsday and competition to avoid domestic regulation.

Graylin says Washington holds the misconception that technology is the key to global domination or to maintaining the supremacy the US has had for the last 70-80 years.

He adds that even if one of the countries achieves AGI, leading to artificial superintelligence (ASI), they wouldn’t want to use it offensively, as it would trigger a chain of escalation and wouldn’t end well for anyone.

“This is the reason we haven't had another nuclear weapon used offensively since 1945. I think those same forces still hold whether or not you have an advanced AI,” Graylin says.

Advice to Europe: stop trying to reinvent the wheel

As the race between the US and China intensifies, Europe is caught in the middle, with only one company, Mistral AI, capable of creating frontier AI models.

At the same time, Europe is seeking ways to reduce its digital dependence on the US and China amid fears of espionage and a potential “kill switch.”

Mistral CEO Arthur Mensch recently addressed Europeans, saying that now is the time to choose between building the frontier models or surrendering.

He warned that relying on foreign AI tech will cost Europe “what remains of its place in the world” and lock the continent into a regime of “economic dependence unprecedented in the history of civilizations.”

Graylin argues that unless Europe is ready to spend “hundreds of billions” of dollars, it won’t be able to match or significantly surpass China’s and the US’s AI capabilities.

Instead, Europe could focus on using, diffusing, and customizing models with open licenses, rather than “reinventing the wheel.”

Graylin adds, “It costs almost no money and can be diffused directly into the economies of every country in Europe almost immediately. The real value of technology has always been how well it diffuses into your society and the productivity of your people, not who invented it.”

Revolut, an EU-regulated neobank, recently said it moved approximately 2,000 engineers from Anthropic’s Claude Code and OpenAI’s Codex to the self-hosted GLM 5.3 model developed by China’s Z.ai, resulting in 90% lower coding costs.

Graylin says that as open-source models – whether developed by Chinese companies or not – are becoming more capable and significantly cheaper, it “doesn’t make sense” to pay a lot for frontier models, especially because closed models can’t guarantee that data isn’t stored.

Nor do most people need frontier models.

He adds, “Most countries and companies can do very well with models that are 2 to 5% less, and it is already more than most people need. So to spend extraordinary amounts of money on investment in CapEx to catch up on something that does not necessarily make the world more fragile.”