On October 27th, Qualcomm officially announced its entry into the field of artificial intelligence (AI) chips in data centers, and launched two brand-new chips, AI200 and AI250, with the intention of opening up its own path in the current market dominated by NVIDIA. These two chips are expected to be listed in 2026 and 2027 respectively, marking an important strategic extension of Qualcomm from the wireless communication and mobile device chip market to the large-scale data center chip field.
 

The newly released chip adopts Hexagon Neural Processing Unit (NPU) technology independently developed by Qualcomm, which focuses on AI reasoning tasks rather than model training. Among them, the AI200 single card will support up to 768GB of low-power LPDDR memory, far exceeding the memory capacity of 288GB HBM3e equipped by NVIDIA's latest GB300 GPU.
 

What is more noteworthy is that the AI250 will introduce an innovative "near storage computing" memory architecture, which is expected to increase the effective memory bandwidth by more than 10 times, while greatly reducing power consumption. Qualcomm also announced the details of its rack-level solution: using direct liquid cooling technology, a single cabinet consumes 160 kilowatts of power, and supports flexible expansion through PCIe and Ethernet.
 

In terms of market layout, Qualcomm has reached a cooperation agreement with Humain, a Saudi artificial intelligence company. Humain will become the first customer of AI200 and AI250, and plans to deploy AI solutions with a total power of 200 MW from 2026. This series of measures clearly shows that Qualcomm is actively expanding strategically from the traditional field of mobile communication chips to the AI computing power market in data centers.