For a decade, the AI build-out meant one thing for silicon: GPUs. This week the market repriced the other half of the equation. AMD's stock rose about 3% to a fresh high of $649.42 on Tuesday, leaving the company above a $1 trillion market capitalisation it first crossed on 21 September, and up roughly 32% in 30 days from about $478 at the start of September.

The immediate trigger is agentic AI. Personal agents such as OpenAI's Dots and Meta's Muse run long, multi-step tasks — reading files, driving browsers, calling tools, waiting on each other — and that work leans on general-purpose CPUs in a way single-shot model inference never did. AMD's EPYC server processors power both agents, and Morgan Stanley analysts estimate that Muse alone could account for around 20% of AMD's 2026 chip sales, according to market coverage.

Analysts followed the price. Citi lifted its AMD target to $800 and Stifel to $700, according to reports; Intel, the other large beneficiary of the CPU side of the agent boom, has risen by about a fifth over the same stretch. Both companies have outperformed the megacap GPU story over the past month.

It is a demand story with a supply answer attached. AMD chief executive Lisa Su has been outlining a multi-year capacity plan, telling investors the company is planning 2027 supply on a three-to-five-year horizon rather than a quarterly one — a signal that agent-driven CPU demand is being treated as structural rather than a spike.

The caveat is that the why is still contested. CPUs are cheap per core and easy to buy at scale; agents that spend much of their time orchestrating rather than multiplying matrices naturally skew there. But whether 'more CPU' stays true as agent frameworks move to specialised accelerators — or as inference providers consolidate onto whichever silicon is cheapest per useful token — is an open question. What is not disputed is that the trillion-dollar club now has a second chip company in it, and that the reason is not a training run.

Sources