Why Is META Stock Falling After Its Muse AI Rally?

Meta’s Muse AI helped lift the stock to a new 52-week high on September 24. Two trading sessions later, Meta shares had lost about 8%.

The surprising part is what happened in between: Meta did not report weaker earnings, and it even announced a new business unit for selling its AI tools to companies. The reversal looks less like a verdict that Muse failed and more like a fast reassessment of how soon a popular agent can earn back the money being spent to build and run it.

Let's break down the two-day drop, the news that did and did not change Meta’s prospects, and the numbers that can test whether this is a brief reset or a deeper problem.

How Much Has Meta Stock Fallen From Its 52-Week High?

Muse was introduced on September 8, when Meta closed at $613.48. After a surge around the agent and Meta Connect, the stock closed September 24 at $777.59. That was a 26.75% gain from the launch-day close. It touched $779.82 during the September 24 session, its new 52-week high. The next two closes went the other way. The figures below use unadjusted daily closes from S&P Global Market Intelligence’s historical data; the Nasdaq figures are index closes.

From September 24 close to September 28 close, the decline was 7.97%. From the September 24 intraday high of $779.82 to Monday’s close, it was 8.23%. Both explain an “about 8%” headline; they are different comparisons. Meta was still 16.65% above its September 8 close on September 28. This is a substantial retreat from a fast rally, not a return to its pre-Muse price. The Nasdaq Composite slipped only about 0.44% from Thursday’s close to Monday’s, so the overall market cannot explain the whole move.

For the original explanation of why Muse sent Meta stock higher, see our earlier analysis. For the product announcements and a detailed revenue model from the event, read our Meta Connect 2026 breakdown. The question here is what changed after that high.

Why Is Meta Stock Down After the Muse AI Rally?

First, the stock ran ahead of proof about profit. Muse’s early appeal gave investors something Meta’s large AI budget had lacked: a consumer product people sought out. But an app ranking tells us that people tried it, not how often they return or what Meta earns after paying for the computers that keep each agent running. The jump from “people want this” to “this meaningfully lifts Meta’s profit” is where expectations can outrun the accounts. Meta’s September 24 Connect recap expanded Muse’s planned reach to AI glasses and more retail, travel and work services, but did not provide Muse-specific revenue or profit figures. Those announcements made the product more plausible, while leaving its economics open.

Second, Friday refocused attention on the AI spending bill. Reporting on a Goldman Sachs analysis of the broader AI buildout brought the question of returns back into view. Goldman Sachs Research estimates that the largest US companies building AI infrastructure are on track for about $800 billion of capital spending in 2026. That is an industry estimate, not a new bill for Meta alone or a newly reported Muse loss. It matters because Meta was one of the names whose share price had risen on the promise that its spending would soon pay off. Friday’s 3.33% Meta decline against a rising Nasdaq is consistent with that reassessment.

Third, a promising agent will have competition. Ahead of OpenAI’s September 29 DevDay, The Verge reported on a possible rival consumer agent. OpenAI confirmed the event, but it had not, as of the September 28 close, confirmed the reported product’s features or launch. This distinction matters. A rumoured rival cannot erase Muse’s existing adoption, but it can make investors less willing to assume Meta will keep the entire opportunity or retain pricing power. The uncertainty is especially relevant when users could have several assistants available on the same phone.

Fourth, agent safety became an industry concern. OpenAI’s pause in training some advanced models amid concerns about agents acting outside instructions weighed on AI-related shares more broadly on Monday, according to Associated Press. This was not a disclosure that Muse itself had been shut down. It does, however, remind investors that an agent handling email, purchases or private information has to earn a higher level of trust than a chatbot answering a question. A slower rollout, more safeguards or hesitant users could all delay revenue across the category. The immediate competitive effect cuts both ways: delays for a rival may preserve Muse’s lead, while doubts about agents can hurt enthusiasm for everyone.

Finally, Monday’s market offered little support. The Nasdaq lost 0.92% as oil and US Treasury yields rose; the 10-year yield finished near 5.23%, according to AP’s market recap. When bond yields rise, investors often demand a better earnings payoff from shares valued on profits expected far into the future. That helped intensify Monday’s move. It cannot be the sole explanation because Meta had already fallen sharply on Friday when the Nasdaq rose.

One often-cited limit on Muse’s shopping ambitions also needs the right date. Amazon’s restriction on Muse shopping access was known before the September 24 high. It is a genuine constraint on where the agent can complete a purchase, but it was not a fresh Monday announcement. Our separate look at why Amazon blocked Muse covers the platform dispute. A retailer allowing an agent to enter its store and Meta earning a meaningful fee are still two different steps.

Why Did Meta Fall Even After Announcing an Enterprise AI Business?

On September 28, Mark Zuckerberg announced the Meta Enterprise Platform, bringing the Muse agent, Meta Business Agent, Muse API, Muse Code and other tools together for businesses and developers. Former MongoDB chief executive CJ Desai will lead it. That gives Meta another route to turn its models and infrastructure into products customers could pay for. It is a real strategic development, not merely a new label on Muse.

Yet the announcement included no enterprise sales target, customer count, margin, price list or timetable for a material contribution. On a day when investors were asking how the existing spending earns a return, a new division supplied a possible answer but no measurable result. That is why the negative share-price reaction should not be read as proof that the enterprise idea is bad. Our reading is narrower: after the run to $777.59, a further promise was no longer enough to extend the rally.

There is a useful distinction here. Meta can have a valuable consumer agent, a sensible enterprise plan and a stock that became too confident about the speed of both payoffs. Those statements can all be true at once.

Is Meta’s AI Spending Becoming a Financial Problem?

Meta’s latest reported quarter gives the answer more clearly than a debate about whether Muse is exciting. Its advertising business is still large and growing, but more of the cash it produces is going straight back into infrastructure.

These are company-wide figures, not a Muse profit-and-loss statement. In Q2, revenue grew 28% but total costs and expenses grew 55%, partly because the quarter included a $2.40 billion legal charge and $1.18 billion of severance costs. Free cash flow fell about 91% from Q2 2025. That does not mean the ad business stopped producing cash: Meta reported $90.26 billion in cash, cash equivalents and marketable securities at the end of June, alongside $83.66 billion of long-term debt. The near-term issue is the return on incremental spending, not an imminent inability to fund operations. Meta’s figures also show why a profitable-looking AI product has to be judged after its computing and operating costs, not just by users or gross sales.

The timing creates a trap for investors. Meta pays to build capacity now. A successful Muse may bring more usage quickly, but additional work can require more computing before enough users or businesses pay. Stronger adoption could therefore coincide with thin free cash flow for a while. Conversely, if AI improves ad performance across Facebook and Instagram, the return may arrive inside advertising revenue rather than as a separate Muse line. Anyone demanding a large, neatly labelled Muse revenue figure could miss that benefit; anyone counting every new agent user as profit could make the opposite mistake.

What Would Make This Pullback Look Temporary or More Serious?

The next test is a bridge from product activity to company cash, not a return to an app-store ranking. Meta has already given a Q3 2026 revenue guide of $61 billion to $64 billion. That guidance was issued in July, before Muse’s September launch, so the next results and management commentary can show whether advertising keeps funding the buildout and whether new AI products are beginning to contribute. These are the specific questions to put beside the next report:

Our view is that the selloff has removed some of the certainty investors attached to Muse, and that is sensible. The bullish part of the case remains: Meta has a large, cash-generating advertising business, broad distribution and an agent that found immediate consumer interest. The weak part is equally concrete: the September announcements have not yet shown that Muse and the new enterprise platform can produce enough incremental profit to justify a lasting re-rating while spending stays high. The stock’s two-day fall does not prove the AI strategy failed. It also does not, by itself, make the shares inexpensive.

The most useful question after the 8% drop is therefore simple: Is Meta turning more completed AI tasks into more cash per dollar of infrastructure, or merely turning excitement into more spending? The answer will come from retention, paid usage, advertising performance and free cash flow over several reporting periods. September’s price action tells us expectations moved much faster than those measurements.