America is inventing AI. China is deploying it.

In dollar terms, the American AI economy is without historical precedent. Last year, private investment in US-based AI companies reached $285.88 billion (roughly twenty-three times China's $12.41 billion), and more than every other country on earth combined. American organizations released fifty top-tier AI models in 2025, against thirty from China. They fund more AI researchers and more AI startups than anywhere else on the planet.

By the metric that has always defined the race, the United States is not losing. It is lapping the field.

*An important caveat: China's government did commit to invest $98 billion last year.

The Lead Is Dissolving

In May 2023, the top American model led the top Chinese model by roughly 300 points on the Arena leaderboard. By March 2026, that gap had narrowed to 39 points — 2.7 percent. Stanford describes the prior "substantial lead" as having been "nearly erased."

The inputs underneath tell a similar story. China now publishes more AI research than any other country and accounts for 20.6 percent of global AI citations, compared with 12.6 percent for the United States. In 2024, Chinese entities received 74 percent of the world's AI patent grants. American patents received 12 percent. The pipeline that feeds future breakthroughs — researchers — is thinning as well: the number of AI scholars moving to the United States has fallen 89 percent since 2017, with the decline accelerating sharply in the past year.

China Installs the Machines

Consider a different measurement. In 2024, China installed 295,000 industrial robots. The United States installed 34,200. China installed more than the next nine countries combined, and its share of global robot installations climbed from 20.8 percent in 2013 to 54.4 percent last year.

The factories exist. The integrations exist. The workforce trained to operate them exists.

Reporting for the Guardian earlier this year, Chang Che visited eleven robotics companies across five Chinese cities. What he found was not a policy ambition but a built-out industry: quadruped robots patrolling power lines, humanoid prototypes rehearsing factory tasks, a regulatory and financial apparatus designed to move embodied AI out of the lab. The technology is being delivered into operating environments at speed.

Consumer adoption follows the same pattern. The New York Times has documented the mainstream Chinese embrace of tools like OpenClaw, an open-source autonomous agent whose rapid spread gave rise to a domestic slang term — "raising lobsters" — for the process of installing and tuning it. The United States, despite building most of the models the rest of the world uses, ranks 24th globally in generative AI adoption at 28.3 percent.

The Shortcut

In a recent memo, the White House accused Chinese entities of running "deliberate, industrial-scale campaigns" to distill American frontier AI systems — querying proprietary models through tens of thousands of proxy accounts, using jailbreak techniques to expose training data, and extracting capabilities at a fraction of the cost of original development. OpenAI and Anthropic have publicly named DeepSeek, Moonshot AI, and MiniMax as suspected actors.

The mechanism resolves the apparent paradox. A twenty-three-to-one investment gap is less daunting if you can query the outputs of the winning investment for pennies per call. The US funds the frontier. Others harvest it.

Invention Is Not Deployment

A model at the top of a leaderboard generates headlines. A robot on a factory floor generates output. Invention without deployment is a technology demonstration. Deployment at scale, even if derivative, is an economy.

The question is not who will build AI. It is who will be using it when the decade closes.

The Takeover Tracker newsletter covers the economic and labor consequences of AI deployment in daily briefings. Subscribe here.