Z.AI's new facility marks a turning point in China's push to build AI infrastructure without Nvidia silicon
A Beijing-based AI company has flipped the switch on a data center that runs entirely on domestically produced semiconductors. Z.AI’s 1-gigawatt facility is now operational, purpose-built to train the company’s GLM (General Language Model) AI platforms without a single Nvidia chip in sight.
Since 2025, US export controls have forced Chinese companies to get creative about sourcing AI compute. Washington’s restrictions specifically targeted high-end Nvidia GPUs like the H100 and its successors, chips that had become the backbone of AI training worldwide.
In April 2026, Alibaba launched its own data center equipped with 10,000 Zhenwu AI chips, its proprietary silicon designed specifically for AI workloads. Z.AI’s facility dwarfs it in scale and ambition.
Beijing is putting serious money behind this #
The Chinese government has proposed a five-year plan with an estimated budget of approximately $295 billion to create a network of interconnected AI data centers. The target: 80% domestic technology sourcing across the entire network.
New policies now mandate the use of domestic chips in state-funded projects. This creates a guaranteed demand floor for Chinese chip manufacturers, essentially a captive market that lets them scale production, iterate on designs, and drive down costs without having to compete head-to-head with Nvidia for commercial customers.
What this means for investors #
For Nvidia, Z.AI’s data center represents proof that China can build AI infrastructure at massive scale without Nvidia’s products. Every gigawatt of Chinese compute capacity that runs on domestic silicon is a gigawatt that will never generate Nvidia revenue, even if export controls are eventually relaxed.
For investors in the AI infrastructure space more broadly, the key metric to watch isn’t whether Chinese chips match Nvidia’s benchmarks on paper. It’s whether Chinese AI models trained on domestic hardware can compete with models trained on Nvidia silicon. DeepSeek’s earlier breakthroughs already demonstrated that clever engineering can partially compensate for hardware limitations.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our