CPU and GPU have become a symbiotic relationship, and TAM has expanded significantly.

```htmlCPU and GPU Become Symbiotic, TAM Expands SignificantlyAgent AI Requires More Than Just GPUs. GPUs are responsible for generating tokens, while CPUs handle tool invocation, program execution, retrieval, and database interactions. Therefore, overall system computation is crucial—the latency of the CPU and task completion speed affect GPU utilization and user experience.Server CPU TAM is expected to grow from $61 billion in 2026 to $211 billion in 2030, with about $180 billion related to AI. As a result, analysts reiterate their buy ratings for AMD, NVDA, and INTC, and raise target prices and earnings expectations for AMD.Demand Should No Longer Be Measured by “Cores per User”CPUs are not replacing accelerators; rather, they are expanding the overall AI system market. There are significant differences between vendors, workloads, architectures, and utilization rates; thus, it cannot be simplified into cores per user.More useful metrics are: how many tasks a GPU initiates, how many a CPU can handle simultaneously, how quickly results are returned, and the value added at the rack level.AI CPU shipments are expected to rise from 16 million units in 2026 to 53 million units in 2030, about 1.5 times the volume of accelerators (around 0.7 times in 2026).Different Approaches Give AMD, NVDA, and ARM Room to GrowNVDA opts for fewer but faster cores to return results with lower latency to the GPU; ARM emphasizes increasing cores for higher parallelism; AMD boasts the broadest product line, and Intel is moving towards broader workload support.Ultimately, the key is not single CPU specs, but completion time, number of agents per rack, memory, and cooling efficiency. AI CPU packaging value is expected to rise, with average server CPU prices climbing from around $1,600 to about $2,600.ARM Pursues New Markets, x86 Defends the Established MarketARM’s server market share could approach 40% by the end of this decade and may reach 50% if hyperscale cloud providers accelerate migration. However, x86 remains robust in traditional cloud and enterprise software, and ARM is better suited for new workloads. NVDA benefits from Rubin deployment, while a heterogeneous environment may favor AMD or Intel. Custom hyperscale cloud CPUs can help ARM earn licensing fees but may not directly drive commercial CPU revenue; there is also an $80 billion opportunity in on-premises enterprise CPUs.The server CPU market will expand rapidly and shift clearly towards AI. Total sales are projected to reach about $35 billion in 2025 and $211 billion in 2030, with a compound growth rate of about 36% from 2026 to 2030. Traditional/IaaS growth will be slower, dropping from nearly half the share to around 14%; AI-related (computing/head nodes + agents) will dominate, accounting for about 86% by 2030.Agent AI nodes will see the fastest growth. From about $6.4 billion in 2026, they are expected to reach $90.2 billion in 2030—a compound growth rate of about 94%—and their share will rise to about 43%, nearly matching that of computing/head nodes. The latter will grow from $36 billion to $90.2 billion over the same period. This means that future incremental server CPU demand will primarily come from AI, especially agent workloads, rather than traditional cloud needs.```