Qualcomm recently launched its AI200 and AI250 artificial intelligence chips, designed specifically for data centers. This development sparked a strong response in the capital market, with Qualcomm's stock price surging by 20% in a single day. This breakthrough marks Qualcomm's official entry into the AI infrastructure market, directly challenging Nvidia's monopoly in the data center GPU market.
The AI200 and AI250 chips utilize Qualcomm's Neural Processing Unit (NPU) technology, each featuring 768GB of LPDDR memory, support for direct liquid cooling, and 160kW of rack-scale power consumption. They are optimized for large language model (LLM) inference. Saudi AI startup Humain has become its first customer, with plans to deploy 200 megawatts of computing power based on these chips by 2026. Qualcomm executives have described them as "redefining what's possible for rack-scale AI inference," highlighting their low total cost of ownership (TCO) and energy efficiency.
The market generally believes Qualcomm's move is aimed at breaking Nvidia's monopoly in data center AI chips. Nvidia currently holds a 98% market share in this field, with data center GPU revenue expected to reach $36.2 billion in 2023. Qualcomm, leveraging its low-power expertise in mobile chips, is attempting to enter the edge AI server market through a "customized CPU + NPU" collaborative architecture. Its strategic goal is to differentiate itself from Nvidia through high-performance, low-power AI infrastructure.
Analysts point out that the surge in Qualcomm's stock price reflects investors' optimistic expectations for its transformation. In addition to breakthroughs in chip technology, the company's $15 billion share repurchase program and strong performance in its automotive and IoT businesses (both with revenue growth exceeding 20% year-over-year) have also boosted confidence. However, challenging Nvidia will be challenging—the latter has a strong ecosystem in the AI training sector, and Qualcomm needs to further demonstrate the stability and cost-effectiveness of its chips in large-scale deployments.
