Qualcomm moves into AI data centres with AI200 and AI250 inference chips; first customer Humain to deploy 200MW

Qualcomm on October 26, 2026, unveiled two artificial intelligence chips for data centres, the AI200 and AI250, in its biggest move yet beyond smartphones and into the AI infrastructure market dominated by Nvidia. Shares in the company jumped sharply on Monday after the announcement.

Two chips, due in 2026 and 2027

The AI200 is due to be commercially available in 2026 and the AI250 in 2027. Both are inference-optimised accelerators, designed to run already-trained AI models rather than train them, and are built on Qualcomm's Hexagon neural processing unit (NPU) architecture adapted for data-centre workloads. They will be sold as accelerator cards and as full rack-scale systems, using direct liquid cooling, PCIe interconnects for scale-up and Ethernet for scale-out, at 160kW of rack-level power.

Memory is the selling point

Qualcomm's pitch centres on memory: the AI200 card supports up to 768GB of LPDDR memory, while the AI250 uses a near-memory computing design that Qualcomm says delivers more than 10 times higher effective memory bandwidth, though it did not disclose a baseline for comparison. The company says the systems support common AI frameworks and tools and will lower the total cost of ownership for enterprises running large language and multimodal models.

First customer Humain, 200MW deployment

Saudi AI startup Humain was named as the first customer, planning to deploy 200MW of Qualcomm AI200 and AI250 systems in Saudi Arabia and other locations from 2026, to support its ALLaM models and customer-specific solutions for enterprises and governments. Durga Malladi, Qualcomm's senior vice president and general manager for technology planning, edge solutions and data centre, said the new chips redefine what is possible for rack-scale AI inference at unprecedented total cost of ownership.

From phones to data centres

Qualcomm remains the largest maker of smartphone processors, but the growth of data-centre AI spending has pushed it to carry its mobile power-efficiency expertise into servers. In June it agreed to buy Alphawave, a data-centre semiconductor technology firm, for about $2.4 billion, laying groundwork for this push. Nvidia chips still underpin much of the current AI boom, so whether Qualcomm can take share from Nvidia and AMD will depend on how the AI200 performs once deployed.