In a moment that speaks to the deepening fracture in global technological order, Chinese AI firm Z.ai has unveiled GLM-5.3-Flash — a competitive language model it claims runs entirely on domestically produced chips. The announcement, which lifted the company's shares by more than 8%, is less about a single product launch and more about a civilization-scale question: whether nations can build sovereign technological foundations in an era of deliberate interdependence. China's answer, still unverified but symbolically potent, is arriving chip by chip.
Z.ai shares surge 8% on homegrown chip AI model launch
A top-ten model running on entirely Chinese chips
Mark
So Z.ai built this model on entirely Chinese chips—how significant is that, really?
Mimi
It's significant because it proves the concept works at scale. They're not just talking about self-sufficiency; they're actually running a top-ten model on their own hardware.
Mark
But they won't say which chips they used. Doesn't that seem like they're hiding something?
Mimi
Maybe. Or maybe they're protecting their suppliers from U.S. scrutiny. If you name the chip maker, you invite attention to that company.
Mark
The model was already live for a week before they announced it. Why the secrecy?
Mimi
Testing in the real world without the noise. They wanted to see if it could actually handle demand before making claims about it.
Mark
And it ranked first on OpenRouter during that week. That's real usage, not a benchmark.
Mimi
Exactly. That's the part that matters most. A ranking on a test is one thing. Actual people choosing to use your model over others—that's validation.
Mark
What does this mean for Nvidia?
Mimi
It means the threat to their dominance is becoming concrete. This isn't theoretical anymore.
Le Pouls
- Z.ai quietly deployed its new model a full week before announcing it, stress-testing real-world demand under the codename 'Ox Alpha' — and it topped global usage rankings on OpenRouter before anyone knew its name.
- The company claims 100,000 Chinese-made chips powered every request to GLM-5.3-Flash, a direct provocation to the U.S. export restrictions that have cut off Nvidia's most advanced hardware from Chinese buyers.
- Critical details remain in shadow: Z.ai has not named its chip suppliers, disclosed hardware specifications, or allowed independent verification — leaving the technical achievement as much assertion as fact.
- The model landed 10th globally on the Artificial Analysis Intelligence Index, outranking DeepSeek V4 Pro Max, signaling that China's parallel AI ecosystem is maturing faster than many Western observers anticipated.
- With Huawei and others racing to fill the Nvidia-shaped void, Z.ai's launch lands as both a commercial milestone and a geopolitical signal — that semiconductor self-sufficiency is no longer a distant ambition but an active, accelerating project.
In a moment that speaks to the deepening fracture in global technological order, Chinese AI firm Z.ai has unveiled GLM-5.3-Flash — a competitive language model it claims runs entirely on domestically produced chips. The announcement, which lifted the company's shares by more than 8%, is less about a single product launch and more about a civilization-scale question: whether nations can build sovereign technological foundations in an era of deliberate interdependence. China's answer, still unverified but symbolically potent, is arriving chip by chip.
On Wednesday, Chinese AI company Z.ai announced GLM-5.3-Flash, a new language model it claims was built and run entirely on domestically manufactured semiconductors. The company's Hong Kong-listed shares surged more than 8% when markets opened the following day.
The launch had already been underway for a week. Z.ai deployed the model on August 20 under the internal codename 'Ox Alpha,' running it through global channels without public announcement. In that quiet interval, the model ranked first by usage volume on OpenRouter, a platform aggregating access to AI systems worldwide — meaning the company had already proven real-world viability before making any official claim.
Z.ai says 100,000 Chinese-made chips powered all incoming requests to the model. On the Artificial Analysis Intelligence Index, GLM-5.3-Flash ranked 10th globally, placing it ahead of DeepSeek V4 Pro Max. The company positioned it as a lower-cost alternative to its flagship offering. Notably, the chip suppliers were not named, technical specifications were not disclosed, and CNBC could not independently verify the claims. It also bears mentioning that serving a deployed model demands far less computing power than training one — a distinction relevant to assessing the scale of the achievement.
The announcement lands inside a larger contest. U.S. export restrictions have blocked Nvidia from selling its most advanced chips to China, pushing Chinese firms including Huawei to accelerate domestic semiconductor development. Meanwhile, leading American AI models remain officially unavailable inside China, compelling Chinese companies to build their own parallel ecosystem.
GLM-5.3-Flash is a concrete, if still unverified, step in that direction — a public signal that Chinese-made chips can power a model competitive on the world stage. How Z.ai scales, sustains, and improves from here will help determine the shape of global AI infrastructure for years to come.
On Wednesday, the Chinese artificial intelligence company Z.ai announced a new model called GLM-5.3-Flash, built entirely on semiconductors manufactured within China. The announcement sent the company's Hong Kong-listed shares up more than 8% when trading resumed the following day.
The timing of the release carried its own story. Z.ai had actually deployed the model a week earlier, on August 20, under the internal code name "Ox Alpha," running it quietly through global channels before making the official announcement. During that week of stealth operation, the model ranked first by usage volume on OpenRouter, a platform that aggregates access to various AI systems worldwide. When the company finally went public with the launch, it had already proven the model could handle real-world demand.
According to Z.ai's claims, the company used 100,000 domestically produced chips to power all the online requests flowing to GLM-5.3-Flash. On the Artificial Analysis Intelligence Index, a ranking system that measures model performance, the new release landed at number 10 globally—a position that placed it ahead of DeepSeek V4 Pro Max, another significant player in the competitive landscape of large language models. The company positioned GLM-5.3-Flash as a low-cost alternative to its flagship offering, designed to serve users who need capable AI without premium pricing.
The specifics of which Chinese semiconductor manufacturers supplied the chips remain undisclosed. Z.ai declined to name its suppliers or provide technical details about the hardware. CNBC was unable to independently verify the company's assertions about the chips or their performance characteristics. It is worth noting that running an AI model to serve user requests requires substantially less computing power than the process of training a model from scratch—a distinction that matters when evaluating the technical achievement.
The announcement arrives against a backdrop of intensifying technological competition between the United States and China. Nvidia, the American semiconductor giant that has dominated the market for AI computing, has faced severe restrictions on selling advanced chips to China, imposed through coordinated policy between Washington and Beijing. In response, Chinese companies including Huawei have accelerated their own semiconductor development programs, racing to build viable alternatives that can reduce dependence on foreign technology.
China's broader strategy centers on achieving technological self-sufficiency, particularly in semiconductors and artificial intelligence. The country has invested heavily in domestic chip manufacturing and AI research as a direct response to U.S. export controls. Meanwhile, leading American AI models remain unavailable for official use within China, creating a parallel ecosystem where Chinese companies must develop their own solutions to serve their domestic market and compete internationally.
Z.ai's move with GLM-5.3-Flash represents a concrete step in that direction—a public demonstration that Chinese-made chips can power a competitive AI model that ranks among the world's best. Whether the company can sustain this advantage, scale production, and continue improving its models will shape how the global AI infrastructure landscape evolves in the years ahead.
Citations marquantes
Z.ai used 100,000 China-made chips to handle all online requests to use GLM-5.3-Flash— Z.ai company statement
Envie de l'histoire complète ?
Lire l'original sur CNBC ↗
Technology