Qualcomm Unveils 768GB AI200 Servers, Secures $1B 200MW HUMAIN Deal
QCOM•Qualcomm has launched AI200 and AI250 inference servers offering 768GB memory per card, undercutting competitors' 288GB and 180GB capacities to reduce energy costs and footprint. It secured a 200 MW, roughly $1 billion deployment deal with Saudi Arabia's HUMAIN initiative, positioning inference as a major growth driver.
1. Qualcomm’s Strategy Shifts to AI Inference
Qualcomm is targeting the growing AI inference market by adapting its power-efficient smartphone processing expertise to data centers. The company emphasizes performance per watt, leveraging its custom Oryon CPU architecture and integrated chip designs to address rising electricity costs and power constraints in AI infrastructure.
2. AI200 and AI250 Server Specifications
The AI200 and AI250 servers feature 768GB of memory per card, more than double AMD’s MI350X and over quadruple Nvidia’s comparable GPUs. These systems prioritize capacity over peak speed, allowing deployment of large language models on fewer, lower-power systems, reducing total cost of ownership.
3. HUMAIN Deployment and Financial Impact
Qualcomm secured a 200 MW deployment deal with Saudi Arabia’s HUMAIN sovereign AI initiative, estimated at roughly $1 billion. This marquee contract underscores Qualcomm’s inference strategy and is expected to boost royalty and product revenue as the company scales infrastructure chip shipments.
