Nobody at Arm, Intel or AMD announced anything on Monday. No earnings, no guidance, no new contract. What moved the three biggest names in CPUs was a consumer app: Meta’s Muse agent, which has sat at the top of the free chart on Apple’s US App Store for three straight days, according to Sensor Tower data cited by Barron’s. By the close, Arm had climbed 17%, Intel 12% and AMD 10%, with AMD ending the session above $1 trillion in market value for the first time, per Bloomberg. Back-of-envelope math puts the combined gain at roughly $200 billion in a single session.
The rally was sharply concentrated. 24/7 Wall St. clocked Arm at $312.46 (up 13%), Intel at $121.38 (up 12%) and AMD at $611 (up 9%) late Monday morning, while the iShares Semiconductor ETF rose only 3% and the Invesco QQQ Trust 2%. The Philadelphia Semiconductor Index finished up 4.3% for its fifth straight gain. The gap between the three CPU names and the broader chip index is the point: this was not a general semiconductor bid but a specific repricing of the processors that run the orchestration, data movement and tool-calling that agents generate when they act rather than merely answer.
Muse turns a demand theory into a demand signal
Meta launched Muse earlier this month as a personal agent that sends emails, books travel and works on longer-running projects, with each instance running inside its own dedicated cloud virtual machine. That architecture is why the app matters to chipmakers. Agentic workloads are heavier on general-purpose compute than chat, because every step of a multi-step task involves scheduling, memory management, network calls and code execution that land on CPUs alongside the accelerators doing the model math. The Register reported earlier this month that harnesses such as OpenClaw were already reshaping CPU demand in data centers; Muse is the first consumer product with mass adoption to make that shift visible to equity investors.
Meta has been saying as much for months. In April the company said that as its agentic AI work advances, “compute requirements are evolving to demand more CPU,” and it has described its data centers as “outgrowing the abilities of traditional CPUs.” Meta is also the lead partner and co-developer on Arm’s AGI CPU, the first data-center processor Arm designed itself, unveiled in March for agentic workloads and meant to run alongside Meta’s MTIA accelerators. Arm has since said customer demand for that part exceeded $2 billion across fiscal 2027 and 2028, and that CPU requirements are rising by at least four times as application workloads become more agent-driven. Stocktwits noted, however, that neither company has said Muse actually runs on Arm’s AGI CPU.
Intel cannot build enough of them
The supply side of the story came from Intel CEO Lip-Bu Tan, who told Cisco president Jeetu Patel at Splunk’s .conf26 conference in Denver last week that the company is badly capacity constrained. “CPU demand is so high that we can only supply 50% of customers,” Tan said, according to CTech. “Many CEOs are calling me apologizing for not being able to produce enough,” he added. Tan tied the shortfall directly to agents: “There are numerous agents, and in the future millions or trillions of agents will require resources. Sufficient computing power and security are needed to respond to this.”
That admission gave Monday’s buyers a reason to believe incremental demand lands on an order book that is already backlogged. Intel’s second-quarter revenue rose 25% year over year to $16.1 billion, its strongest growth in 15 years, with the data center and AI segment up 59% to $6.3 billion. Adjusted gross margin has climbed from 29.7% a year ago to 41.8%. Chief financial officer Dave Zinsner said alongside those results that “we are meaningfully increasing our investments in equipment, clean room space, and substrates,” and Tan said the 14A process begins production in the first quarter of 2027.
The three companies reach the same thesis by different routes. Arm earns royalties on architecture and owns no fabs, so a broad rise in server CPU volume reaches it without winning individual sockets; Arm estimates the server CPU market can grow more than 35% a year to $120 billion by 2030. Intel designs and manufactures its own chips, so demand must pass through its own constrained factories. AMD, which outsources manufacturing, has projected an even larger $220 billion server CPU opportunity through 2030.
Why it matters
For two years the AI trade has been a GPU trade, and the CPU was treated as commodity plumbing. Monday was the clearest signal yet that the market is re-rating the plumbing. If agents rather than chatbots become the dominant consumer interface, the ratio of general-purpose compute to accelerator compute in every inference deployment rises, and that shifts value toward Arm, Intel and AMD without requiring Nvidia to lose. The risk is equally clear: the trigger was a download chart, not a purchase order. A consumer signal sits two removes from Arm’s royalty line and can fade as fast as it appeared. Arm was already up more than 178% this year before Monday, and Intel trades above 50 times next year’s expected earnings, so the market is pricing in capacity that does not yet exist.
What to watch
Watch whether Muse holds the top of the App Store past its launch-week novelty, and whether Meta discloses anything about the CPU mix behind it. Watch hyperscaler capital expenditure commentary in the coming weeks for signs that agent inference is changing server orders, not just sentiment. And watch Intel’s third-quarter report in October: guidance of $15.8 billion to $16.8 billion was set in July, before Tan said the company can fill only half its demand, so any update on capacity, substrates and the 14A timeline will tell investors how much of Monday’s rally the fabs can actually deliver.
“CPU demand is so high that we can only supply 50% of customers.”— Lip-Bu Tan, CEO, Intel