The release of personal AI agents such as Meta Muse is shifting artificial intelligence from passive, responsive chatbots toward autonomous agents that operate around the clock, a transition that could drive up consumption of CPUs, GPUs, memory and networking infrastructure.

According to a research report released by Citi on October 6, agent applications represented by Meta Muse are pushing market expectations for computing hardware to new heights. The bank estimates that by 2030, the total addressable market for CPUs globally will expand from $29 billion in 2025 to $300 billion, a compound annual growth rate of 60%.

On the GPU side, agents are likewise creating structural incremental demand. The bank calculates that if Meta Muse reaches 100 million daily active users, it would require roughly 200,000 to 390,000 Blackwell-class GPUs, potentially generating $7 billion to $19 billion in one-time revenue for NVIDIA (NASDAQ: NVDA).

This trend directly benefits core computing power suppliers. The bank notes that because Meta is one of the largest customers of AMD's server business, AMD will be a primary beneficiary of the CPU revival, with its target price raised to $800; at the same time, NVIDIA (NASDAQ: NVDA) will also continue to gain from surging GPU demand.

The Rise of Agents Makes CPUs the New Bottleneck

Before the explosion of agentic AI, CPUs were largely confined to traditional workloads and to serving as head nodes for AI applications. In the head node role, the CPU was only responsible for "management," sending user requests to the GPU and returning results, while the heavy matrix multiplication and inference work was handled by the GPU.

Agents, however, have changed this division of labor. Citi said in the report: "Compared with traditional chatbots, we believe agentic AI is a potential order-of-magnitude driver of computing demand." Agents need to handle orchestration, reasoning loops, data processing and security components, which makes the CPU the new bottleneck.

As AI shifts from model training to inference and then to autonomous agent workflows, the CPU-to-GPU ratio is changing significantly. During the model training phase, the ratio was 1:8; during inference it was 1:4; and in agentic AI, the ratio is moving toward 1:1 or even higher. The bank expects that by 2030, CPUs dedicated to agents will grow at a 247% compound annual growth rate, accounting for 52% of the entire CPU market.

Meta Muse's GPU Computing Ledger

Meta Muse is the first consumer AI agent released at true social network scale, and its computing footprint is one of the most important variables in today's AI infrastructure. Unlike standard chatbots, Muse operates continuously as an autonomous agent, and its underlying logic dictates far greater GPU consumption.

Citi built a bottom-up GPU demand model. Under the base case, it assumes a typical user makes 4 simple requests and 12 agent tasks per day, such as finding a restaurant, checking a calendar and drafting an invitation, with each agent task containing about 8 model calls. Because the model has no memory between calls, it must reread the growing conversation context each time.

Based on this, the bank estimates that each Muse user requires roughly 0.0020 of a GB200-class GPU. In short, one GPU can serve only about 500 users. The bank stressed: "Agents are structurally heavier inference workloads than chatbots." Under the same framework, the GPU capacity required by chat-style users is only 12.5% to 25% of that required by agent users.

The Knock-On Effects on Memory and Networking

Agent workflows not only place higher demands on logic chips; their knock-on effects also spread to memory and networking. On the memory side, CPUs have a high attach rate for LP, DDR and SSD. Micron recently noted that agent workflows represented by Meta Muse are helping consumers capture more value, while CPUs are placing more constraints on DRAM.

On the networking side, agentic AI substantially increases network traffic. NVIDIA (NASDAQ: NVDA) management said in communications with Citi: "Agents can run for hours or continuously... thereby driving far more token generation across consumer and enterprise workloads."

This continuous operation means that more and more of the entities accessing infrastructure are agents rather than humans. This drives growth in "north-south" traffic within data centers, such as storage access, security and configuration services, creating new acceleration opportunities for the use of DPUs such as BlueField 4 and Spectrum-X Ethernet.