From Qualcomm's vantage point, the AI200 and AI250 deliver rack-scale performance and superior memory capacity at an industry-leading total cost of ownership (TCO), with optimized performance for large language models (LLMs) and large multimodal models (LMMs). The cards interface to the host via PCIe and to the BMC via SMBus/I2C. The Cloud AI 100 Ultra SKU has 4. Qualcomm is moving from mobile NPUs into rack-scale AI infrastructure, positioning its AI200 (2026) and AI250 (2027) to challenge Nvidia/AMD on the economics of large-scale inference. Qualcomm is entering this evolving market with its Dragonfly platform, aiming for $15 billion in revenue by 2029. The stock soared 11% following the news.
[PDF Version]We Look Forward to Working with You