The obvious question about an upgrade is whether the analyst is right, and it cannot be answered. A price target is a claim about earnings that do not exist yet; nobody settles a 2028 earnings claim in September 2026, and pretending otherwise is how coverage of analyst notes goes wrong. There is a question next to it that can be settled, because this analyst published his own numbers on the way down and again on the way up: what changed between the two calls? That one has an answer, and it is not the one the tape is reporting.
JPMorgan's Doug Anmuth upgraded Meta to Overweight from Neutral before the open on 10 September 2026 and raised his price target to $820 from $640. The same analyst cut the same name twice this year on the same theme — the returns on Meta's AI spending. What has moved since the second cut is his own capital-expenditure forecast: $243 billion for 2027, up from the $202 billion he published with the downgrade, and free cash flow of negative $65–70 billion a year where he had modelled negative $24 billion.
Four actions, one analyst, one theme
- 30 April, cutting to Neutral, he wrote that "full-stack AI competition is intensifying and Meta has a more challenging path to returns on heavy AI capex beyond advertising." That note carried 2027 capex of $202 billion and 2027 free cash flow of −$24 billion, and set $725 on 21x a 2027 GAAP EPS estimate of $34.01.
- 30 July, cutting the target to $640, the reason given was limited visibility into how Meta would monetise AI spending beyond advertising. He had, per the wires, "not come away learning much incremental on the developer API, consumer/business agents, or potential plans to directly monetize compute."
- 10 September, upgrading, the products named are the Muse AI agent and the Meta Model API — the developer API and the consumer agent from the July sentence.
- No multiple and no earnings year has been published for $820. The 21x and the $34.01 belong to April's $725. Nothing in the circulating excerpts says what $820 is 21 times of.
The spending objection did not get resolved. It got bigger.
This is the part the coverage has skipped. Anmuth downgraded in April over the path to returns on heavy AI capex. Between that note and this one, his own view of the capex got $41 billion worse for 2027 alone — from $202 billion to $243 billion, a 20% increase in his own forecast — and he added a 2028 figure of $284 billion.
The cash consequence moved with it. In April he modelled 2027 free cash flow at −$24 billion. He now models −$65 billion to −$70 billion in each of 2027 and 2028, against trailing levered free cash flow of +$41 billion. That is a swing of $106–111 billion from what the company is currently generating to what he expects it to burn — and it is his own arithmetic, not a bear's. The direction is already visible in the reported numbers: capex reached 49.5% of revenue in the June quarter and Meta's Rule of 40 fell to 30.83, its first sub-40 reading since 2023.
The internal check holds: $243 billion at the stated +70% implies a 2026 base of about $142.9 billion, inside the $130–145 billion Meta guided to with its June quarter. The forecast is coherent. It is simply worse, on the exact metric that justified two downgrades.
The 6.55% was not the upgrade
Meta rose 6.55% to $653.69 on Wednesday 9 September, on volume of 35.2 million shares against a three-month average of 17.8 million. That was the session after Meta launched Muse, and it happened roughly sixteen hours before Anmuth's note was distributed at 03:31 Eastern on Thursday.
The upgrade's own move was about 1% in Thursday's pre-market. Coverage crediting a $180 target raise with a 6.55% rally is crediting it with the previous day's product launch.
$820 is not a Street high. It is a catch-up.
Roth MKM had already printed $820 on 3 August, five weeks earlier. Four of the six most recent analyst actions this site carries sit above it, and the six average $840.83; the full-panel consensus of 62 analysts stood at $754.14 in mid-August.
More telling is what happened on Muse day itself. Three desks published on the launch and none of them raised a number.
Most recent published price targets, and the three desks that published on the Muse launch
| Desk | Target | As of |
|---|---|---|
| Monness | $890 | 30 Jul |
| CMB International | $880 | 31 Jul |
| Mizuho | $835 | 31 Jul |
| Jefferies | $825 | 3 Aug |
| Roth MKM | $820 | 3 Aug |
| Phillip Securities | $795 | 3 Aug |
| KeyBanc | $780 | 9 Sep · held |
| Morgan Stanley | $775 | 9 Sep · held |
| TD Cowen | $750 | 9 Sep · held |
The top six rows are the most recent sourced analyst actions this site carries, as of 19 August 2026; they average $840.83, and the full-panel consensus per S&P Global was $754.14 across 62 analysts on the same date. The bottom three are press-reported reiterations published on the 9 September Muse launch — each desk held its existing target rather than raising it. All nine sit outside JPMorgan's $640, which was the lowest of the group before 10 September.
JPMorgan's $640 was $110 to $140 below Morgan Stanley, KeyBanc and TD Cowen before Thursday. The upgrade closes that gap and steps modestly past it. And $820 sits $5 below the $825 the same desk carried in January. The round trip is the story, not the destination.
What Muse actually disclosed was a price, not a quantity
Meta launched Muse on 8 September in the United States with three tiers: free, Power at $20 a month and Maximum at $100 a month. It published no subscriber count, no adoption figure and no revenue. Anmuth's own framing, per the wires, is that "monetization is not the near-term priority for Muse."
Revenue is a price multiplied by a quantity. Meta has now given one of the two terms, which is genuinely more than it had given in July — and it is still not revenue. We worked what a paid Meta agent would have to earn in August, when the only price in circulation was a reported one.
Set the disclosed price against the disclosed business. Meta's advertising line ran $59,363 million in the June quarter, 97.6% of a $60,801 million company — $19.8 billion a month. Adding 1% to it takes about 9.9 million subscribers paying $20, or 2.0 million paying $100, gross of compute, payments and support, none of which Meta has disclosed for this product.
Meanwhile the $180 raise adds $461.9 billion of implied market value across 2.566 billion diluted shares: 7.6 quarters of Meta's total revenue, attached to products the company reports under no segment and has never printed a dollar for. Meta's reported segments are Family of Apps and Reality Labs. Muse, the Model API and business agents are not among them.
What would settle it
One dollar figure, attached to any one of the named products, in a Meta disclosure. That is the whole test. Until then the upgrade rests on an app-store rank — Muse reached number three in the United States on its second day, with early usage reported at ten times training cohorts — and a price list.
The counter-case is real. Anmuth may be early rather than wrong: a product cycle is visible before it is countable, waiting for disclosed revenue means buying it later and dearer, and July's complaint was that he could not see the pipeline — a product shipping with prices attached is a partial answer. Nothing here says $820 is unreachable. What the evidence says is narrower: between the downgrade and the upgrade, the only new observable is a price list, while the spending forecast that drove both cuts got $41 billion worse.
Sources and provenance: JPMorgan's rating changes, price targets, capital-expenditure and free-cash-flow forecasts, the EPS estimate and multiple behind the April target, the Muse app-store rank and usage figures, and every phrase quoted from Doug Anmuth are press reporting of a note this site has not seen — dpa-AFX, Investing.com, Seeking Alpha and TipRanks/TheFly — and are attributed to those wires rather than to the note itself; no multiple or earnings year has been published for the $820 target. Muse's launch date, tiers and prices are Meta's own, reported by Reuters; the company published no subscriber, adoption or revenue figure with it. The reiterations from Morgan Stanley, KeyBanc and TD Cowen on the launch are press-reported. Segment revenue is as Meta reported for the June quarter, and the capex guidance is the company's own. The six recent analyst actions and their average are the panel this site carries as of 19 August 2026, alongside the S&P Global full-panel consensus. Share prices are the 9 September close and the 10 September pre-market. Every ratio, subscriber conversion, implied market-value and swing figure here is our arithmetic on those numbers.
