The Two Sides of the U.S.-China AI Race: Innovation and Diffusion

“Slowing down is not an option.”

That is Nobel laureate Professor Michael Spence’s assessment of the artificial intelligence race.

At Primavera Capital’s 2026 Global Investment Conference, Professor Spence joined Primavera Capital Founder Dr Fred Hu for a conversation spanning AI, global trade and the dilemmas facing middle powers. Running through these seemingly disparate subjects was a pragmatic view of a more fragmented world: competition is not going away. The challenge is to manage it while preserving room for cooperation.

On AI, Spence drew attention to two distinct dimensions of the race. One is about pushing out the frontier of intelligence. The other is about how quickly AI can diffuse through the economy. The first determines who leads at the frontier; the second, whether AI translates into productivity and growth.

 

The race at the frontier

In Spence’s view, frontier AI has become overwhelmingly a contest between China and the United States. Neither country will allow itself to become “a distant second”, and he rejects the idea that AI will necessarily be a winner-take-all race. By the measures he has seen, he said, whatever gaps once existed between the two countries have now closed.

Much the same logic applies to companies competing at the frontier. In game-theory terms, Spence calls it an “arms race”. For leading technology companies, the cost of investing too much is small compared with the cost of falling behind. The result, he argues, is likely to be an excessive investment boom.

AI may be attracting too much capital, and its rapid development brings genuine safety risks. But neither is likely to stop the race.

Spence’s answer is therefore not to ask China and the US to stop competing, but to identify the most dangerous risks and treat them separately. The two countries will continue to compete, including over semiconductors. Yet they could establish a framework for limited cooperation on extreme risks that could have catastrophic consequences for both sides.

“When it comes to mutually destructive, real risks,” he said, “I think we can agree to have a conversation with each other.”

It is a deliberately pragmatic proposition: competition where competition is inevitable, cooperation where the risks are too great to ignore.

 

The race to diffuse AI

For an economist, however, technological leadership is only part of the story.

Spence argues that productivity growth and the diffusion and adoption of AI are “flip sides of the same coin”. If adoption remains concentrated in a handful of sectors while large parts of the economy remain untouched, even dramatic advances at the frontier will struggle to translate into economy-wide gains in productivity and growth.

Markets are already finding more practical ways to deploy the technology. In the US, Spence has observed companies moving away from relying exclusively on expensive closed models towards what he calls a “hybrid model”: the most advanced models are reserved for the most demanding tasks, while open-weight, smaller and cheaper models are used where they are sufficient to get the job done. DeepSeek, he noted, is being deployed by some US companies and is already widely used in India and parts of the developing world.

This is the other side of the AI race: not how far the frontier can advance, but how quickly AI can spread through the economy.

On this measure, Spence sees something significant in China. He noted that China’s policy planning explicitly emphasizes AI adoption and diffusion, including the sectors in which policymakers want it to take place. By contrast, he believes the US is underinvesting in diffusion and adoption. In some areas, he said, its approach amounts to a “spectator sport”: waiting to see whether the market can make it work.

Technological breakthroughs matter. Economies are transformed not simply when a handful of companies push the frontier forward, but when the technology spreads far enough for businesses across the economy to put it to work.

 

China Is Not the Only One That Needs to Adjust

Spence’s distinction between competition and coexistence extends beyond AI.

China’s manufacturing sector illustrates the shift. Fifteen or 20 years ago, Chinese manufacturing was largely complementary to many of the economies with which it now competes. As China has diversified its industrial base and moved up value chains, it has come into direct competition with countries such as Germany and South Korea.

At the same time, Spence argues, both the level and composition of Chinese domestic demand remain out of balance, tilted away from consumption and towards investment. His preferred response is not a stronger renminbi. China, he argues, could make a greater contribution by strengthening domestic demand in a structurally sustainable way, while more outward direct investment could help ease some of the tensions created by its export strength.

When Chinese companies invest and produce overseas, host economies share in the investment, technology and employment generated by that activity. As China evolves from a major beneficiary of globalization into one of the world’s strongest manufacturing competitors, the next phase of its global integration may therefore involve not just exporting more, but investing more abroad.

Middle powers face their own version of the adjustment. Spence pointed to Canada, whose exceptionally high dependence on the US economy he regards as a poor long-term strategy. Diversifying away from that dependence, while rebuilding relations with China and developing ties with Europe, will take years and may carry costs for growth and employment. But he sees it as a necessary structural adjustment.

The circumstances differ, but the underlying lesson is similar: old patterns of dependence are being reshaped, even as interdependence itself remains.

 

Finding room for Limited Cooperation

What emerges may be a less orderly international system — one built not around a single comprehensive framework, but through countries reaching agreements and forming coalitions issue by issue.

Spence describes the likely result as a “slightly messy but nevertheless workable, practical architecture” for the global economy.

AI may offer an early test. China and the US are unlikely to stop competing for technological leadership. Nor will countries stop trying to make their economies more resilient. But competition does not eliminate the need for cooperation — whether in managing extreme AI risks or adapting to shifting patterns of trade and investment.

The challenge is no longer how to avoid competition. It is how to compete without losing the capacity to cooperate.