Qualcomm CEO Cristiano Amon revealed on May 1 a dedicated CPU for agentic AI and a custom silicon deal with an unnamed hyperscaler. The company also teased 'agentic smartphones' as the next mobile computing paradigm.
Key facts
- Qualcomm building dedicated CPU for agentic AI in data centers
- Custom silicon deal with unnamed hyperscaler shipping December 2026
- Alphawave acquisition enabled custom ASIC capabilities
- Qualcomm expects 70% of Samsung SoC business in 2026-2027
- Agentic smartphones from ZTE and Xiaomi cited as examples
Qualcomm is repositioning itself as a custom silicon supplier for the agentic AI era. On its Q2 FY 2026 earnings call, CEO Cristiano Amon disclosed three strategic moves that signal a sharp pivot from the company's traditional mobile-chip identity.
Custom Hyperscale Silicon
Amon said Qualcomm will provide custom product to 'a leading hyperscaler,' with shipments expected 'in the December quarter' and planning for 'a multi-generation engagement.' The company gained custom ASIC capability through its acquisition of Alphawave [According to The Register]. Qualcomm is now working on a data center CPU and high-performance AI inference accelerators.
Dedicated CPU for Agentic AI
Amon revealed Qualcomm has built 'a dedicated CPU for agentic experiences in the data center.' His reasoning: AI kicked off with GPUs for training, then dedicated inferencing hardware, but the market now needs to 'generate demand for tokens' to power agentic AI. 'When you think about agents, CPU becomes very important,' he said. This contrasts with the GPU-centric narrative from Nvidia and AMD.
Agentic Smartphones
The CEO also introduced 'agentic smartphones,' citing Chinese handset-makers as early adopters. He mentioned a ZTE phone with ByteDance's Doubao personal assistant and Xiaomi's miclaw, an OS-kernel-integrated AI that divines user intent and drives third-party tools. Amon said smartphone designs are 'moving towards products [that] have much more capable CPU' and potentially more memory, though DRAM is currently in short supply.
Unique Take
The AP wire would report this as Qualcomm entering the custom chip business. The more interesting angle: Qualcomm is betting that agentic AI shifts the compute bottleneck from GPU memory bandwidth to CPU instruction throughput. If Amon is right, the hyperscaler's token-generation economics will favor Qualcomm's custom CPU over Nvidia's GPU clusters. That's a structural bet against the current AI infrastructure consensus, which has spent $250-300 billion annually on datacenter capex [per KG Intelligence].
Memory and Samsung Dynamics
Amon flagged a memory shortage hurting Qualcomm as Chinese smartphone makers build fewer units. He noted 'new memory players coming and building capacity' but said the situation needs monitoring into 2027. Separately, Qualcomm expects to win 70 percent of Samsung's SoC business this year and next, up from its usual 50 percent, as Samsung struggles with Exynos SoC quality.
What's Next
Qualcomm will hold an investor day in June to reveal more details. The hyperscaler's identity remains undisclosed, and the company did not share pricing, performance benchmarks, or volume commitments for the custom silicon deal.
What to watch
Qualcomm's June investor day will likely reveal the hyperscaler partner and performance targets for its custom CPU. Watch for benchmark comparisons against Nvidia's Grace Hopper or AMD's MI300 series, and whether the hyperscaler commits to multi-year volume guarantees.
