Qualcomm has introduced new AI chips designed for data centres, giving direct competition to Nvidia and AMD.
The tech giant's shares surged more than 20% after the announcement, showcasing the AI market's keen interest in the product.
About the new Qualcomm AI chip:
Qualcomm has released its AI200 and AI250 artificial intelligence accelerators, with the former being available in 2026 and the latter in 2027, while a third chip is planned for 2028.
Similar to GPU racks offered by Nvidia and AMD, both models are designed as fully liquid-cooled rack systems, capable of functioning as a single high-performance computing unit.
The cutting-edge technology behind Qualcomm's new chips:
The new AI200 and AI250 chips are built on Qualcomm's custom Hexagon Neural Processing Unit (NPU), already used in its Windows PC and smartphone processors.Ā
With these chips, Qualcomm is scaling that technology to compete in the field of AI-driven data centres.
Comparison to Nvidia and AMD's chips:
Memory is one of the main differentiating factors between Qualcomm's new models, as the AI250 will reportedly offer 10 times the memory bandwidth of the AI200.
In addition, Qualcomm claims its AI accelerator cards support up to 768 GB of memory, exceeding the current offerings from Nvidia and AMD.
In terms of power consumption, each Qualcomm rack will require around 160 kilowatts, and while its comparable to Nvidia's racks, the tech firm has claimed improved efficiency and cost.
Qualcomm is also aiming to offer flexibility to its customers, allowing them to purchase individual chips, system components, or a complete rack.Ā
