Meta's Muse lights another fire under the AI trade

Key points:

• Meta’s Muse surged in downloads and user ratings, helping drive a sharp rally in Meta shares.
• Investors extrapolated strong demand for agentic AI into greater data-centre CPU needs, lifting Arm, Intel and AMD.
• Arm reacted most strongly on its direct Meta partnership and broader exposure to cloud CPUs.
•  The move highlights how sensitive tech stocks remain to both AI narratives and swings in oil prices and geopolitical risk.

What happened?

A new AI product from Meta suddenly became one of the biggest market stories of the week.

Meta launched Muse, its new personal AI agent, in the US on September 8. Adoption initially looked solid rather than spectacular, but has since accelerated rapidly. By Monday September 21, Muse had reached the top of the US free-app rankings on both Apple’s App Store and Google Play. Sensor Tower estimated that the standalone app was downloaded more than 900,000 times in its first six days, already ahead of the equivalent launch period for Meta’s previous standalone Meta AI app. Muse is also accessible through WhatsApp and the web, which are not captured in the standalone-app figures.

Early user ratings have also been unusually strong. At the latest reading, the US Apple App Store showed Muse with a 4.9 out of 5 rating from around 18,000 ratings, while Google Play ratings were also close to five stars, although these numbers are changing quickly as the user base expands.

Investors reacted aggressively. Meta shares jumped 11.4% on Monday, their strongest one-day performance in well over a year, adding roughly $190 billion of market value. The move helped carry the Nasdaq Composite up 2.3% to a record closing high, while the S&P 500 gained 1.5% and finished within about 0.4% of its record.

That is a large market response to what is so far primarily evidence of strong early consumer adoption rather than a material new source of reported earnings. Still, Meta’s vast user base makes for a vast total addressable market of potential customers. And the app is free to try and use, only requiring subscriptions of USD 20 per month and up to USD 100 per month depending on the intensity of usage.

What is Meta Muse?

Rather than simply answering questions, Meta describes Muse as a personal AI agent that can perform tasks on a user’s behalf. Users can ask it to manage email and calendars, make restaurant or travel reservations, fill in forms, research products, track spending or subscriptions and shop online. It can continue working after the user closes the app and can break larger jobs like planning a holiday or an event into multiple subtasks.

An important part of the architecture is that each Muse runs in what Meta calls a Muse Secure VM, essentially a dedicated virtual computer in the cloud, with its own browser and access to services that the user chooses to connect. Muse can therefore navigate websites and interact with applications rather than relying exclusively on specially designed AI interfaces or APIs.

Meta says users must approve sensitive actions such as sending an email or making a purchase, while passwords and payment information are kept separately from the AI model itself.

That design has already exposed one of the potential complications of agentic AI. Amazon has blocked Muse from shopping on its site, arguing that the agent has not been authorised and does not adequately identify itself when browsing. The dispute echoes Amazon’s earlier fight with Perplexity over its Comet agent and points toward a potentially important future battle over whether AI agents should be treated as extensions of the user or as third-party commercial services requiring permission from websites.

What were the knock-on effects?

The most striking market reaction outside Meta itself was in semiconductors, and particularly CPUs rather than the GPUs that normally dominate the AI story.

Arm Holdings rose 17.1%, Intel 12.1% and AMD almost 10% on Monday. The broader Philadelphia Semiconductor Index gained a still-impressive 4.3%, making the outperformance of the CPU names particularly conspicuous. AMD also closed above a $1 trillion market capitalisation for the first time.

The logic behind the move is that agentic AI could require much more conventional computing capacity than the current chatbot model.

Large AI models still rely heavily on GPUs and other accelerators to perform inference. But an agent like Muse also needs CPUs to operate its virtual machine, browser, APIs, databases, networking and storage, execute code and coordinate the many individual tasks taking place around calls to the underlying AI model. It is important to point out that the CPUs required to run Muse agents are server CPUs in the cloud, not CPUs in a personal computer or on a smartphone.

One physical server CPU can host many virtual machines (the Muse Secure VMs), so one Muse user does not remotely translate into one CPU. Nevertheless, if persistent AI agents eventually number in the tens or hundreds of millions, investors are extrapolating that this could create a substantial new source of demand for general-purpose data-centre computing alongside continued demand for GPUs.

Why did Arm react the most?

Arm had an additional reason to attract attention. Meta is already Arm’s lead partner and co-developer for the Arm AGI CPU, Arm’s first finished data-centre processor. The companies have committed to developing multiple generations of CPUs together for Meta’s AI infrastructure, with the Arm processors designed to work alongside Meta’s own MTIA AI accelerators.

Muse therefore provides a particularly convenient narrative for Arm: Meta launches a consumer AI agent whose architecture potentially creates large amounts of CPU demand, while Meta is simultaneously working directly with Arm on CPUs designed specifically for large-scale agentic AI.

There is a broader angle as well. Arm architecture already underpins Amazon’s Graviton, Google’s Axion and Microsoft’s Cobalt cloud CPUs, as well as Nvidia’s Grace and next-generation Vera processors. So if agentic AI increases overall CPU demand, Arm potentially participates across several competing data-centre ecosystems rather than relying on one manufacturer winning market share.

The smartphone angle is less important to Monday’s reaction. Muse is accessed from a phone, but much of its work is performed in the cloud. Again, the more direct potential impact is on demand for data-centre CPUs, not for processors inside smartphones.

The enthusiasm also spread into other parts of the AI infrastructure complex, including networking, cloud and edge-security stocks. Here the picture is less clean: companies including Fastly and Akamai also had their own AI-related product and positioning developments, so not every move should be attributed directly to Muse.

Zooming out: an important signal, but also an example of the market we are in

Monday's price action illustrates how hungry many investors are for anything that appears to validate the next stage of the AI investment story.

Muse's early adoption is potentially significant. It provides some evidence that consumers may be interested in moving from AI systems that merely answer questions toward agents that can actually carry out tasks. If that transition happens at scale, given that agentic AI is far requires far more network capacity and computing resources than chatbot queries, the implications would stretch far beyond Meta, touching CPUs, GPUs, cloud infrastructure, networking, security, payments and e-commerce.

For now, the market has grabbed the ball and run with on this story, suggesting significant extrapolation of where we are headed. But app-store rankings and initial ratings do not tell us what Muse's long-term user retention will be, how frequently people will use agents, how much computing capacity each active agent will consume or how successfully Meta can monetise that usage. Nor does a successful consumer application immediately translate into additional orders for Arm, Intel or AMD.

This latest bout of enthusiasm comes against an already strong market backdrop. The Nasdaq Composite closed at a new record on Monday (the more closely followed Nasdaq 100 index came up a bit short of its record high, as did S&P 500). That leaves markets enjoying considerable momentum, but it also means baked-in growth expectations are also very high.

There was another important tailwind for technology stocks on Monday and early Tuesday that should not be overlooked: oil prices fell sharply. Brent dropped well below $100 per barrel early Tuesday and WTI below $94 as markets priced some possibility of diplomatic progress in the Middle East. Falling oil helped pull bond yields lower and eased concerns that higher energy prices would feed another round of inflation, a particularly helpful combination for long-duration growth and technology stocks.

That backdrop could improve further or take a sudden turn for the worse, a risk we have become well acquainted with since the war with Iran broke out over six months ago. At the time of this writing, investors are holding their breath on whether US President Trump will meet Iranian President Masoud Pezeshkian on the sidelines of the UN General Assembly in New York. Trump has said he is open to meeting the Iranian president, and news agencies have circulated talk of Iran’s openness to allowing shipping traffic to proceed undisturbed through the Strait of Hormuz if the US changes its stance.

That makes energy prices an important near-term swing factor for broader risk sentiment. A credible diplomatic opening could further remove some of the geopolitical and inflation risk premium embedded in oil, while a setback or renewed escalation could quickly reverse that dynamic.

For investors, the Muse episode therefore contains two separate messages. The first is technological: agentic AI could broaden the AI infrastructure opportunity well beyond GPUs and into general-purpose CPUs and the wider internet infrastructure stack. The second is about markets: with major equity indices already close to record highs, both AI expectations and the oil-driven inflation backdrop are capable of producing unusually large and rapid shifts in sentiment.

Muse may eventually prove to be an important milestone in consumer AI. For now, the extraordinary reaction across Meta and the semiconductor sector tells us at least as much about the market's willingness to extrapolate the next phase of the AI story as it does about the commercial value of an app that has been available for only two weeks.