Qualcomm used its Investor Day event to lay out its most detailed data center strategy yet, announcing a multigenerational CPU deal with Meta, two hyperscale customer wins, and a new AI inference server architecture. The company projects more than $15 billion in annual data center revenue by fiscal 2029, positioning itself as a broad AI infrastructure supplier rather than a chipmaker focused mainly on edge and mobile devices.
Meta Deal Adds Credibility
Meta CEO Mark Zuckerberg confirmed an agreement under which Qualcomm will supply CPUs for Meta’s next-generation server fleet, though Meta did not disclose deployment timing, processor specifications, or target workloads. A Meta spokesperson said the company is taking a “flexible, portfolio-based approach” that combines outside hardware partners with its own MTIA silicon program, suggesting Qualcomm will be one piece of a larger infrastructure mix rather than a replacement for in-house chips.
Analyst Matt Kimball of Moor Insights & Strategy said the deal matters less for its immediate revenue than for the validation it provides. A single hyperscale win does not upend the server CPU market on its own, he said, but it strengthens Qualcomm’s position when pitching additional cloud customers.
A Second, Unnamed Hyperscaler
Qualcomm executives said the company has signed two major hyperscaler agreements expected to generate at least $1 billion in combined revenue within a year, starting late this year. Meta is one of the two; the second customer was not named. Microsoft CEO Satya Nadella praised Qualcomm’s new High-Bandwidth Compute (HBC) architecture during the event but did not announce a commercial deployment.
Beyond a Single Chip
Rather than a standalone server processor, Qualcomm described a full portfolio spanning CPUs, AI accelerators, networking, custom silicon, and a software stack bolstered by its acquisition of Modular. Executives argued that conventional server designs cannot keep pace with agentic AI workloads and that the industry needs a structural shift in how infrastructure is built.
The centerpiece of that shift is HBC, which Qualcomm says pairs SRAM-class speed with HBM-class capacity to ease memory bottlenecks during AI inference. The company did not release performance benchmarks or technical implementation details, but Kimball said the technology could be significant if it delivers as described, particularly in disaggregated AI infrastructure where efficient data movement matters as much as raw compute.
What It Means for Operators
Qualcomm’s fiscal 2029 targets include more than $15 billion from data centers, alongside billions more from automotive, IoT, industrial and robotics, and personal AI compute. Handsets would fall to roughly a third of total chip revenue under that forecast. For hosting providers and enterprises watching the CPU and AI accelerator market, the emergence of a credible new entrant, backed by a hyperscaler commitment and a broader software strategy, could eventually mean more competition and more architectural options as AI infrastructure needs continue to grow.