- Joined
- Jan 20, 2026
- Messages
- 345
- Reaction score
- 2,471
Alibaba showed the processor Zhenwu M890 and the plan three years ahead.
Alibaba is preparing iron for AI agents, which are not enough to respond to requests. The Chinese company introduced the new processor Zhenwu M890 and showed how it intends to reduce dependence on American chips against the background of export restrictions of the United States.
The new processor was presented at the annual cloud summit of Alibaba in China. The development was carried out by the chip division of T-Head. According to the company, the Zhenwu M890 is about three times faster than the previous processor Zhenwu 810E.
Alibaba calls the Zhenwu M890 a specialized chip for autonomous software agents. By agents, the company understands systems that are able to perform complex multi-step tasks almost without human participation. Such work requires a lot of memory, constant data exchange between models and the ability to hold the context for a long time.
The launch of the processor coincided with the accelerated race of Chinese technology companies for their own replacements with Nvidia chips. U.S. restrictions on the supply of advanced semiconductors to China are forcing the local market to develop domestic developments faster.
The Zhenwu M890 is different from processors that primarily optimize large language models for training. The new chip is designed for the continuous operation of agents: coordination of actions, processing of large data streams in real time and long tasks without constant restart.
Alibaba also uncovered a semiconductor roadmap until 2028. The next V900 chip is scheduled to be released in 2027. The manufacturer promises another threefold increase in performance compared to the M890. The third J900 processor is scheduled for 2028.
The company does not show a single chip, but an attempt to assemble its own ecosystem of computing for the cloud and corporate AI. The market is gradually shifting from chatbots that respond to individual requests, to agent systems capable of clock-executing related operations. Such a load requires more memory, a quick connection between accelerators and sustainable operation without pauses.
Last year, Alibaba promised to invest more than 380 billion yuan, about $53 billion, in cloud and computing infrastructure over three years. The company expects a sharp increase in demand for agent systems, as the business is increasingly introducing autonomous software tools.
Together with the new chip, Alibaba introduced the Panjiu AL128 server system. The platform combines 128 accelerators in one rack and is designed for corporate implementations of a large scale. According to the company, the system is already available to Chinese customers through the Bailian cloud platform.
T-Head reported that it has already delivered more than 560 000 processors Zhenwu. Chips are used by more than 400 external customers from 20 industries, including the financial sector and auto production.
Alibaba also showed the Qwen 3.7-Max, a new version of the flagship big language model. The company says that the model is designed for complex programming and long agent tasks, and can also work up to 35 hours without reducing productivity.
Announce shows how major Chinese tech companies are rebuilding the computing strategy around their own processors. Huawei and other manufacturers have also accelerated the development of domestic hardware after tightening American control over access to advanced chips.
Alibaba is preparing iron for AI agents, which are not enough to respond to requests. The Chinese company introduced the new processor Zhenwu M890 and showed how it intends to reduce dependence on American chips against the background of export restrictions of the United States.
The new processor was presented at the annual cloud summit of Alibaba in China. The development was carried out by the chip division of T-Head. According to the company, the Zhenwu M890 is about three times faster than the previous processor Zhenwu 810E.
Alibaba calls the Zhenwu M890 a specialized chip for autonomous software agents. By agents, the company understands systems that are able to perform complex multi-step tasks almost without human participation. Such work requires a lot of memory, constant data exchange between models and the ability to hold the context for a long time.
The launch of the processor coincided with the accelerated race of Chinese technology companies for their own replacements with Nvidia chips. U.S. restrictions on the supply of advanced semiconductors to China are forcing the local market to develop domestic developments faster.
The Zhenwu M890 is different from processors that primarily optimize large language models for training. The new chip is designed for the continuous operation of agents: coordination of actions, processing of large data streams in real time and long tasks without constant restart.
Alibaba also uncovered a semiconductor roadmap until 2028. The next V900 chip is scheduled to be released in 2027. The manufacturer promises another threefold increase in performance compared to the M890. The third J900 processor is scheduled for 2028.
The company does not show a single chip, but an attempt to assemble its own ecosystem of computing for the cloud and corporate AI. The market is gradually shifting from chatbots that respond to individual requests, to agent systems capable of clock-executing related operations. Such a load requires more memory, a quick connection between accelerators and sustainable operation without pauses.
Last year, Alibaba promised to invest more than 380 billion yuan, about $53 billion, in cloud and computing infrastructure over three years. The company expects a sharp increase in demand for agent systems, as the business is increasingly introducing autonomous software tools.
Together with the new chip, Alibaba introduced the Panjiu AL128 server system. The platform combines 128 accelerators in one rack and is designed for corporate implementations of a large scale. According to the company, the system is already available to Chinese customers through the Bailian cloud platform.
T-Head reported that it has already delivered more than 560 000 processors Zhenwu. Chips are used by more than 400 external customers from 20 industries, including the financial sector and auto production.
Alibaba also showed the Qwen 3.7-Max, a new version of the flagship big language model. The company says that the model is designed for complex programming and long agent tasks, and can also work up to 35 hours without reducing productivity.
Announce shows how major Chinese tech companies are rebuilding the computing strategy around their own processors. Huawei and other manufacturers have also accelerated the development of domestic hardware after tightening American control over access to advanced chips.
