In brief

At its Apsara conference on September 22, Alibaba showed the Zhenwu V900, a chip for training and running AI. It says the chip has three times the performance of May’s M890 and will be mass-produced from early 2027. It also plans Qwen models with 5 to 10 trillion parameters.

New to this? Read it in simple words
  • On September 22, Alibaba showed the Zhenwu V900, a chip for training and running AI. Its chief executive called it the most powerful AI chip in China.
  • Alibaba says it has three times the performance of the Zhenwu M890, released in May. Mass production is planned for early 2027.
  • China’s AI chips are improving, but claims like this need tests.
  • Alibaba did not say what it measured for the three times claim. It did not name the chip’s factory, and there are no independent tests.
Words to know
Training
Teaching an AI model by showing it huge amounts of data.
Mass production
Making a product in very large numbers in factories.

A chip with big claims and few details

The Zhenwu V900 comes from T-Head, Alibaba’s chip design unit. It is built both to train AI models and to run them. Chief executive Eddie Wu called it “the most powerful AI chip in China today”.

Alibaba says the chip has three times the performance of the Zhenwu M890, released in May. It lists 216 GB of memory and 1,200 GB per second of bandwidth between chips. Mass production is planned for the first quarter of 2027.

Alibaba did not say what it measured for the “three times” claim. It did not name the factory that makes the chip or give its computing speed. There are no independent tests.

In May, the research firm TrendForce reported a later date for the same chip, the third quarter of 2027. Chip plans can change.

Sources135

CHIP CLAIMS 06
Alibaba’s chip claims are big, and the public details are few.

The factory, computing speed, and outside test results are not public. Qwen 4.5 and Qwen 5 could reach 5 to 10 trillion parameters.

Much bigger Qwen models are planned

Alibaba says Qwen 4 is now in training. It expects its later Qwen 4.5 and Qwen 5 series to reach 5 trillion to 10 trillion parameters. Parameters are the internal values that a model learns during training.

Its current top model, Qwen3.8-Max, has 2.4 trillion parameters in total, of which 95 billion are active at a time. So the planned models would be about two to four times larger. Alibaba gave no release dates.

Sources124

Twenty gigawatts by 2032

Wu set a target of more than 20 gigawatts of data-centre capacity for Alibaba Cloud by 2032. Alibaba did not say how much capacity it has today.

It announced no new investment amount. In February 2025, Alibaba promised at least 380 billion yuan, about $53 billion, for AI and cloud over three years. Caixin reports that it had spent about half of that by June 2026.

Investors reacted with some caution. Reuters reported a jump of about 5 percent in Hong Kong during the day, but the shares closed 1.95 percent higher, according to Caixin.

US export rules still keep the most advanced chips and chipmaking machines out of China. An analyst told AP that closing the gap with the US also depends on China’s chip factories, not only on chip design.

Sources12367

Sources

Every fact in this story comes from the sources below. Open them to check our work.

  1. 1
  2. 2
    Research · September 22, 2026Alibaba deepens AI push with new chip, bigger model; shares jump 5% Reuters via Investing.com
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
How we checked this story

We read Alibaba’s press release and its August model announcement, then compared them with Reuters, AP, Caixin, the South China Morning Post, and TrendForce. The “most powerful” and “three times” claims come from Alibaba, and we found no independent tests.