Key Points:
- Alibaba announced a comprehensive upgrade to its artificial intelligence stack to help developers build autonomous agents.
- The new Qwen3.7-Max software model can run complex tasks entirely on its own for up to 35 hours and manage over 1,000 tool calls.
- The Panjiu AL128 Supernode Server packs 128 powerful artificial intelligence accelerators into one single hardware rack.
- T-Head successfully delivered over 560,000 Zhenwu processor units to more than 400 external customers across 20 different industries.
Alibaba announced a massive upgrade to its artificial intelligence technology on Wednesday. The company revealed new cloud infrastructure, brand-new model services, custom computer chips, and a highly powerful foundation model. Executives shared these updates during the annual Alibaba Cloud Summit. The technology giant wants to give developers and business leaders the exact tools they need to build and deploy independent artificial intelligence agents.
The star of the presentation was Qwen3.7-Max. Alibaba designed this brand-new large language model specifically for agentic coding and complex reasoning. Unlike older software that needs constant human supervision, this model handles long-horizon tasks from start to finish. The company plans to release the model to developers and corporate enterprises worldwide very soon.
Qwen3.7-Max brings incredible stamina to the digital workplace. The software can generate code, find hidden bugs, and fully automate tedious office workflows. It handles massive multi-step projects that require hundreds or even thousands of specific actions. Amazingly, the model can execute tasks totally on its own for up to 35 hours without dropping its performance. Imagine a digital worker who starts a massive coding project on Friday evening and finishes it by Sunday morning without ever taking a break. During that long shift, it can manage more than 1,000 different tool calls flawlessly.
To make the software work seamlessly with existing systems, Alibaba optimized Qwen3.7-Max for popular agent frameworks. Developers can easily plug the new model into OpenClaw, Hermes Agent, Claude Code, Qwen Paw, and Qoder. Users will access all these powerful features directly through the Alibaba Model Studio platform.
Powerful software needs massive physical hardware to run smoothly. To solve this computing problem, Alibaba Cloud launched the Panjiu AL128 Supernode Server. Engineers built this massive machine specifically to support complex inference for artificial intelligence agents and large-scale model training sessions.
The Panjiu AL128 packs an incredible amount of power into a very small physical space. The system integrates exactly 128 artificial intelligence accelerators inside a single hardware rack. This tight design delivers data speeds at the petabyte-per-second scale, allowing massive amounts of information to flow without any delay. Right now, customers in the Chinese market can access the Panjiu AL128 directly through the Model Studio platform.
Alibaba also designs its own silicon chips to reduce its reliance on outside hardware vendors. T-Head, the semiconductor design subsidiary of Alibaba, introduced its newest creation on Wednesday. The team proudly revealed the Zhenwu M890, a highly advanced artificial intelligence training and inference processor.
The Zhenwu M890 delivers exactly 3 times the computing performance of its older sibling, the Zhenwu 810E. When artificial intelligence models grow larger, they require massive amounts of memory to store their parameters. The new chip holds 144 GB of dedicated GPU memory to ensure it can hold these giant models without crashing. The chip also provides 800 GB/s of inter-chip bandwidth. The processor supports multiple data precision formats, ranging from FP32 down to FP4, giving software developers extreme flexibility.
Moving data quickly between chips matters just as much as raw processing power. For this exact reason, T-Head also unveiled the brand-new ICN Switch 1.0. This specialized switching chip delivers up to 25.6 Tbps of total aggregate bandwidth. The switch allows engineers to create a full-bandwidth interconnection among 64 accelerators simultaneously.
These hardware projects already show massive real-world success for the technology giant. T-Head proudly announced that it has delivered over 560,000 total Zhenwu units to the market so far. More than 400 external customers currently buy and deploy these chips to run their daily business operations. These loyal buyers span 20 global industries, proving that the Alibaba silicon strategy actually works at a commercial scale.
By building everything from the base silicon chips up to the final software model, Alibaba controls its entire technological destiny. The company no longer has to wait for outside suppliers to innovate. Companies worldwide currently race to build the smartest computer systems, and Alibaba wants to offer a complete end-to-end package. With the release of the Qwen3.7-Max model and the new Zhenwu hardware, Alibaba sends a clear message to the global technology market that it plans to dominate the future of autonomous artificial intelligence.
