Alibaba is taking a broader approach to AI by tying together the model layer, server infrastructure, and chip design into one connected stack. The move is aimed at the growing demand for AI systems that are not only capable, but also stable, efficient, and able to operate with greater autonomy.
At the center of that push is Qwen3.7-Max, a new model positioned as a general-purpose foundation for a range of AI needs. Alibaba says it is built for advanced agentic coding and office workflow automation, with the ability to carry out long-running tasks independently for up to 35 hours and handle more than 1,000 tool calls without performance degradation.
A model built for agentic workloads
That level of endurance matters as AI agents are asked to manage longer chains of work. In industrial settings, the ability to keep pace across many sequential steps is becoming a key requirement, especially when consistency is needed over extended sessions.
Qwen3.7-Max is designed to fit that use case, giving Alibaba a model layer that can support more demanding agent workflows. The company is clearly targeting scenarios where AI must do more than respond quickly and instead maintain reliable execution across complex tasks.
Infrastructure built to keep pace with heavier compute loads
To support those workloads, Alibaba Cloud introduced the Panjiu AL128 Supernode Server. The system combines 128 AI accelerators in a single physical rack and delivers petabyte-per-second-scale data transfer capacity.
That kind of infrastructure is meant for large AI models that need to be trained and run with stability. It also addresses the needs of companies processing very large volumes of data, where speed and coordination across systems become critical.
Alibaba positions the server as a core part of the foundation for AI Agent growth. For agentic systems, the compute environment must remain fast and stable while multiple processes continue to work together smoothly.
T-Head pushes the hardware side further
On the chip side, Alibaba’s subsidiary T-Head added another layer to the strategy with the Zhenwu M890 AI processor. The company claims the chip delivers three times the performance of the previous generation.
The processor comes with 144 GB of GPU memory and 800 GB per second of inter-chip bandwidth. Those specifications are aimed at workloads that require large context retention, high speed, and better cost efficiency.
Alibaba also says the chip supports multiple precision data formats natively. That matters for multi-agent coordination, where systems need to process information quickly while keeping resource use under control.
Already in the hands of customers
T-Head said it has shipped more than 560,000 units of Zhenwu chips to more than 400 customers. Those customers span sectors including automotive and financial services.
The scale of deployment suggests the chip program has moved well beyond a product announcement stage. It is already being used in industries that depend on high-performance computing, which strengthens Alibaba’s position as a digital infrastructure provider serving AI transformation across sectors.
A full-stack push aimed at global developers
Taken together, the announcements show Alibaba’s intent to connect AI models, servers, and chips inside a single ecosystem. That full-stack approach reduces reliance on just one part of the technology chain and creates a more complete base for modern AI development.
Qwen3.7-Max will soon be available to global developers through Model Studio. Panjiu AL128 will be offered for the China market to help meet rising demand from AI workloads across industries.
Source: www.medcom.id
