He Xiaopeng cut open the leg of his robot in front of a live audience, even though he didn’t have to. The humanoid moved with such eerie grace when XPeng unveiled IRON in Guangzhou last November—a tiny hip sway, a flicker of weight shifting from one foot to another—that online viewers accused him of stuffing a person inside a costume. He thus unveiled it on stage. servos, joints, and wires. The audience chuckled. He appeared relieved. This brief, somewhat dramatic moment seems to encapsulate the current state of China’s AI industry, which is defensive about being questioned, eager to prove itself, and subtly more developed than most outsiders realized.
Beijing was shaken three years ago. Chinese policymakers were caught off guard when ChatGPT crashed through a window. Then, in January of last year, DeepSeek released the R1, which was purportedly constructed on a small portion of the chips that American labs were using. Although the claim is still disputed in Silicon Valley back channels, the market responded right away. Investors changed course after becoming alarmed. In a matter of months, six startups—DeepSeek, Moonshot, MiniMax, Zhipu, StepFun, and 01.AI—were named the “AI Tigers,” each supported by a province or city in dire need of a local champion. When asked about the origin of the nickname, Yan Junjie, CEO of MiniMax, dismissed it with a shrug. He feels that there is no competition at all between the tigers. It’s against the world’s superpowers.
| Field | Details |
|---|---|
| Topic | U.S.–China Artificial Intelligence Competition |
| Estimated Size of China’s Core AI Industry (2025) | $160–$170 billion |
| Number of Chinese AI Enterprises | More than 5,300 |
| China’s Generative AI Patent Lead | Roughly 6× more registered patents than the U.S. |
| Key Chinese AI Players | DeepSeek, Alibaba (Qwen), Moonshot, MiniMax, Zhipu, StepFun, 01.AI, Baichuan |
| The “AI Tigers” | Six rising startups dubbed by investors and the media |
| Alibaba’s Three-Year AI Investment Plan | $53 billion |
| U.S. Hyperscaler Spend (2026, projected) | $650 billion combined |
| China’s Strategic Target Year | 2030, per Xi Jinping’s leadership push |
| Key Policy Frameworks | 2017 New Generation AI Plan; “AI Plus” Initiative |
| Notable U.S. Adopters of Chinese Models | Airbnb, Social Capital, Cursor, Cognition AI |
| Main U.S. Constraint on China | Export controls on advanced Nvidia GPUs |
That framing is important. Because Beijing is in a completely different race than Washington, which continues to refer to a “Manhattan Project” for artificial general intelligence. China’s policymakers appear to be more concerned with integrating AI as quickly as possible into government services, factories, hospitals, and classrooms than they are with AGI as a philosophical goal. Diffusion, not discovery, is the foundation of this strategy. It remains to be seen if that proves to be more prudent or just less expensive.
It’s worth stopping to look at the numbers. Over three years, Alibaba intends to invest about $53 billion in AI. In 2025, Microsoft alone spent about $80 billion. Almost $650 billion is being budgeted for this year by the four American hyperscalers: Alphabet, Amazon, Meta, and Microsoft. That is a chasm by all reasonable standards.
Nevertheless, Chinese models continue to appear in unexpected places. Brian Chesky of Airbnb acknowledged that he moved to Alibaba’s Qwen because it was less expensive and faster than ChatGPT; similarly, Chamath Palihapitiya moved Social Capital to Moonshot’s Kimi K2. Composer and Windsurf, two well-known American coding assistants, seem to have been covertly developed on Chinese foundations, despite the developers’ silence. It has stuck with me since researcher Nathan Lambert referred to it as the tip of the iceberg.
It’s difficult to ignore the irony. Export restrictions were intended to cut off Nvidia’s top chips, suffocating Chinese AI labs. Rather, the limitation might have compelled Chinese engineers to become truly proficient in efficiency, maximizing where Americans merely scale. As you pass the Beijing Innovation Center for Humanoid Robotics’ production line, you can see rows of partially completed machines with ribbed metal torsos that are awaiting limbs. The scene has an almost industrial revolution feel to it. Not glitzy. Just unrelenting.
By 2030, will China surpass Silicon Valley? Most likely not in terms of raw frontier capability; American labs continue to lead benchmarks by months or even more. However, Beijing may be competing in a completely different race. less expensive deployment. quicker uptake. AI is integrated into everyday business, models in the clinic, and robots on the floor. In a few years, the answer will be different if that is the definition of victory. And it already seems to be changing to anyone who is paying close attention.
