Amazon Shut Muse Out of Its Store. Meta's Own Compute Is the Tighter Limit.

Article published Sep 25, 2026. Prices below use latest available snapshots.

META $709.33 +5.3% 30d AMZN $250.00 +1.6% 30d

Meta is doing well, and has been since it launched Muse, a personal AI agent, on September 8. The stock is up 30.46% in a month, and its one-year return turned positive again this week. Two limits now sit on the story. Amazon blocked Muse from shopping its store (reported September 21). Meta is also reported to be holding back Muse's WhatsApp rollout because it lacks the processors and memory to serve it. The call is to wait for a retest of the September 21 gap at $670–$690, not to buy Friday's 3.33% reversal from the high.

Some claims about Muse usage, capacity and suppliers go beyond what Meta and Amazon have said on the record. They come partly from X posts — @ChrisCamillo, @demian_ai, @amitisinvesting, @jukan05 (quoted by @ConnorJBates_), @ParadisLabs, @aleabitoreddit and others — and from TheBuildout's summaries of such posts. They are attributed, not independently verified.

Quick Snapshot (as of 2026-09-25)


The call	🟡 Wait for the September 21 gap to be retested. Entry $670–$690 · stop $665 · first target $778 · medium conviction.
What it is	The advertising business behind Facebook, Instagram, WhatsApp and Messenger, plus Reality Labs. Since September 8 it also makes Muse, an agent that runs on Meta's servers and completes tasks for users.
Why now	Muse passed 2.5 million downloads in about 13 days and reached No. 1 on Apple's free chart (CNBC). Amazon blocked it from shopping amazon.com and Shopify let it in, both in the week of September 21. At Connect on September 23 Meta showed the Muse Charm pendant, the agent's first dedicated hardware (CNBC).
Proof it is working	June-quarter revenue grew 28.0%, and trailing-twelve-month growth is 27.7% (SEC filings; Massive agrees). Gross margin has held between 81.4% and 82.1% for eight quarters. Muse is the first consumer product outsiders can see the spending produce.
The objection	Operating margin fell to 30.9% from 43.0% a year earlier. Meta bought back no stock in the first half of 2026 and borrowed $24.91 billion (SEC 10-Q). Accounts of the July 29 release say it also raised its 2026 capital-spending guide to $130–145 billion. A $16.7 billion state settlement (CNBC, August 26) falls after the June quarter.
Stock setup	20.0% above the 200-day at RSI 71.3, 3.61% below the 52-week high after Friday's reversal from the September 24 close. The 50-day is still under the 200-day because the month's rise outpaced both averages.

Action Matrix (as of 2026-09-25)

Signal	If	Then
🟢 Buy	A pullback holds $670–$690 on a daily close. The September 21 gap's top ($679.60), the $677.08 support shelf and the $682.49 value-area high all sit inside it	Entry. A gap that holds on its first retest is the flag in the bottom → pop → flag pattern, and it turns a sentiment re-rate into a base.
🟡 Wait	Anywhere from $690 up to $778, including Friday's $751.66 close	RSI 71.3 after a 30% month, a reversal from the high on above-average volume, and the October 28 print still ahead. There is no edge here.
🔴 Out	A daily close below $665	The gap has filled back to its $665.75 base and the market has sold the Muse re-rate. The setup is void.
🟠 Target	A daily close above $778, through the $777.59 September 24 close	Trim or trail. That close is the highest of the past year and cleared the prior $765 closing high from September 2025.
🔴 Thesis breaks	At the October 28 print, all three: operating margin below 30% excluding the August settlement charge, the capital-spending guide raised again, and no Muse usage figure or WhatsApp date	The build is then outrunning both the ad engine and the one product outsiders can measure.

The Story Right Now (as of 2026-09-25)

Meta is doing well: the Muse launch turned a flat year into the stock's best month since 2013 (Bloomberg, September 24), and its one-year return turned positive again this week. On September 15 the shares were down 12.4% over twelve months, the weakest one-year showing in their mega-cap group.

Data as of 2026-09-25 close: META $751.66, RSI 71.3, −3.33% on the day, +12.90% on the week, +30.46% on the month, +0.37% over one year, +13.7% vs SMA20, +20.0% vs SMA200, 3.61% below its 52-week high.

On the long windows the trend is repaired, not new. The one-year return trails the S&P 500 by 16.85 points. The two-year and three-year returns are +32.26% and +149.86%, per the desk's September 25 settled-close summaries. The one-year figure was −1.50% on September 23 and +0.37% on September 25 (the AI scans of those dates). The price sits well above the 200-day average, while the 50-day is still under the 200-day because the month's rise outpaced both.

The sequence has sources at each step:

Launch, September 8. Muse launched as an agent that runs on a computer on Meta's servers and can act inside a user's email and calendar (Semafor). It came with a free tier and $20 and $100 monthly plans (CNBC).
Adoption. It logged about 730,000 downloads in five days and 2.5 million in thirteen, and reached No. 1 on Apple's free chart (CNBC).
September 21. Wells Fargo raised its target to $796 from $640 without changing its revenue forecast (Investing.com), and the shares rose 11.34% on 2.6 times normal volume. Arm, Intel and AMD rose 17.16%, 12.16% and 9.95% that day. CNBC's explanation was that agents run on general-purpose processors (the September 21 market brief).
Connect, September 23. Meta showed the Muse Charm pendant and $1,299 VR Glasses (CNBC). On September 24 the stock closed at $777.59, its highest close of the past year.

Friday's reversal has no sourced cause. The shares fell 3.33% on 1.14 times normal volume while the S&P 500 fund rose 0.54%. Names exposed to agent traffic fell too: Cloudflare −2.73%, Fastly −6.45%, Shopify −2.00%. The only dated item is a Motley Fool headline, "Why Meta Platforms Stock Reversed Today," and its text is not in hand. The September 25 news check returned only quote pages, and no company event is dated to the session.

Two constraints now frame the story.

The first is commercial. Amazon blocked Muse from its retail site. According to Bloomberg and GeekWire, its grounds were that Meta gave no notice, the agent does not identify itself, and it appears to capture customers' credentials. Coverage reads the motive as advertising: an agent that picks the product skips the browsing that Amazon sells ads against (Forbes). Shopify let Muse in the same week (MarketWatch, Semafor).

The second is physical. Each Muse user gets a dedicated virtual machine (TheBuildout), so usage grows with processors and memory. Meta is reported to have delayed the WhatsApp rollout, its path to two billion users, for lack of CPUs, GPUs, DRAM and storage (TheBuildout, tagged as rumor). The filings show the cost of the build: $49.11 billion of property and equipment purchases in the first half, no share repurchases in 2026, and $24.91 billion of new debt (SEC 10-Q).

The call: wait, medium conviction. Entry is $670–$690 on a daily close that holds the September 21 gap, whose top ($679.60), the $677.08 support shelf and the $682.49 value-area high all sit inside the zone. The stop is $665, just under the gap's $665.75 base. The first target is $778, a close through the September 24 high.

The September 22 view set the same zone. The stock never closed nearer than 6.8% above it (September 22 market brief), and that view's $765 target was reached on September 24 without an entry. A 30% month at RSI 71 followed by a reversal on volume is not the flag this pattern buys; a retest that holds the gap is.

What changes the view: a daily close under $665 voids the setup. The October 28 print is scheduled to answer three questions:

whether Meta restates Muse usage with a number or dates the WhatsApp rollout;
whether operating margin holds 30% excluding the August settlement charge;
whether the capital-spending guide rises again.

A second large retailer following Amazon would cap Muse's commerce path. A disclosed paid-tier subscriber count would give the engagement story a revenue line.

Extended analysis

Why This Moment Is Different

For two years the open question on Meta was what more than $100 billion a year of capital spending would produce beyond better ad targeting. The only outside readout was operating margin, and it fell as planned. Muse is the first product outsiders can watch the spending produce.

The market treated it as a platform, not a feature, because it moved other companies' stocks:

Bloomberg reported banks, insurers and online travel agencies falling on September 22 on fears that agents erode "consumer inertia," people's habit of staying with a provider they no longer compare.
CNBC named a brokerage as traders' next target on September 23.
Airbnb fell 7.56% and Expedia 7.72% that day. The September 23 market brief treats the theme as context for the category, not as the cause of those two moves.

Meta's financing changed in the same period, and the filings show it more clearly than the headlines did:

The capex guide rose. The annual report guided 2026 capital expenditures to about $115–135 billion. Contemporaneous accounts of the July 29 second-quarter release put the guide at $130–145 billion (@ParadisLabs, @aleabitoreddit).
Buybacks stopped. Meta has bought no stock since the September 2025 quarter. Repurchases were $26.25 billion for both the first nine months and the full year of 2025, and both 2026 quarterly filings report zero.
Debt rose. Meta raised $24.91 billion of debt in the first half of 2026.
The dividend continues at $0.525 a quarter.

The build is now funded by operating cash, cash balances and borrowing, with nothing returned through buybacks.

The legal overhang cleared at a price. On August 26 Meta settled the state attorneys general's social-media addiction case for $16.7 billion (CNBC; other outlets round to $17–18 billion). The settlement adds limits for users under 18, including a two-hour daily cap on Instagram and Facebook (CNBC, August 28). It falls after the June quarter. The filings in hand do not show how much was already accrued, so the third-quarter report is where it will show.

Amazon and Muse

What Muse is

Muse is an agent. It takes a request and works through a browser and connected accounts to finish it — email, calendars, forms, travel, shopping — and it keeps working after the app closes (Motley Fool, September 21). Semafor called it "a computer sitting on Meta's servers that connects to and can control your email, your calendar, or your health information." Each user's agent runs on its own virtual machine (TheBuildout, from @RihardJarc, September 8). One analyst's sizing puts each account at 2 vCPUs, 8 GB of memory and 100 GB of disk, with the service US-only at thirteen days (@demian_ai).

Muse launched free, with $20 and $100 monthly tiers by usage (CNBC, September 8). According to coverage of the Amazon dispute, it asks the user's approval before buying anything. Commerce integrations named in the week of September 21 include Shopify, Instacart, $PYPL (PayPal) and $EXPE (Expedia), per @amitisinvesting; MarketWatch and Semafor separately confirm the Shopify arrangement.

The hardware follow-on is confirmed. At Connect on September 23, Zuckerberg showed the Muse Charm, a pendant built for the agent, alongside $1,299 VR Glasses (CNBC). CNBC's read the next day was that Meta reached the consumer AI-device market before OpenAI but that "Zuckerberg's strategy remains unproven." The Charm's price and ship date are not in the coverage in hand.

What Amazon did, and why

Amazon blocked Muse from shopping amazon.com. The Register, Forbes and Bloomberg reported the block on September 21, and CNBC on September 23 called the standoff one that "could be a sign of things to come."

Amazon's stated grounds, as Bloomberg and GeekWire reported them:

Meta did not notify Amazon.
The agent does not identify itself as automated.
It appears to capture and store customers' Amazon credentials.

The block took the form of a Conditions of Use warning. Coverage adds that Amazon has moved against shopping agents before, including a lawsuit over Perplexity's Comet browser.

The commercial motive is the one coverage leads with. Forbes headlined its September 23 piece "Amazon's $68 billion reason to block Meta's Muse." Its argument is that an agent choosing the product skips the search-and-browse session in which Amazon sells sponsored placements. The headline figure is Forbes's, and this piece does not verify it.

Amazon's own annual report supports the logic. It lists "virtual assistants, and other online and app-based means of discovering, using, or acquiring goods and services, either directly or in collaboration with other retailers, including through artificial intelligence" among its competitors. It describes its advertising as "sponsored ads, display, and video advertising." The stated reasons and the ad motive point the same way, since both protect the shopping session. The stated reasons are what a court would weigh; the ad motive is what investors should price.

What it means for Meta

Muse's commerce path now runs through merchants that opt in. Shopify's merchants, Instacart and Expedia give the agent inventory to act on, while the largest US store is closed to it.

For Meta's advertising business, which the annual report says produces "substantially all" revenue, the stakes run the other way. An agent that completes purchases puts Meta nearer the transaction than an ad impression does. None of this is monetized yet. The paid tiers are Muse's only revenue line, and Meta has disclosed no subscriber count.

What it means for Amazon

Amazon is on both sides:

Its store is defending the browsing session that its retail-media business sells.
Its cloud is reported to be a Meta supplier. On September 21 @ChrisCamillo wrote that Meta "already uses Amazon Bedrock at scale and has committed to tens of millions of AWS Graviton CPU cores for its agentic AI workloads." TheBuildout (September 15) argued that Muse's growth is an AWS capacity story.

Meta's annual report partly supports the supplier side without naming vendors. It says most of its $131.05 billion of non-cancelable commitments at year-end 2025 relate to "third-party cloud capacity arrangements" and to its own servers, data centers and hardware. Amazon's own shopping agents are not covered in the material behind this piece.

How the two stocks traded

Over the week ending September 25, Meta rose 12.90% and $AMZN fell 1.59% (the desk's settled-close summaries). Day by day, from the market briefs:

Date	Meta	Amazon
September 21 (the block's news day)	+11.34%	+1.87%, to $258.45
September 22	−0.6%	−1.3%
September 23	+1.02%	−2.24% on 1.29 times volume
September 24	not in hand	+0.04%
September 25	−3.33%	+0.12%

On September 23 the brief treated CNBC's standoff report as context, not a cause, for Amazon's decline. Amazon's drift predates Muse: the September 23 Amazon profile describes it stalling since its July 31 earnings pop.

On the long windows Amazon's trend is intact: +14.45% over one year, 3.7% above its 200-day average, 2.77 points behind the S&P 500. This is a shallow pullback, and the tape has not priced the block as damage to Amazon.

What is unresolved

Legal. The coverage in hand reports no lawsuit between Meta and Amazon. Amazon's Perplexity suit is the precedent.
A deal. None has been reported. Neither side has publicly proposed one possible resolution: an agent that identifies itself and uses sanctioned access.
Other retailers. Shopify is in, and Semafor (September 22) called the split "the first sign of battle lines." Whether other large retailers follow Amazon or Shopify is open.
Security. Hacker News surfaced an Ars Technica report titled "Meta's Muse has a serious 0-day." Meta's own annual report lists "prompt injection, errors, and other issues related to AI agents" among its cyber risks. Amazon's credential complaint falls on the same weak point.

The Power and Compute Under Muse

An agent is a processor-and-memory problem

A chatbot answers a prompt and stops. Muse holds a machine for each user and keeps working, which shifts load toward CPUs and memory. That is why CNBC called September 21 "a CPU kind of day," with the processor makers leading. @demian_ai put the arithmetic plainly: "If Meta actually leaves those boxes on, user growth turns into a chip and memory problem."

Meta is short of it now

WhatsApp rollout. TheBuildout (September 21, a report its source tagged as rumor, via @zephyr_z9) says Meta delayed Muse's WhatsApp launch for lack of CPUs, GPUs, DRAM and SSD storage to serve agents across WhatsApp's two billion users.
Strain at current scale. @jukan05, quoted by @ConnorJBates_ on September 24: "So Muse hasn't even reached a million users yet, and it's already struggling to serve them."
Purchasing. TheBuildout (September 10, also tagged as rumor) reports Meta as the most aggressive buyer of server CPUs since the third quarter of 2025 and of memory since the second quarter of 2026.
Meta's own filing names the dependency: its "ability to continue to develop and effectively deploy AI technologies is dependent on access to … processing hardware, network capacity, models, computing power, and related energy requirements, as to which we cannot control the availability or pricing."

Custom silicon: MTIA

Meta's in-house accelerator line, MTIA (Meta Training and Inference Accelerator), was built first for ad ranking and recommendation. TheBuildout (September 21, via @mindmoon_108) reports Muse has been optimized to run on it. TheBuildout (September 16, via @dnystedt) reports two new custom chips planned for 2027 with more than a gigawatt of custom-silicon capacity. @wallstengine (September 14) named the MTIA 450, code-named Arke, for the first half of 2027. The annual report does not name the chips.

Custom silicon lowers the cost of serving each user. It does not add capacity before 2027, and the WhatsApp gap is a 2026 problem.

Capital: the spend and how it is paid for

Spending and cash. Purchases of property and equipment were $49.11 billion in the first half of 2026, against $29.48 billion a year earlier. Operating cash flow was $64.09 billion, against $49.59 billion (SEC 10-Q). Against a guide of $130–145 billion, as reported, the second half has to run heavier than the first.
Commitments. Meta had $131.05 billion of non-cancelable commitments at year-end 2025 (SEC 10-K).
Louisiana campus venture. Meta holds 20% of a venture co-developing a Louisiana data-center campus, with about $27 billion of estimated development costs. Its leases there begin in 2029, with an initial commitment of about $12.31 billion, and it has given residual value guarantees with an aggregate threshold of about $28 billion (SEC 10-K).
Hyperion. CNBC (July 13) reported Meta's Hyperion campus in Richland Parish, Louisiana at 5 GW and more than $50 billion. The filing does not name the campus.
Funding. As covered above, the money comes from operating cash, cash balances and $24.91 billion of new debt, with buybacks stopped.

Power: what Meta has contracted, and what it has not

What the filing says. Meta's annual report describes "multi-year agreements ranging from three to 25 years to purchase clean and renewable energy that do not specify a fixed or minimum volume commitment." It names "related energy requirements" as a risk it cannot control.

Nuclear deals, as reported:

NuttyCLD (March 30) wrote that Meta announced a nuclear portfolio combining existing plants with next-generation reactors still in early development (attributed; the portfolio size is not verified here).
@amitisinvesting (August 25) reported 20-year power agreements across $VST (Vistra)'s nuclear fleet (attributed; the capacity figure is not verified here).
Vistra's Stacey Doré, via TheBuildout (September 17), cited a 433 MW nuclear uprate with Meta as a model for fast capacity additions. She also said chip supply supports roughly 60 GW of US data-center growth by 2030, and that grid interconnection, not generation, is the bottleneck.

Uprates and power-purchase agreements deliver grid power where the grid can reach. That is an inference about timing; none of these deals is described as closing a 2026 capacity gap. How Hyperion will be powered is not in the material in hand.

Where Bloom Energy fits

No direct relationship found. Bloom Energy's annual report, its news receipt and the desk's captured coverage name no Meta contract, and Meta's filings name no fuel-cell supplier.

The link is industry-level. Bloom's annual report says "Grid Constraints and Permitting Delays are Extending Time to Power" and "Shift Toward Onsite Power Generation is Occurring." It lists hyperscalers, colocation operators and neoclouds among its data-center channels. According to the September 23 Bloom profile, management says all major US hyperscalers use its on-site power; that statement comes from a summary of its second-quarter call.

This week's example of the constraint involved Oracle, not Meta. Oracle sent a force-majeure notice on its New Mexico Project Jupiter campus (CNBC, September 24) after local officials delayed permits needed to power it (Semafor, September 25). Morgan Stanley said the notice puts data-center loan and lease documents under scrutiny (Bloomberg, September 25). NuttyCLD's May 1 framing fits: the market moved from asking "how many GW announced" to "who can power the GPUs now."

The tape has paid for speed-to-power this year. $BE is up 320.29% over one year and rose 8.27% on Friday. Vistra is down 31.33% and $CEG (Constellation) 19.32%, and both nuclear operators sit below their 200-day averages (the desk's September 25 summaries). This piece does not attribute those moves.

For Meta, the direct link would be an on-site generation contract, and none has been disclosed.

How the Business Makes Money, and What It Says It Will Do

Meta reports two segments:

Family of Apps (Facebook, Instagram, WhatsApp, Messenger, Threads) generates "substantially all of our revenue from selling advertising placements," per the annual report.
Reality Labs reduced 2025 operating profit by about $19.19 billion, and Meta expects 2026 losses there "to remain similar."

The ad engine grew on both volume and price. Daily active people across the family averaged 3.58 billion in December 2025, up 7%. Ad impressions rose 12% in 2025 and average price per ad 9%. Headcount was 78,865, up 6%.

The strategy is stated in the filing's own words. Meta is building AI "to advance our vision to deliver personal superintelligence for everyone." Muse is the first consumer product to carry that line.

The filed risks that bear on this read:

competition for engagement;
the EU regulatory stack and the FTC consent order;
founder voting control through Class B shares;
new language on the cyber risks of AI agents;
infrastructure and energy the company "cannot control."

What the Street and the Public Conversation Say

Sell side:

Wells Fargo's September 21 note lifted its target to $796 from $640 on an unchanged revenue forecast, raising the multiple on 2027 earnings from 20 to 25 times (Investing.com).
Yahoo Finance carried "Bank of America backs Meta stock after Muse surprise" on September 23; the note's details are not in hand.
JPMorgan's view that Muse "could become the most-used AI app since ChatGPT" reached the desk only through @wallstengine (September 22).
Yahoo's September 21 headline, "Meta's Muse Agent Has Wall Street Divided," has no text in hand.

Usage claims beyond CNBC's download counts are social and unaudited. @ConnorJBates_ charted Muse at 642,000 US daily users against ChatGPT's 231,000 at launch, and US daily Muse downloads above every Meta core app (September 24); the charts' data source is not stated.

The most useful skeptics:

@ChrisCamillo: "Too early to know if Muse adoption fades or has staying power. But one thing is already obvious: distribution is an unfair advantage."
@RichTerry123: the free tier includes 100 million tokens a week, heavy users cost $20–40, and "how Meta recoups that remains an open question."
@jukan05, on capacity (above).

Hacker News surfaced Meta's Muse page (666 points), a post titled "I asked Meta's Muse for its filesystem and it sent…" (352), Ars Technica's "serious 0-day" report (122) and a post saying Muse "appears to use an OpenAI model" (58). Only the headlines are in hand. Security and model provenance are open questions, not findings. Today's Amazon-and-Muse social sweep returned nothing usable.

Insiders sold into the run (SEC Form 4s):

Chief product officer Chris Cox sold 20,000 shares on September 9 (about $13.01 million), 20,000 on September 15 (about $13.50 million) and 40,000 on September 21 (about $28.51 million).
Chief operating officer Javier Olivan sold 1,575 shares on September 21 (about $1.07 million).
An August 12–20 cluster covered five filings by the chief technology, financial, legal and operating officers.

Equal lots on a schedule usually indicate pre-arranged plans, which carry no view. It is also true that senior officers sold into the month.

Short interest was 30.99 million shares at the September 15 settlement, 1.57 days to cover (Massive).

The Conviction Chain

Claim 1 — the ad engine is compounding. Revenue growth accelerated from 16.1% in the March 2025 quarter to 33.1% a year later and was 28.0% in June, with gross margin between 81.4% and 82.1% throughout.

Confirms it: third-quarter revenue inside the $61–64 billion guide with growth above 20%.
Breaks it: growth under 20% while the capital-spending guide rises.

Claim 2 — Muse is a real consumer product.

Evidence for: downloads, the No. 1 ranking, paid tiers at launch, merchant integrations and dedicated hardware.
Evidence against: no company usage figure since launch beyond the 60% rise in daily assistant users the Motley Fool reported on September 21; retention is unknown; the security headlines; Amazon's block.
Confirms it: a usage, retention or subscriber number on October 28.
Breaks it: no figure, and the download ranking fades.

Claim 3 — capacity arrives in time.

Evidence for: first-half capital spending of $49.11 billion, the MTIA roadmap and $131.05 billion of commitments.
Evidence against: the reported WhatsApp delay, the strain at under a million users, custom chips arriving in 2027, and power contracts that are long-dated.
Confirms it: a dated WhatsApp rollout.
Breaks it: the rollout slips past year-end while the spending guide rises.

Claim 4 — the build is financeable.

Evidence for: first-half operating cash flow of $64.09 billion; $15.46 billion of cash and $74.80 billion of marketable securities at June 30.
Evidence against: zero buybacks, $24.91 billion of new debt, the Louisiana guarantees and a $16.7 billion settlement.
Confirms it: 2026 spending lands inside $130–145 billion without further borrowing.
Breaks it: a guide above $145 billion alongside more debt.

This claim's September 15 break condition — "a guide raise … accompanied by a buyback cut" — had already been met in the July 30 10-Q, and the earlier views missed it.

The weakest link is Claim 3. Muse's value depends on reaching WhatsApp scale, and the reported constraint is physical: chips, memory and power cannot be bought instantly at any price. That keeps conviction at medium.

Financial Proof

Quarterly series from the filed statements, with growth and margins computed by script from the same rows. Revenue growth for the 2024 and early-2025 quarters uses SEC XBRL prior-year quarters; the December 2024 quarter is derived as full year minus nine months.

Quarter	Period end	Revenue	Rev YoY	Op income	Op margin	Gross margin	Net income	Diluted EPS	R&D % rev	Source
2024 Q3	2024-09-30	$40.59B	+18.9%	$17.35B	42.7%	81.8%	$15.69B	$6.03	27.5%	SEC 10-Q / Massive financials
2024 Q4	2024-12-31	$48.38B	+20.6%	$23.36B	48.3%	81.7%	$20.84B	$7.98	25.2%	SEC 10-K / Massive financials
2025 Q1	2025-03-31	$42.31B	+16.1%	$17.55B	41.5%	82.1%	$16.64B	$6.43	28.7%	SEC 10-Q / Massive financials
2025 Q2	2025-06-30	$47.52B	+21.6%	$20.44B	43.0%	82.1%	$18.34B	$7.14	27.2%	SEC 10-Q / Massive financials
2025 Q3	2025-09-30	$51.24B	+26.2%	$20.54B	40.1%	82.0%	$2.71B	$1.05	29.6%	SEC 10-Q / Massive financials
2025 Q4	2025-12-31	$59.89B	+23.8%	$24.75B	41.3%	81.8%	$22.77B	$8.87	28.6%	SEC 10-K / Massive financials
2026 Q1	2026-03-31	$56.31B	+33.1%	$22.87B	40.6%	81.9%	$26.77B	$10.44	31.4%	SEC 10-Q / Massive financials
2026 Q2	2026-06-30	$60.80B	+28.0%	$18.77B	30.9%	81.4%	$15.85B	$6.18	35.6%	SEC 10-Q / Massive financials

Two quarters' bottom lines are distorted by tax:

September 2025: a one-time $15.93 billion charge on enactment of the One Big Beautiful Bill Act, $14.03 billion of it a valuation allowance (SEC 10-K). Operating income, which the charge does not touch, was $20.54 billion.
March 2026: a $5.02 billion tax benefit (SEC 10-Q), which is why net income exceeded operating income.

The June 2026 quarter's operating line carries about $2.4 billion of legal charges (Q2 call coverage). Research and development rose to 35.6% of revenue from 27.2% a year earlier. The compression is entirely below gross margin.

Cash flow and balance sheet, SEC filings:

Item	Period	Value	Comparison
Operating cash flow	H1 2026	$64.09B	$49.59B in H1 2025
Purchases of property and equipment	H1 2026	$49.11B	$29.48B in H1 2025
Share repurchases	H1 2026	$0	$22.92B in H1 2025; $26.25B for 2025
Dividends paid	H1 2026	$2.70B	$0.525 per share per quarter
Debt issued	H1 2026	$24.91B	$29.91B for all of 2025
Long-term debt	June 30, 2026	$83.66B	$58.74B at December 31, 2025
Cash and equivalents / marketable securities	June 30, 2026	$15.46B / $74.80B	$35.87B / $45.72B at December 31, 2025
Operating cash flow	FY2025	$115.80B	SEC annual figure, single source
Capital expenditure	FY2025	$69.69B	SEC annual figure, single source
Free cash flow	FY2025	$46.11B	SEC annual figure, single source
Unused repurchase authorization	December 31, 2025	$25.03B	—

Trailing-twelve-month revenue growth through June is 27.7% (SEC XBRL and Massive agree). The full-year 2025 revenue headline is omitted here because the two sources compare it on mismatched bases.

Price Data (live)

Live values from the desk's settled-close summaries. They update with each close; the dated sections above keep the numbers of their own date.

Symbol	As of	Price ($)	RSI	1D (%)	7D (%)	30D (%)	1Y (%)	1Y vs S&P 500 (pts)	vs 200-day (%)
$META	2026-10-16	709.33	50.2	-3.72	-1.3	5.35	-0.38	-15.92	12.2
$AMZN	2026-10-16	250	45	-1.87	-4.74	1.64	16.57	1.03	3
$SHOP	2026-10-16	169.52	61	-4.84	-0.78	30.44	8.27	-7.26	30.7
$ARM	2026-10-16	227.8	34.5	-6.94	-14.45	-6.63	33.07	17.54	1.9
$NET	2026-10-16	296.3	35.9	-6.33	-17.91	-8.73	40.16	24.63	22.3
$BE	2026-10-16	221.31	36.5	-6.34	-21.1	-18.04	92.29	76.76	2.3
$VST	2026-10-16	152.02	52.4	-1.39	-5.86	8.28	-27.75	-43.28	-1.7
$CEG	2026-10-16	271.37	47.1	-1.32	-8.96	4.56	-31.56	-47.1	-4.5
$SPY	2026-10-16	763.25	43.8	-1.11	-1.97	1.22	15.53	0	5.3

Catalysts

Third-quarter results, scheduled for October 28 after the close. Revenue is guided to $61–64 billion. The questions that matter:
- operating margin excluding the settlement;
- how the $16.7 billion settlement is booked;
- any change to the $130–145 billion spending guide;
- any Muse usage, retention or subscriber figure;
- a WhatsApp date.
Amazon's third-quarter report, scheduled for October 29 after the close. This is Amazon's first earnings chance to address shopping agents and the Muse block.
The WhatsApp rollout. The reported delay turns into a date or it does not.
Muse Charm pricing and shipping. Neither is disclosed yet.
Resolution with Amazon, and other retailers' choices. Any lawsuit, any access agreement, and any large retailer choosing Amazon's side or Shopify's.
Custom silicon in 2027. Any slip pushes cost relief past the depreciation ramp.
Power contracts. A disclosed on-site generation deal would be the first direct link between Muse's compute needs and the fast-power suppliers.

Risks and Kill Tests

Setup fails. Kill test: a daily close below $665.
Long trend fails. Kill test: a daily close back below the 200-day moving average. That average is the level to watch, and the September 22 view's July-era structural stop at $563 remains the deeper line.
Spending does not convert. Kill test: operating margin below 30% at the October 28 print, excluding the settlement charge.
The ad engine slows. Kill test: revenue growth under 20% with the spending guide unchanged or higher.
Muse fades. Kill test: no usage figure at the print, and the download ranking gone.
Commerce is fenced off. Kill test: a second major retailer blocks the agent.
Security. Kill test: a confirmed credential or data incident involving Muse. That would validate Amazon's stated grounds and invite regulators in.
Governance. Class B voting control rests with the founder and chief executive, and outside holders have no way to slow the capital program.

Refresh History

Date	Price	RSI	vs SMA20	30D	What changed
2026-09-22	$736.60	76.0	+16.4%	+34.0%	Muse launched September 8, and Wells Fargo raised its target to $796 from $640 on an unchanged revenue forecast, gapping the shares on September 21. Entry $670–$690, stop $665, first target $765. The target was reached on September 24 (close $777.59) without the zone ever trading; the nearest close was 6.8% above it. Corrected in this view: that view quoted the January capital-spending guide ($115–135 billion) where accounts of the July release put it at $130–145 billion; it said no paid Muse tier had been announced, but CNBC reported $20 and $100 tiers at launch; and it did not note that buybacks were zero in the first half of 2026.
2026-09-15	$670.24	71.6	+7.5% vs 200-day	+13.6%	Third attempt at reclaiming the moving averages, and the first to hold two closes above $665. Entry $665–$680, stop $563, target $780. Operating margin 30.9% against 43.0% a year earlier. The insider panel showed zero buys against 100 sells.

Cross-references

Supersedes the Meta deep dive of September 15, 2026 (refreshed September 22). Its numbers are history.
Earlier desk call 2026-07-09-meta-second-try-at-the-turn: watch above $665, invalidation under $563. History.
Companion work on the connected story: the September 23 profiles of Amazon, Bloom Energy, Cloudflare and Palantir, and this set's companion dives on Bloom Energy, Cloudflare and Palantir.
- The Cloudflare and Fastly angle, how agents reach websites through edge networks, belongs there. @citrini's September 23 tests of Muse traffic through Fastly are unverified, and no direct Cloudflare–Muse relationship was found here.
- No direct Palantir–Muse relationship was found.
Tracked on the Mag 7, FAANG, Tech, AI Scan, Bargain Bin and Insider lists, and on three AI build-out lists: data-center operators, custom silicon, and nuclear energy for data centers.
Momentum Board / washout-turn: not in any Momentum Board tier on the September 25 board, and not on the washout-turn board.
Campaigns: the desk's AI-power and AI-spending campaigns were last reviewed September 6–11, before the re-rate. They are context, not a lens for this read.

Sources

Primary filings

Meta Platforms Form 10-K for fiscal 2025, filed 2026-01-29 — segments, Family of Apps metrics, the 2026 capex guide of $115–135 billion, the $131.05 billion of commitments, the Louisiana venture and guarantees, energy agreements, the tax charge, the AI-agent and energy risk language, and headcount.
Meta Form 10-Q for the quarter ended 2026-06-30, filed 2026-07-30 — first-half cash flow, zero repurchases, debt issuance, long-term debt, cash and securities.
Meta Reports Second Quarter 2026 Results, 2026-07-29 — the release whose capex guide was not read directly; see the attributed accounts below.
Meta Q2 2026 earnings call coverage, AlphaStreet — the $61–64 billion third-quarter guide and about $2.4 billion of legal charges.
Form 4s: Cox, 2026-09-21; Olivan, 2026-09-21; Cox, 2026-09-15; Cox, 2026-09-09.
Amazon.com Form 10-K for fiscal 2025, filed 2026-02-06 — competition language on virtual assistants and AI shopping; advertising services.

Muse, Connect and the re-rate

Amazon's block

Settlement, infrastructure and power

Attributed, not independently verified (X and TheBuildout)

Desk receipts

mag7, tech.json, ai-scan.json, war-sensitivity.json, focus.json, insider-universe.json, holdings-etfs.json (SPY row only), thebuildout-semiconductor-ip-licensors.json, web-cloud.json — the 2026-09-25 settled close for all tape values
2026-09-21-market-brief, 2026-09-22-market-brief.md, 2026-09-23-market-brief.md, 2026-09-24-market-brief.md — dated daily moves and closes
2026-09-23-ai-scan, 2026-09-25-ai-scan, 2026-09-25-monster-scan, 2026-09-25-washout-scan
income-statements, cash-flow-statements.json, balance-sheets.json, short-interest.json, 10k-business.json, 10k-risk_factors.json
company-facts, key-financials.json, 10K-2026-01-29.html
latest (fetched 2026-09-25T20:44Z, ok, not stale; quote pages only); 2026-09-25
10k-business; 10k-business, 10k-risk_factors.json, latest
2026-10-28-meta-earnings, 2026-10-29-amzn-earnings, 2026-07-29-meta-earnings
filing-watch — Form 4 rows
2026-09-08-*, 2026-09-09-semafor-meta-launches-muse-*, 2026-09-10-thebuildout-meta-aggressively-*, 2026-09-15-thebuildout-meta-s-muse-agent-*, 2026-09-16-thebuildout-meta-plans-*, 2026-09-17-thebuildout-vistra-*, 2026-09-19-thebuildout-oracle-project-jupiter-*, 2026-09-21-* (Muse, Amazon/Shopify, TheBuildout), 2026-09-22-*, 2026-09-23-*, 2026-09-24-*, 2026-09-25-* (Meta, Oracle), 2026-08-26/28-* (settlement), 2026-07-13-cnbc-meta-s-louisiana-*, 2026-03-30-nuttycld-*, 2026-05-01-nuttycld-*
2026-09-25-social-search-meta-muse, 2026-09-25-social-search-amzn-amazon-muse.md, and the X captures listed above
2026-09-23-AMZN, 2026-09-23-BE; 2026-09-15-meta-deep-dive; CALL-LESSONS