Alibaba Sets 10-Trillion-Parameter Model Goal Alongside Its New AI Chip

Alibaba is tying a planned 5-trillion- to 10-trillion-parameter AI model to a new in-house chip, commercial supernodes and a major data-center expansion. The ambition is large; so is the bottleneck: CEO Eddie Wu said supply shortages are already limiting how quickly Alibaba Cloud can expand.

At its annual Apsara Conference in Hangzhou, Alibaba presented the model, chip and cloud plans as one strategy. The company plans to train a model with 5 trillion to 10 trillion parameters, the learned variables that are a rough measure of a model’s size. Its current flagship, Qwen 3.8 Max, has 2.4 trillion parameters, making the proposed system roughly two to four times larger.

A model target that depends on more than model research

Wu said the Qwen team is working on model architecture and data optimization for more complex, longer-horizon tasks, as part of Alibaba’s stated progress toward artificial superintelligence. But a larger model also raises the practical question of where it will be trained and run. Alibaba says its existing M890 supernode can handle inference, or the process of using a trained model to generate responses, for models above 2 trillion parameters.

The hardware claim reaches from chip to cluster

Alibaba also introduced the Zhenwu V900, a next-generation AI chip from its T-Head semiconductor unit. Alibaba says the V900 delivers three times the performance of its predecessor, the M890, and that a single V900 cluster can support as many as 500,000 cards for frontier-model training and inference. Those are company performance and scale claims, not independently reported benchmarks.

The infrastructure Alibaba says it is assembling

• An M890 supernode that Alibaba says can run inference for models above 2 trillion parameters.
• The Zhenwu V900, which Alibaba says is three times as powerful as the M890 chip it succeeds.
• V900-based clusters that Alibaba says can scale to 500,000 cards for training and inference.

Capacity is the limiting variable

Alibaba Cloud aims to exceed 20 gigawatts of global data-center capacity by 2032 and says it will begin bringing AI supernodes online at commercial scale this quarter. Wu said customer demand for AI remains strong, but global shortages across the AI data-center supply chain are limiting the company’s expansion speed.

The industry’s mid-to-long-term demand far outpaces our supply capabilities.

Eddie Wu, Alibaba CEO

Alibaba expects significant growth in annual AI-chip shipments. It framed the hardware and cloud push within a wider effort by Chinese companies to develop domestic alternatives to Nvidia processors amid U.S. export restrictions. For Alibaba, the strategy’s near-term proof point is concrete: whether commercial supernode deployment begins on the timetable it gave, even as component constraints persist.