Chinese AI Chips Gain Momentum as Z.ai Claims New Model Runs Without Nvidia

What Z.ai Actually Announced

GLM-5.3-Flash is a low-cost variant of Z.ai’s flagship model. The company says it deployed 100,000 domestically produced chips to handle all online inference requests for the model, including during its initial release under the code name “Ox Alpha” on August 20.

The model ranked 10th on the Artificial Analysis Intelligence Index — ahead of DeepSeek V4 Pro Max — and topped usage charts on the global OpenRouter platform in its first week. Those are not small numbers for a model positioned as a budget-tier option.

Z.ai’s Hong Kong-listed shares rose more than 8% following the announcement.

The Chip Question Everyone Is Asking

Z.ai declined to name which chip suppliers it’s using. CNBC could not independently verify the claims.

That said, analysts aren’t guessing blindly. Counterpoint Research’s Ivan Lam suggests Z.ai is likely running on Huawei Ascend chips alongside other domestic suppliers. He noted a broader pattern: Chinese AI developers are increasingly investing in infrastructure built entirely on homegrown hardware, with tighter collaboration across the hardware and software stack.

This matters because running inference — serving a live model to users — is less compute-intensive than training. It’s a more achievable milestone for domestic chips, but it’s still a milestone.

Why This Is Happening Now

U.S. export restrictions have cut off Chinese companies from Nvidia’s most advanced chips. That’s not new. What’s changed is the response.

Huawei has accelerated its Ascend chip program. Domestic AI server manufacturers have scaled up. And Chinese AI labs have started building their software stacks to work with what’s available locally rather than waiting for restrictions to lift.

Z.ai’s announcement is the latest visible signal that this infrastructure is maturing — at least enough to handle production inference workloads at scale.

The Competitive Picture in China

Z.ai isn’t the only one moving fast. Rival MiniMax reported a 283% revenue surge in the first half of the year, though its adjusted net loss more than doubled to $293 million. MiniMax’s flagship M3 model ranks 18th on the same intelligence index.

Both companies listed in Hong Kong in January. Z.ai shares have climbed over 800% since the IPO. MiniMax is up around 80%. The gap reflects market confidence in Z.ai’s technical positioning — and announcements like this one reinforce that narrative.

What This Means for the AI Tools Ecosystem

For anyone tracking AI tools and infrastructure, a few things are worth watching:

• Domestic chip viability is being tested in production. If Z.ai’s claims hold up, it signals that Chinese AI products can operate independently of Western chip supply chains at the inference layer.
• Model availability outside China remains limited. Leading U.S. models aren’t officially available in China, and Chinese models face their own distribution constraints globally. The OpenRouter ranking suggests GLM-5.3-Flash is gaining traction internationally despite that.
• Cost-efficient models are where the action is. GLM-5.3-Flash is explicitly a low-cost variant. The race isn’t just about the most powerful model — it’s about the most deployable one.

The practical takeaway: China’s domestic AI stack is no longer purely theoretical. Whether it can scale to training workloads — not just inference — is the next question worth watching.