Meta's Muse Agent Lifted Intel 12.5%, Nvidia Only 2.3%

Three chip companies were repriced at once on Monday.

Intel rose 12.53%, Arm 10% and AMD 9.97%. AMD closed at $614.91 and crossed $1 trillion in market value for the first time during the session.

None of the three announced anything that day. In the same session Meta gained 11% to $741.25, the PHLX Semiconductor Index rose more than 3%, and the Nasdaq closed at a record 27,122.09.

What set the move off was not a chip.

What Got Priced Was a Phone App

Meta released a personal agent called Muse on September 8. It is priced between $20 and $100 a month and is meant to carry out errands on a user's behalf: writing email, booking travel, making purchases.

On September 18, ten days after launch, it reached No. 1 on Apple's US free-app chart. ChatGPT fell to second, Gemini to eighth, Claude to thirteenth.

By Sensor Tower's count, more than 730,000 US downloads sit behind that ranking.

Nothing else has been disclosed. No subscriber figure, no revenue.

That all three companies moved by such similar amounts on one signal is worth noting, because their businesses are not alike. Arm makes no chips and collects royalties per unit on its designs. Intel designs its processors and prints them in its own fabs. AMD designs and outsources manufacturing. They share one thing: all three sell server CPUs.

The market did not price Meta's subscription revenue on Monday.

It priced how many processors those subscriptions would take to run.

Why an Agent Keeps the CPU Busy

A chatbot runs once: a question arrives, the model answers, the job ends. The heavy load sits on the GPU and it is brief.

An agent does not work that way. It breaks a job into parts, queues them, calls outside tools, judges what comes back and retries when it fails.

That coordination is not GPU work. Queue management, memory layout and tool calls sit on the CPU — and they sit there without pause. In a chatbot the CPU steps in for short bursts; in an agent that runs continuously it never goes quiet.

Intel chief executive Lip-Bu Tan said at a conference in Denver last week that the company can meet only about half of what customers are asking for. He pointed to the growth in work models do after training.

Nvidia Stopped at 2.30%

The most instructive figure of the day is not among the risers but in the one that did not rise.

Nvidia closed at $227.38. On a day explained by AI demand, the largest beneficiary of that demand lagged the index.

This does not mean money left the AI story. It means money moved inside it: a rotation within the sector rather than between sectors.

Muse was not the only reason for the day, either. Oil fell 4.42% to $91.83, the 10-year Treasury yield slipped below 5%, and executives from AI companies will attend the Trump-Xi dinner at the White House on Thursday, September 24. All three fed the appetite for risk.

The Analyst Who Did the Math Disagrees

The scale Muse needs to turn into revenue had been calculated the day before Monday.

According to a September 20 note from Oppenheimer analyst Jason Helfstein, Meta would need 115 million subscribers paying $20 a month to earn roughly $28 billion a year from its agent.

The same note does not expect that in the near term. Its reasons: the paid conversion rate, competition from ChatGPT and Gemini, and users' reluctance to hand Meta their passwords.

The note's benchmark is ChatGPT's free-to-paid conversion rate of about 6%. At that rate, 115 million paying subscribers would require roughly 1.9 billion free users.

Downloads counted so far: 730,000.

What Is Left

The demand Intel cannot half-fill is real and measured. How much Muse added to it is not.

The chart shows only the past month's path, not the reason for Monday.

This article draws on US closing data for September 21, 2026, Bloomberg and CNBC reporting on AMD's $1 trillion market value, accounts of Muse's App Store ranking, and CPU-to-GPU ratio estimates from TrendForce and Arm. The download figure is reported by Sensor Tower; Meta has disclosed no subscriber or revenue figure. Lip-Bu Tan's remarks are summarized from conference accounts and are not a verbatim quotation.