Alibaba shipped nearly 500,000 AI chips, but admitted it lags behind Nvidia

Alibaba has revealed for the first time the scale of its own AI chip production: the T-Head division shipped nearly 470,000 devices. At the same time, the company did not embellish the situation and directly acknowledged that in terms of performance, these solutions still lag behind Nvidia's products.

How Many Chips Are Already There

The figure was announced during Alibaba's quarterly report for the third quarter of fiscal year 2026. According to management, T-Head over the past two years has brought its chips to the commercial stage and has shipped approximately 470,000 units in total. These are not laboratory samples, but real hardware already being used for AI workloads.

For the Chinese market, where access to the most powerful American accelerators is limited by U.S. export regulations, such volume in itself looks like an important signal. Even more important is that Alibaba is not just talking about internal use. According to the company, more than 60% of T-Head chips are already going to external customers, and revenue from this direction has reached approximately 10 billion yuan per year. This means the story is no longer a purely corporate R&D project. Alibaba now has its own semiconductor business, which should not just support cloud operations within the group, but function as a commercial element of its entire AI strategy.

Where Alibaba Lags Behind

The most notable aspect of the announcement is the rare directness for large tech companies. Alibaba did not pretend that it had already caught up with global leaders. The company's CEO Eddie Wu acknowledged that T-Head chips still lag behind foreign counterparts. For the Chinese market, where news about domestic hardware is typically presented as a demonstration of technological sovereignty, this is an unusually high degree of candor. And this is an important marker: the company is selling the market not a myth about a breakthrough, but an honest picture of the current level.

"Our chips still lag behind foreign counterparts," acknowledged

Alibaba CEO Eddie Wu.

But Alibaba is not betting on a frontal race in benchmarks. Wu explained that the company wants to design its chips more deeply together with Alibaba Cloud infrastructure and Qwen family models. The idea is to win not on the absolute speed of a single accelerator, but on the ratio of price, availability, and efficiency in specific scenarios. For customers, this can be more important than simply the highest test scores, especially if alternative chips are difficult to buy or take a long time to obtain.

Why Alibaba Needs Its Own Stack

For Alibaba, owning chips is not just a matter of margins, but also a matter of access to computing power. Against the backdrop of American restrictions, Chinese companies cannot count on stable supplies of the most advanced accelerators. Therefore, T-Head within Alibaba is viewed as a way to guarantee itself computing power for years to come. The company specifically emphasizes that it remains the only major cloud provider in China with its own chip development, which means it can more closely control the entire chain—from hardware to model and service.

Against this backdrop, it becomes clearer why Alibaba is investing so aggressively in the AI + Cloud combination. The company's cloud division grew 36% year-over-year in the quarter to $6.19 billion, and revenue from AI products shows three-digit growth for the tenth consecutive quarter. In parallel, Alibaba is setting a major goal: to bring annual external revenue from cloud and AI to $100 billion within five years.

What does such vertical integration give the company:

• more predictable supplies of computing power;
• ability to optimize inference for Qwen models;
• reduction of AI services costs in the cloud;
• sale of not a single chip, but an entire stack for corporate customers.

What This Means

Alibaba has not yet shown a chip that can compete on equal terms with Nvidia's best solutions. But it has shown something else: the Chinese market has already moved from talking about technological sovereignty to mass production of its own AI accelerators and their real monetization. For the industry, this is a signal that the next phase of competition will go not only along the line of "who is faster," but also along the line of "who will assemble a cheaper, more accessible, and more manageable AI stack."