Every AI conversation in America is about language models. Who has the smartest chatbot. Who scores highest on benchmarks. Who can write better code. Meanwhile, China has been quietly winning a different race entirely: the race to give AI a body.

A new TIME analysis published this week argues that China is pulling ahead in "physical AI," the integration of artificial intelligence with robotics, manufacturing, and the physical world. While American companies debate safety frameworks and prepare IPO filings, Chinese firms are deploying humanoid robots in factories, testing autonomous systems at scale, and building the infrastructure that will define the next decade of AI.

Software Wins Headlines. Hardware Wins Wars.

The split is striking. America's AI giants, OpenAI, Anthropic, Google, are locked in a language model arms race that consumes hundreds of billions in compute. China's AI ecosystem is focused on something different: robots that work, drones that deliver, factories that run themselves.

Google DeepMind's recent partnership with Agile Robots for industrial humanoid testing is the exception that proves the rule. It took one of America's most well-funded AI labs partnering with a German-Chinese robotics company to even enter the game. China has dozens of companies already there.

The Uncomfortable Question

If you had to choose between the country that builds the best AI chatbot and the country that builds the best AI-powered manufacturing base, which one would you bet on over a 20-year horizon? Language models are impressive. They write poems, debug code, and pass bar exams. But they don't make anything. They don't move atoms. They don't build.

Physical AI does. And the country that masters it controls the supply chains, the manufacturing capacity, and the military robotics that will define the 2030s. China seems to understand this. America is still arguing about whether Claude or GPT writes better emails.

This isn't to say language models don't matter. They clearly do. But the narrative that America is "winning" the AI race because it has the best chatbots ignores the possibility that chatbots aren't the race that matters. The country that gives AI hands and legs and puts it in factories at scale may end up with a far more durable advantage than the country that gives AI the best vocabulary.

Brett Adcock, the founder of Figure AI, seems to get this. He just left his humanoid robotics company to launch Hark, a new AI hardware venture with $100 million in stealth funding and a former Apple designer leading product. The smart money is starting to move toward physical AI. The question is whether it's moving fast enough.