UncoverAlpha argues that as agentic AI workloads scale, CPUs—not just GPUs—are becoming the binding constraint in datacenter buildouts, since orchestration, tool calls, and memory-bound serving steps stress general-purpose compute. This favors high-core-count server CPU suppliers (AMD EPYC, Intel Xeon, Arm-based hyperscaler silicon) and CPU-adjacent infrastructure (memory, networking) over the next capex cycle.
Why it matters: Sector-wide thesis reframing CPU demand within the AI infra cycle—relevant for AMD, Intel, Arm and adjacent memory/networking names but not a discrete event.
MRVL
$218
-7.75%