Chinese AI company Z.AI has completed a massive AI data center that reportedly runs only on Chinese-made chips.

The company, formerly known as Zhipu, has already started partial operations at the site, which is designed to support the training and development of its GLM models. The facility is described as a 1GW-class AI data center, making it one of the largest server hubs built by a Chinese AI lab.

The most important detail is not just the size of the facility. It is the hardware.

According to Bloomberg-linked reports, Z.AI runs several computing clusters, each containing more than 10,000 chips. The site reportedly uses domestic Chinese chips instead of Nvidia hardware.

The exact chip suppliers were not named. China’s AI accelerator market includes Huawei, Cambricon, Moore Threads, Kunlunxin, and other local players. Reuters previously reported that Zhipu’s GLM systems had used domestically made chips for inference, including Huawei Ascend chips and products from Moore Threads, Cambricon, and Kunlunxin.

The data center is a major test of China’s push to reduce dependence on Nvidia.

US export controls have restricted Chinese companies from buying the most advanced Nvidia AI chips. Beijing has responded by encouraging local AI firms to rely more on domestic chips for training and inference.

A large training facility built entirely around local silicon suggests Chinese AI labs are no longer only testing domestic hardware. They are now trying to scale it for frontier model development.

The timing is important.

The report comes shortly after Moonshot AI released Kimi K3, a large open-weight model that added fresh pressure to China’s AI race. Reuters reported that Chinese firms including Moonshot, Z.AI, and MiniMax are releasing increasingly capable models at lower cost, challenging older assumptions that Chinese developers remain far behind US rivals.

That makes compute capacity a bigger issue. If Chinese AI labs want to keep launching larger models, they need reliable access to large-scale training hardware. Z.AI’s new data center appears to be part of that answer.

Z.AI’s project also fits into China’s wider AI infrastructure push.

China is preparing to spend around 2 trillion yuan, or about $295 billion, over five years to build data centers across the country, Bloomberg News reported. Reuters also reported that the plan is aimed at strengthening China’s position in the global AI race.

Cloud giants such as Alibaba and China Telecom remain among the biggest builders, but startups like Z.AI are also racing to control more of their own compute.

Z.AI is not just competing on model performance.

The company is reportedly on track to reach $1 billion in annual recurring revenue, which would make it the first Chinese AI startup to hit that level. Bloomberg reported that the company sells cloud access to its AI models and builds customized AI systems for enterprise customers, including state-owned firms.

This positions Z.AI as one of China’s closest equivalents to an enterprise AI supplier like Anthropic.

The report still comes with important limits.

Z.AI has not publicly confirmed the chip details, and the exact hardware inside the data center remains unknown. Building a large domestic-chip cluster also does not prove it can train frontier models as efficiently as Nvidia-based systems.

Domestic chips may still lag Nvidia in raw performance, software maturity, and cluster stability. Training large AI models also requires more than chip count. Networking, memory bandwidth, software optimization, and reliability all matter.

The real test will be Z.AI’s next GLM model and how it performs against new releases from US and Chinese rivals.

For now, the message is clear: China’s alternative to Nvidia is moving from policy goals and slide decks to working, gigawatt-scale infrastructure.

Get the latest tech news, telecom insights, and product launches wherever you prefer.

Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.

Shares