Qualcomm to Unveil Dragonfly AI Data Center Strategy on June 24 as Agentic AI Lifts CPU Demand 4x
Updated
Updated · Forbes · Jun 17
Qualcomm to Unveil Dragonfly AI Data Center Strategy on June 24 as Agentic AI Lifts CPU Demand 4x
3 articles · Updated · Forbes · Jun 17
Summary
June 24 is when Qualcomm plans to detail its data-center push at Investor Day, centering on Dragonfly-branded AI products aimed at inference rather than model training.
Agentic AI is driving the move because multi-step systems need far more orchestration, lower latency and heavier CPU control-plane work than one-shot generative AI workloads.
CPU demand per gigawatt of accelerator capacity could rise about 4x for agentic services, with deployments shifting from roughly 1 CPU per 4-8 accelerators toward 1-2 CPUs per accelerator.
Dragonfly includes AI200 and AI250 products that Qualcomm says use near-memory computing for more than 10x effective memory bandwidth, lower power draw and rack-scale inference efficiency, with direct liquid cooling on AI250.
The strategy would extend Qualcomm's strength in phones, edge AI and autos into hyperscale infrastructure, where power costs and political backlash over data-center energy use are creating an opening for efficiency-first designs.