A year on, one question hangs over China’s rapidly developing AI models: what do they actually work on? IZ.AI recently provided part of the answer. It has built a large data center that uses only chips made in China.
The company, formerly known as Zipu, has begun operations in the area, Bloomberg reported, citing a person familiar with the matter. It is a 1 gigawatt hub built to train GLM models for Z.AI. IZ.AI did not respond to a request for comment.
What was built
Scale topic. A gigawatt roughly powers 750,000 homes at any given time. That puts the site among the largest server hubs Chinese AI labs have built.
The chip count is equally impressive. AI now uses several computer clusters, each with more than 10,000 chips, the person said. Most importantly, none of them belong to Nvidia.
Bloomberg did not specifically mention the chips. China’s leading designer of AI accelerators is Huawei, which is competing with local companies such as Cambricon and Alibaba to close the gap on Nvidia.
TNW City Coworking Space – Where your best work happens
A workplace designed for growth, collaboration, and endless networking opportunities at the heart of technology.
Why is it important
US export controls have blocked Chinese labs from buying powerful Nvidia chips. An open question was whether domestic units could take on the responsibility of training border models, not just running them.
A 1GW cluster built entirely on domestic silicon is the real answer to that question. It suggests that Chinese labs can continue to measure even if they are disconnected from the chips that the rest of the industry treats as default.
Time sharpens the point. The build came days after Beijing’s Moonshot Kimi K3 matched the top US models, then ran out of computers and halted new registrations. IZ.AI runs similar competitors, and bets on owning the hardware.
A large building
AI does not do this alone. China is preparing to spend an estimated 2 trillion yuan, about $295bn, over five years on data centers across the country. Cloud giants Alibaba and China Telecom are still the biggest builders so far.
This company also has a competitive edge. Fresh from a Hong Kong listing and subsequent IPO, Z.AI is on track to achieve $1bn in annual recurring revenue, making it the first Chinese AI company to reach that mark. Its shares jumped nearly 20% on the day.
It positions itself as an enterprise AI supplier, a role that begs to be compared to Anthropic.
Catching up
A few caveats should be observed. The details of the chip come from an unknown source, and Z.AI has not confirmed it. Building a cluster is not the same as proving that it can train a frontier model as efficiently as Nvidia.
Home chips can still lag behind in raw performance, and stitching thousands of them into a stable system is difficult. The real test is the following GLM model, and how it compares to what US labs are sending.
However, the direction is clear. The effort to build a Chinese alternative to Nvidia has moved from slideware to an active, gigawatt-scale data center. That alone changes the debate about how far export controls can go.
