Alibaba Cloud Ecosystem Update Integrates Qwen 3.7 Max Models With Zhenwu M890 Processors and Panjiu AL128 Infrastructure For Autonomous Digital Agent Workflows
Alibaba has rolled out a massive AI ecosystem update. New hardware and software integration extends from custom chip silicon to agent oriented application models, according to official announcements at the Alibaba Cloud Summit in Hangzhou. It signals a new focus on autonomous digital agents capable of executing complex, multistage workflows.
Qwen 3.7 Max Engineered for Autonomous Engineering
The new Qwen 3.7 Max large language model was engineered to work on ongoing, multi stage projects instead of short text generation or conversation interactions. The model continued running for 35 hours without interruption on a new compute platform in an operational longevity benchmark test, calling over 1,000 different tools throughout to write an optimized compute kernel that ran 10x faster than the manufacturer's official code. The model can manage entire, multi file software projects and plug into common agent platform services including:
- OpenClaw
- Hermes Agent
- Claude Code
- Qwen Paw
- Qoder
Panjiu AL128 Server and Cloud Infrastructure
To serve the high concurrent traffic flows from autonomous software agents, Alibaba Cloud designed the Panjiu AL128 Supernode Server. Key technical specifications include:
- 128 hardware accelerators in a single rack unit box
- 1 petabyte per second of internal bandwidth
- Access through the Bailian service portal known as Model Studio
- Utilization of Agentic RL to train models on successful task completion
T Head Zhenwu M890 Processor and ICN Switch
Alibaba's semiconductor subsidiary T Head announced the Zhenwu M890 as its main AI chip. Performance and hardware features include:
- 3x faster performance compared to the Zhenwu 810E
- 144GB of on chip memory
- 800GB per s inter chip communication bandwidth
- Support for numerical precision down to compressed 4bit integer values
Paired with the processor is the new ICN Switch 1.0, which delivers up to 25.6Tb per s total bandwidth for clusters of 64 accelerators. It is all managed with the T Head SAIL stack, which optimizes performance with custom silicon. Over 560,000 chips have been delivered to more than 400 enterprise customers in 20 different industries, including auto makers and financial companies.
During the summit, Alibaba CEO Eddie Wu gave financial outlooks that predict model and application services revenue of 10,000,000,000 Renminbi, or approximately $1.4 billion US dollars, in the next quarter. He predicts this figure to grow to 30,000,000,000 Renminbi, or $4.1 billion US dollars, by the end of the year. Wu anticipates that the revenue of AI related products will exceed the revenue from standard cloud compute sales to become the Cloud Intelligence Group's top revenue driver within a year.
