Qualcomm unveils AI 200 and 250 chip-based accelerator cards and racks

Qualcomm Technologies, Inc. today announced next-generation, inference-optimized solutions for data centers with the launch of the Artificial Intelligence 200 and 250 chip-based accelerator cards and racks. Building on the companys neural processing unit technology leadership, the products are presented as rack-scale offerings that target fast generative Artificial Intelligence inference while delivering strong performance per dollar per watt. Qualcomm framed the lineup as a step toward scalable, efficient, and flexible generative Artificial Intelligence deployment across industries.

The announcement highlights the Artificial Intelligence 200 as a purpose-built rack-level inference solution designed to reduce total cost of ownership and to optimize performance for large language and multimodal model inference and other Artificial Intelligence workloads. Qualcomm specified support for 768 gigabytes of LPDDR per card, which the company says provides higher memory capacity at lower cost and enables greater scale and flexibility for inference tasks. The Artificial Intelligence 200 is positioned to address the memory and economics of serving large models in a data center context.

The Artificial Intelligence 250 is introduced alongside the Artificial Intelligence 200 as part of the new chip-based accelerator card and rack family, although the company release focused on the Artificial Intelligence 200s memory and cost characteristics. Qualcomm emphasized rack-scale performance and superior memory capacity across the new solutions, framing them as engineered for high-performance generative Artificial Intelligence inference with an emphasis on efficiency. The company described the new products as building on its NPU technology leadership to deliver optimized inference performance and economics for data center operators and enterprises deploying large language and multimodal models.