Alibaba has introduced its new generation AI accelerator, the Zhenwu V900. Developed by the company’s semiconductor unit T-Head, this new processor offers three times higher performance compared to the previous generation Zhenwu M890.

Alibaba Zhenwu V900 features, notable for its 216 GB memory capacity and massive cluster support scalable up to 500,000 chips. This hardware was specifically developed for both training AI models and inference processes.

Technical Specifications and Scalability

The new accelerator features 216 GB of GPU memory, while the inter-chip bandwidth reaches 1,200 GB/s. The system supports different information sensitivities, including FP8 and FP4.

The previous generation Zhenwu M890 was limited to 144 GB of memory and 800 GB/s inter-chip bandwidth. While superior nodes with 128 chips were used in M890-based systems, it is stated that more than 1,000 accelerators can be run under a single system on the V900 side.

On a larger scale, Alibaba’s new magnificent node infrastructure can reach up to 500,000 V900 cards. This new magnificent node server doesn’t just contain processors.

The system brings together Alibaba’s ICN Switch network solution, Panmai SmartNIC network card, and Zhenyue SSD controller in a one-to-one infrastructure. Computing, storage, and networking components are optimized to work together.

This configuration provides efficiency in the training and inference processes of very large artificial intelligence models. Alibaba describes the Zhenwu V900 as the most powerful AI chip in China.

However, the FLOPS value, which indicates the processor’s absolute computing performance, has not yet been declared. Basic technical details such as manufacturing technology, manufacturer foundry, power consumption, and direct memory bandwidth have also not been shared.

Market Competition and Future Roadmap

One of the domestic options competing with the V900 will be the Huawei Ascend 960PR. Scheduled for release in the third quarter of 2027, the Ascend 960PR boasts 192 GB of memory capacity, 2.4 TB/s of memory bandwidth, and 2.2 TB/s of scalability.

Since all comparable performance data for the Zhenwu V900 has not yet been shared, a direct comparison of the computing power of the two accelerators cannot be made at this stage. Mass production and commercial launch of the V900 will begin in the first quarter of 2027.

Alibaba’s previous roadmap indicated the third quarter of 2027 for the V900. Thus, the timeline for the new accelerator has been brought forward by several months.

As of May 2026, T-Head had shipped more than 560,000 Zhenwu chips to over 400 external customers. The current customer base has exceeded 650 across various sectors including automotive, finance, large language models, robotics AI, power, and manufacturing.

Alibaba’s hardware plan is progressing with the new generation Qwen models. Qwen 4 is currently in the training phase.

The plan is to increase the model scale to 5 to 10 trillion parameters in the upcoming Qwen 4.5 and Qwen 5 families. The current Qwen3.8-Max model has 2.4 trillion parameters.

In automated development work conducted on Qwen3.8-Max, the model completed 33 iterative cycles over a month. The process, which included data validation, experiment execution, error detection, and training optimization, resulted in an increase in the model’s Artificial Analysis score from 40 to 45.

In a separate chip design experiment, the model ran for over 60 hours and made more than 10,000 EDA tool calls. During this process, the designed chip area was reduced by 42% without performance loss.

Alibaba Cloud plans to increase its global data center capacity to over 20 GW by 2032. The company also stated that global supply constraints in AI infrastructure are limiting the current growth rate.

The deployment of commercial-scale Alibaba AI Supernode systems will begin in the third quarter of 2026. The Zhenwu V900 and the new Qwen roadmap were announced at a time when the US continues its export controls on advanced AI chips from China.

While Chinese technology companies are working on domestic alternatives to Nvidia accelerators, Alibaba’s announcement comes just before the meeting between US President Donald Trump and Chinese President Xi Jinping on September 24, 2026. AI was one of the topics discussed in the talks between the two countries, and a new AI safety notification mechanism was also brought up during the preparatory meetings in New York.

Do you think Alibaba’s new generation AI chip can change the competitive landscape in the global market?