BofAS published a research report stating that in AI inference, GPUs are responsible for generating tokens for "thinking", while CPUs execute tools for tasks such as calling functions, code execution, retrieval and database interaction. Therefore, Agentic AI requires full-system computing, and CPU latency and task completion speed are becoming increasingly important to GPU utilization and end-user experience. The broker believes Agentic AI makes CPU and GPU demand complementary. It currently forecasts the total addressable market (TAM) for server CPUs to increase from USD61 billion in 2026 to USD211 billion in 2030, of which around USD180 billion will be AI-related.

BofAS also forecasts AI-related CPU shipments to grow from 16 million units in 2026 to 53 million units in 2030, equivalent to around 1.5x accelerator shipments, compared with around 0.7x in 2026. Experts also expect ARM architecture server CPU unit market share to approach 40% by the end of the decade, and potentially reach 50% if hyperscale cloud operators accelerate migration and NVIDIA Corporation (NVDA.US) promotes broader ARM adoption. However, x86 still has an established position in traditional cloud and enterprise software, while ARM is more suitable for new workloads.

BofAS reiterated the "Buy" rating on Advanced Micro Devices, Inc. (AMD.US) , viewing it as its top CPU pick. The broker raised its 2027 and 2028 earnings forecasts by 2% to 3% and lifted the TP from USD620 to USD720. The broker also reiterated the "Buy" rating on NVIDIA Corporation (NVDA.US) , naming it the sector top pick, citing its most comprehensive product portfolio, CPU/GPU interchangeability and supply chain investment, with a TP of USD350. At the same time, it reiterated the "Buy" rating on Intel Corporation (INTC.US) , expecting it may secure incremental CPU foundry capacity services for external clients over the long term, with a TP of USD145. (gc/u)(Real-time Streaming US Stocks Quote; Except All OTC quotes are at least 15 minutes delayed.)