Two events, eight days apart, tell the same story from opposite sides of the Pacific. On July 29, Washington’s Federal Communications Commission banned imports of new humanoid and quadruped robots, citing national security. On Aug 6, Unitree Robotics priced its Shanghai listing at a $9 billion valuation, becoming the first humanoid robot maker listed on China’s A-share market. One government moved to keep China’s machines out. Chinese capital markets moved to fund more of them. They look like separate policy and market stories. They aren’t: The center of gravity in the artificial intelligence (AI) race has shifted from the screen to the factory floor, and the United States knows it.

For three years, AI leadership was measured in a single currency: model quality. Whoever trained the smartest system, on the most data, with the most compute, led the field. That contest is not over, but it has stopped being the whole contest. A large language model can be trained in a data center anywhere with enough chips and electricity. A humanoid robot cannot. It needs motors, actuators, batteries, sensors, and myriad precision components, assembled at a cost low enough that someone will actually buy it — and then it needs a supply chain capable of doing that a million times over without the price collapsing the business. Intelligence, in other words, is no longer the bottleneck. Manufacturing it into an affordable machine is.

China enters this contest with an unusual advantage, built up over decades rather than designed for AI — it was never building this capability from scratch. Unitree shipped roughly 5,500 humanoid units last year, more than any competitor worldwide, at gross margins near 60 percent — profitable numbers that remain rare anywhere in the sector. China supplied an estimated 85 percent of the global humanoid robot market last year, according to industry estimates cited in reporting on the ban, and the advantage was never a secret: Two million industrial robots run on Chinese factory floors, more than four times the total in Japan, the world’s next-largest user. Robotics did not need a new industrial base. 

The ban excludes robots already on the market and carves out federal government purchases entirely — a targeted measure aimed less at removing existing Chinese robots from American soil than at slowing how many new ones arrive. Whatever Washington’s intention, the measure marks a shift: Advanced Chinese robotics is no longer being treated as just another commercial product category. It is being treated as a strategic technology

It needed AI models sophisticated enough to run on the industrial base that existed — and that base, unlike a data center, took decades to build and cannot be replicated with a subsidy check. In July, humanoid robots demonstrated picking and sorting at Shanghai’s World Artificial Intelligence Conference — and, away from the conference floor, real trials were already under way: robots sorting parcels at a postal facility in Guangzhou, others handling fabric on garment production lines in Zhejiang province. The tasks remain narrow, and manufacturers openly admit the robots cannot yet run a full production line unsupervised. The work is narrow. But it is real work, not a demonstration.

Beijing has treated this as a plan rather than a coincidence. China’s 15th Five-Year Plan (2026-30) names “embodied intelligence” — AI systems that perceive and act in the physical world — as a strategic future industry, placed in the same tier as quantum computing and nuclear fusion. The plan is explicit about the logic: Robotics is not a side bet on hardware, but the mechanism by which AI research becomes economic output. Unitree’s initial public offering is the first clean, market-priced test of whether that bet is paying off, and investors have already pushed secondary-market expectations well above the listing price. Chinese banks have moved in the same direction: Several of the country’s largest lenders recently set targets of roughly 30 percent growth in loans to innovation-focused firms, credit aimed squarely at the companies trying to turn robotics research into shipped hardware.

Washington’s ban should be read in that context, not as an isolated trade measure. The ban excludes robots already on the market and carves out federal government purchases entirely — a targeted measure aimed less at removing existing Chinese robots from American soil than at slowing how many new ones arrive. Whatever Washington’s intention, the measure marks a shift: Advanced Chinese robotics is no longer being treated as just another commercial product category. It is being treated as a strategic technology.

None of this guarantees China wins the next phase outright. Humanoid robots remain expensive, fragile and, outside controlled settings, not yet reliable enough to justify the hype attached to them by the market; Unitree’s own prospectus flags US tariffs and export controls as real risks to its overseas growth. Investors often price the promise before the technology delivers, and a $9 billion valuation is still somewhat a bet, not a verdict. But the direction of the contest has changed, and pretending otherwise would be wishful thinking — not because one side is destined to lose, but because the terrain itself has moved.

The question worth asking now is not which country trains the smartest AI model.

It is which one can turn that intelligence into a million affordable machines — reliably, and at a cost someone will actually pay. For three years, that question barely mattered. It now sits at the center of the entire AI race — and the answer depends less on who writes the best algorithm than on who has spent 30 years building the factories capable of running it.

The author is a former investment banker turned technology journalist who writes about Asian business, technology, innovation, and capital markets. He previously worked at GE Capital and Korea Development Bank group.

The views do not necessarily reflect those of China Daily.