Qualcomm outlines new CPU, AI accelerator roadmap, inks deal with Meta

Qualcomm outlined its latest data center offerings including the Qualcomm Dragonfly C1000 CPU, Qualcomm High Bandwidth Compute (HBC) and Qualcomm Dragonfly AI300 inference accelerator. Qualcomm also detailed custom silicon offerings.

The data center roadmap was detailed at Qualcomm's Investor Day. The announcements, which included a CPU deal with Meta, illustrate how Qualcomm is going after Nvidia in the AI data center market, which is shifting as AI inference workloads become dominant.

Qualcomm's broad message revolves around efficiency for AI agents. Qualcomm Dragonfly AI300 is the follow-up to the AI200 and AI250 on the company's annual roadmap. Ahead of the data center roadmap, Qualcomm announced the purchase of Modular, which will beef up its software efforts.

CEO Cristiano Amon said agentic AI is changing data center requirements. "Infrastructure has to deliver much higher performance at lower power and cost," said Amon, who noted that the company has a strong pipeline of multi-year agreements. Qualcomm's focus is on optimized token-per-watt and improved total cost of ownership.

Here's the breakdown of what Qualcomm announced:

• Qualcomm Dragonfly C1000 CPU. The C1000 is designed for AI agent workloads and optimized for power efficiency. The chiplet design has more than 250 cores and is greater than 2x better performance per watt compared to existing server CPUs. Qualcomm's CPU portfolio includes a chip for agentic orchestration and low-latency AI use cases; a general purpose version and AI head node CPU designed for XPUs. The catch is that C1000 will be commercial available in 2028 and that's a long time in AI years.
• Qualcomm HBC. Qualcomm HBC is a near-memory computing architecture that is designed to move data faster and more efficiently. Qualcomm plans to roll out commercial sampling of HBC Gen 1 with AI250 in mid-2027.
• Qualcomm Dragonfly AI300 cards and rack. Qualcomm said the AI300 will power a third-generation air- and liquid-cooled rack for AI inference. It will combine the AI300 with Qualcomm HBC Gen 2 technology and manages memory and bandwidth for AI agent workloads. Commercial sampling will land in 2028.
• Qualcomm Custom Silicon. The company said it will offer bespoke custom silicon for specialized AI workloads with advanced packaging and co-design offerings.

Qualcomm also raised its outlook and projected data center revenue of more than $15 billion by fiscal 2029.