Alibaba launches custom AI chip to power massive data centre expansion

Alibaba's new Zhenwu V900 accelerator represents a significant advancement in the company's infrastructure strategy, offering substantially improved computational capabilities compared to its earlier generation. The chip can be deployed in clusters of up to 500,000 units, enabling the processing power required for training frontier-level artificial intelligence models.

The announcement underscores Alibaba's commitment to expanding its domestic data centre footprint as demand for AI computing resources accelerates across Asia. By 2032, the company plans to have deployed 20 gigawatts of data centre capacity—a massive undertaking that reflects the growing computational requirements of generative AI applications and enterprise machine learning workloads.

The move positions Alibaba to compete more effectively in the competitive market for AI infrastructure, particularly as global demand for computing power continues to surge. Custom-designed chips allow Chinese technology companies to reduce dependence on international semiconductor suppliers while optimizing performance for specific workloads.

The Zhenwu V900's scalability to such large cluster sizes addresses a critical requirement for training increasingly sophisticated AI models, which demand unprecedented levels of parallel processing capability. Alibaba's vertical integration strategy—designing both hardware and software—enables the company to tailor infrastructure solutions to its specific requirements.