Agentic AI reshapes computing power demand! Citi: CPU market size to hit $300 billion by 2030, raises AMD (AMD.US) target to $800

Citi's latest research report points out that with the successive launch of personal AI agents such as Meta Muse, agentic AI is becoming an order-of-magnitude driver of computing power demand. Citi has raised its 2030 CPU total addressable market (TAM) to $300 billion and lifted its AMD (AMD.US) target price from $575 to $800, maintaining a "Buy" rating. Citi believes AMD is the primary beneficiary of the CPU demand surge, as Meta is one of AMD's largest server business customers. Meanwhile, Citi maintains its "Buy" rating and $315 target price on NVIDIA Corporation (NVDA.US), citing rising GPU demand. Agentic AI reshapes computing power demand, CPU market size targets $300 billion Citi stated in the report that recently, multiple personal AI agents including Muse, Dots, Instinct, GrokBot, and GeminiSpark have been launched successively. Unlike chat Siasun Robot&Automation, which remains largely idle between user interactions, "always-on" AI agents like Muse can run continuously, driving significant increases in inference-related computing, memory, and network consumption across enterprise and consumer workloads. Citi has updated its CPU model, believing that agentic AI represents a potential order-of-magnitude driver of computing power demand relative to traditional chat Siasun Robot&Automation. Citi has raised its CPU TAM from $29 billion in 2025 to $300 billion by 2030, corresponding to a 60% compound annual growth rate. Citi still expects AMD to be the primary beneficiary of the CPU demand surge, with Intel Corporation as the secondary beneficiary. Why has CPU become the new bottleneck? In agentic AI, CPU has become the new bottleneck. The CPU-to-GPU ratio is shifting from 1:8 during model training, to 1:4 during inference, and to 1:1 or higher in agentic AI. Citi believes that as agentic AI scales up, this ratio will increasingly tilt toward CPU. Traditionally, CPU has been mainly limited to two major applications: traditional workflows, and serving as the head node for AI applications. In the head node, CPU is only responsible for "management" sending user requests to the GPU and returning outputs to the user. The GPU handles the heavy lifting, such as running matrix multiplications to train large language models, or performing inference when querying chat Siasun Robot&Automation. The third new application is agentic AI. Citi defines agentic CPU as the CPU that handles the orchestration, reasoning loops, data processing, and security components of autonomous AI agents. This new application has exploded over the past year as AI shifts from chat Siasun Robot&Automation inference to agentic autonomous workflows. Consumer AI agent GPU demand estimation Meta's Muse is the first consumer AI agent launched at real social network scale, and its computing power demand is one of the more significant unknowns in AI infrastructure. Under Citi's base case, each user requires approximately 0.0020 GB200-class GPUs. Based on 100 million daily active users, this corresponds to approximately 200,000 to 390,000 Blackwell-class GPUs, approximately 2,800 to 5,500 NVL72 racks, and approximately $7 billion to $19 billion in one-time NVIDIA Corporation revenue (including scale-out networking). On a per-user basis, every additional 1 million AI agent daily active users, if using new servers, would bring approximately $70 million to $190 million in revenue for NVIDIA Corporation. Memory and networking: CPU drives DRAM and networking demand On the memory side, CPU has a high attach rate for LP, DDR, and SSD. Micron expects server units to achieve high double-digit growth in CY26/27, and believes another category of logic chips will also benefit from the AI trend. Micron recently also stated that CPU is placing greater constraints on DRAM, and Meta Muse is an example of consumers deriving value from agentic workflows. On the networking side, agentic AI significantly increases network traffic, as AI agents increasingly access infrastructure, creating new acceleration opportunities for DPU and Spectrum-X Ethernet.