Meta’s Debt-Funded Dividend Raises New Capital Allocation Risks

Meta traded $593.40 in the pre-market Thursday, up $4.63 or 0.79%, after closing Wednesday at $588.77 with a gain of $0.83, or 0.14%, and then adding another $2.65 to $591.42 in after-hours. Wednesday's session opened $600.46, ran to $603.60, and reversed to a low of $580.12 before closing near the middle of that $23.48 range on 14.9 million shares against a 19.58 million average.

That $580.12 low matters more than the close. Applying Fibonacci retracement to the swing between the 52-week low at $520.26 and the 52-week high at $796.25 — a $275.99 range — puts the 78.6% retracement at $579.32. Wednesday's low undercut it by 80 cents and reversed. Buyers defended the deepest retracement level on the chart on the second-lowest volume session of the week.

The recovery from the earnings wreck has been sharp and it has stalled. The stock closed $585.61 on July 29 before the print, fell 9.64% after hours to $529.15, and then dropped nearly 8% in Wednesday's July 30 regular session. Monday August 3 delivered a 6.22% rip. The stock has now recovered from $529.15 to $593.40, a gain of $64.25 or 12.14%, and it is still 25% below the August 2025 all-time closing high of $787.42 and 13% below the July 15 near-term high.

The 52-week arithmetic frames the damage. From $796.25 to $593.40 is a decline of $202.85, or 25.5%. From the $520.26 low, the stock is up $73.14, or 14.06%. That leaves META sitting in the lower third of its own annual range with a $1.50 trillion market capitalization, 2.21 billion shares outstanding, and a beta of 1.25.

Pre-earnings prediction pricing had framed the range tightly: 79.5% probability of closing above $580, 73.5% above $590, and only 10.5% above $600. That made $600 a hard resistance shelf and $580 the support floor. Wednesday's session tested both — $603.60 on the high, $580.12 on the low — and settled between them. Options positioning into the print skewed bullish with a 2.44-to-1 call-to-put open interest ratio and 67,113 contracts of earnings-night volume.

Eleven consecutive down sessions preceded this. That streak erased $17.8 billion from the founder's paper wealth. The stock is trying to build a base at $580, and $580 is the line.

The Free Cash Flow Collapse to $784 Million Is the Story

Every argument about this stock reduces to one number. Free cash flow in the second quarter came in at $784 million against $8.5 billion in the same quarter a year earlier — a decline of 90.8%. That is the figure that produced a 9.64% after-hours drop on a quarter where revenue beat.

The mechanics are brutal in their simplicity. Operating cash flow was $31.86 billion, a healthy figure for a company with $60.8 billion of quarterly revenue. Capital expenditures were $31.08 billion. Capex consumed 97.5% of operating cash flow, leaving $784 million. Dividends paid in the quarter ran $1.353 billion — meaning the payout alone exceeded the entire quarter's free cash flow by $569 million.

That is not a company generating surplus capital. That is a company funding its dividend from the balance sheet while it builds. And the market repriced it accordingly, because the entire investment case for this business over the last decade rested on the combination of 40%-plus operating margins, minimal capital intensity, and a free cash flow machine that funded $26.26 billion of buybacks in fiscal 2025.

The historical comparison is what makes the print so jarring. A year earlier the same company converted operating cash into $8.5 billion of free cash flow in a single quarter. Trailing twelve-month EBITDA still runs $109.66 billion at a 50.70% margin, so the operating business is not the problem. The problem is that every incremental dollar of that EBITDA is being routed into data centers before it reaches the shareholder.

The full-year picture is worse than the quarter. Capex guidance was raised to $130-$145 billion, and the quarterly run rate of $31.08 billion annualizes to $124 billion — meaning the back half carries an even heavier spend. Against trailing EBITDA of $109.66 billion, guided capex at the $137.5 billion midpoint exceeds the entire cash operating profit of the business.

For the forecast, this is the constraint that caps the multiple regardless of what advertising does. A stock trading at 22.2 times trailing earnings on 28% revenue growth would normally be cheap. It is not cheap when the cash those earnings represent never reaches the owner. The re-rating requires either capex to plateau or the spend to show up as identifiable revenue, and neither happens before the October 28 report.

$31.1 Billion of Quarterly Capex and Two Guidance Raises

The capex escalation is the single most important sequence in this story, and it has happened three times in nine months. The original 2026 framework sat under $100 billion. That became $115-$135 billion. In the first quarter it was lifted to $125-$145 billion. With the second-quarter print it moved to $130-$145 billion, raising the floor by another $5 billion.

Each raise has cost the stock money on the day. Management beat on both revenue and earnings in the first quarter and the stock still fell 6% to 7% after hours purely on the capex line. A beat alone has not moved this stock higher in either of its last two reports, which tells you exactly what the marginal buyer is pricing.

The quarterly number itself was $31.08 billion — a figure that on its own would rank among the largest capital programs in corporate history if annualized, and it sits inside a guided range whose midpoint of $137.5 billion is 125% of trailing EBITDA. Management has not provided standalone quarterly capex guidance, only the full-year band, which is precisely where the uncertainty concentrates.

The financing structure has changed to accommodate it. The company added $24.9 billion of debt in the quarter, taking long-term borrowings to $83.7 billion. A business that returned capital for a decade has become one that raises it. That transition alters the equity risk profile: leverage amplifies the downside if the AI investment cycle produces less than promised, and it dilutes the balance-sheet cushion that historically justified a premium multiple.

The strategic argument offered on the call was flexibility rather than certainty. A substantial share of compute goes toward training frontier models, framed as necessary investment to remain a leading lab. Remaining nimble is intended to fund the build-out more efficiently while preserving strategic flexibility to have compute when needed and creating multiple pathways to generate returns on invested capital. Balance sheet strength was cited as the mechanism for attracting capital from a wide range of markets to supplement operating cash flow.

That approach has now been operationalised through structured partnerships. A $14 billion data center joint venture centred on El Paso was announced alongside the results — an off-balance-sheet route to capacity that keeps a portion of the spend outside consolidated capex.

The question nobody has answered: which of these compute investments scales first, in 2026 or 2027, with quantifiable return on invested capital that a shareholder can actually see.

The Ad Engine Did Not Crack — It Accelerated to $59.4 Billion

Almost every headline merged two unrelated stories. The advertising business did not deteriorate. It accelerated.

Total revenue came in at $60.801 billion, up 28% year over year and 27% in constant currency, against a consensus that sat between $60.17 billion and $60.29 billion. Advertising revenue was $59.4 billion, up 27%. The composition is what matters: ad impressions rose 14% and the average price per ad rose 12%, both against the prior-year quarter. Pricing power did not crack — the auction is still clearing 12% higher on 14% more inventory.

That combination is unusual and it is the strongest single datapoint in the release. Most ad platforms scaling impressions at 14% see pricing compress as supply expands. Delivering simultaneous volume and price expansion means the AI-driven targeting and creative tooling built on the in-house model family is measurably improving advertiser outcomes, because advertisers do not pay 12% more per unit for the same result.

The smaller line was arguably more significant. Family of Apps other revenue crossed $1 billion in a quarter for the first time and grew 73% year over year. That is the bucket where messaging monetisation, paid subscriptions and business tooling sit — the only place where non-advertising revenue can eventually scale enough to matter against a $240 billion annualised base. A 73% growth rate off a $1 billion quarterly base is not yet material to the model, and it is the first credible non-ad revenue line this company has produced.

Set that against the sector. Alphabet fell 4.03% Wednesday. Snap dropped 7.94%. Pinterest lost 8.68%. Reddit fell 2.94%. Meta closed up 0.14%. On a session where the entire digital advertising complex was taken apart, the largest player in it finished green — which is the market distinguishing between an advertising problem and a capital allocation problem.

Revenue growth of 28% with impressions up 14% and pricing up 12% is not a business in trouble. Total revenue has increased 22.17% year over year and 7.97% sequentially. Net income declined 3.05% year over year and 40.81% sequentially, and earnings per share held flat year over year while falling 40.85% from the prior quarter.

Two different companies inside one press release. The market priced the worse one.

$3.58 Billion of Charges Made the Miss Look Worse Than It Was

Diluted earnings per share came in at $6.18 against a consensus between $7.17 and $7.40 — a shortfall of 13.8% to 16.5% depending on the estimate used. That headline drove the after-hours collapse. The composition of the miss deserves more scrutiny than it received.

Net income fell 14% year over year to $15.8 billion, weighed down by $2.4 billion in legal-proceedings charges and $1.18 billion of severance costs — $3.58 billion of one-time items in a single quarter. Add a higher effective tax rate, which management raised for the remainder of 2026, and the gap between the reported $6.18 and the expected $7.22 is largely explained by items that do not recur.

Run the arithmetic. On 2.21 billion diluted shares, $3.58 billion of charges represents roughly $1.62 per share pre-tax. Adjusting for those items alone would have put earnings above the consensus rather than 14% below it. The tax rate increase accounts for a further portion of the shortfall.

That is the strongest bull argument available on the earnings line, and it comes with a caveat: legal charges at this company are not obviously non-recurring. Regulatory exposure runs across multiple jurisdictions simultaneously. The Kids Online Safety Act cleared the Senate commerce committee and may reach a full Senate vote, carrying bipartisan support. In India, a case has been registered against the head of the local unit in Hyderabad over user posts, and the chief global affairs officer has apologised to the information technology minister for briefly restricting a post by the prime minister. Youth-safety and European litigation risk were flagged as ongoing concerns.

Severance costs are similarly structural rather than exceptional. A company reallocating headcount from legacy functions into AI infrastructure will book severance repeatedly through this cycle. Announced layoffs across the technology sector fell 46% year over year in July, and the cuts that did land remained concentrated in technology — where AI investment is reshaping org charts rather than expanding them.

Sequential net income tells the cleanest story: $15.85 billion this quarter against $26.77 billion the quarter before, a decline of 40.8%. Even adding back $3.58 billion of charges leaves a substantial sequential deterioration that traces directly to depreciation on the capex build and higher operating expense.

Trailing twelve-month earnings still run $26.56 per share. The earnings base is intact. The trajectory is not.

The Buyback Went to Zero and Debt Went to $83.7 Billion

The capital return story ended this quarter and the market has not fully priced what that means. Share repurchases were zero in the second quarter and zero in the first. Fiscal 2025 buybacks totalled $26.26 billion.

That is a $26 billion annual bid removed from the stock. On 2.21 billion shares at $593.40, $26.26 billion of repurchase represents roughly 44 million shares, or 2% of the float, bought every year by the company itself. Removing that flow changes the supply-demand balance in the shares independent of any fundamental development, and it is a meaningful part of why an 11-session losing streak was able to develop without a natural buyer stepping in.

The dividend survived but became a balance-sheet item rather than a cash-flow item. The quarterly payout of $0.52 per share cost $1.353 billion against $784 million of free cash flow. The yield is 0.36%. Ex-dividend date was June 15. A dividend funded from borrowings rather than surplus cash is a different instrument than a dividend funded from operations, and the sustainability question becomes live if capex runs at the $145 billion top of the guided range.

Long-term debt reached $83.7 billion after $24.9 billion was added in the quarter. Against a $1.50 trillion market capitalization that is modest leverage in isolation. Against $109.66 billion of trailing EBITDA it is under one turn. The concern is trajectory rather than level: three capex raises in nine months and $24.9 billion of new borrowing in a single quarter describe a financing program that scales with the build, and the build has no announced end date.

The structured route is now part of the plan. Balance sheet strength was explicitly cited as the mechanism for attracting capital across a wide range of markets to supplement operating cash flow, with the $14 billion El Paso joint venture presented as a template. Partnerships of that structure keep capacity off the consolidated capex line while committing the company to future payments — which is leverage by another name and harder for shareholders to measure.

The forecast implication is direct. Without a buyback, this stock needs incremental external buyers to move higher. Institutional flow has been mixed: one large manager boosted its stake, another trimmed, and a third added 108,541 shares. That is not accumulation. That is churn.

Q3 Guidance at $61-64 Billion With a Midpoint Below Consensus

Third-quarter revenue was guided to $61 billion to $64 billion, with the guidance assuming an approximately 1% foreign currency headwind to year-over-year growth at prevailing exchange rates. The $62.5 billion midpoint sits $700 million below the $63.2 billion consensus — a 1.1% shortfall.

That guide implies deceleration. Second-quarter revenue grew 28% year over year. The $62.5 billion midpoint against the prior-year comparable implies growth in the low-to-mid 20s. Current Street modelling for the September quarter sits at $63.25 billion of revenue and $6.76 per share of earnings, which places consensus above the guidance midpoint and near the top half of the guided band — a configuration that sets up another miss unless the quarter runs hot.

The forward earnings picture is where the tension lives. Consensus at $6.76 for the third quarter against $6.18 delivered in the second implies a 9.4% sequential recovery in earnings, which requires the $3.58 billion of one-time charges to genuinely not repeat and the higher effective tax rate to be fully absorbed. Underlying earnings growth for the full year was already expected to land close to flat year over year, around 1%, against revenue growth near 27%. That gap between top line and bottom line is the defining feature of this fiscal year.

Depreciation is the mechanical reason. A company spending $130-$145 billion on infrastructure books that spend into the income statement over the following years. The capex accelerating today becomes the earnings drag of 2027 and 2028 regardless of what revenue does, and the market is discounting that forward.

The next print lands October 28, 2026, after the close. Between now and then there is no scheduled catalyst on the fundamental side — no capex update, no revenue disclosure, no segment detail. What fills the gap is product news, regulatory headlines and macro flow, which is precisely the environment that produced an 11-session losing streak followed by a 6.22% single-day rip.

For the forecast that means the stock trades on sentiment and positioning for eleven weeks. The fundamental reset is done; the guidance is on the table; the capex range is public. Nothing changes the numbers until late October, which makes the technical levels the operative framework rather than the model.

Muse Code Is the First Real Answer to the ROI Question

Wednesday delivered the closest thing yet to a monetisation answer. The company launched Muse Code, its first AI coding agent, a terminal-based tool built to handle complete engineering tasks across large repositories, paired with an updated proprietary coding model, Muse Spark 1.2.

The pricing is the strategy. Pay-as-you-go access runs $1.25 per million input tokens and $4.25 per million output tokens, matching the prior model release. A contributor tier charges $0.30 per million total tokens — more than ten times cheaper than pay-as-you-go, a discount exceeding 90% — in exchange for permission to use submitted code to train future models. Enterprise customers can opt into zero data retention. The differentiation is explicitly price rather than capability.

That is a deliberate land grab in a market where two competitors already hold the developer mindshare. Undercutting on price by 90% while harvesting training data is the same playbook that built the advertising business: subsidise access, accumulate proprietary data, monetise the resulting model quality later.

Internal adoption is already producing measurable results. Thousands of engineers have been asked to use the internal version weekly and submit at least one code change through it. Roughly 7,000 weekly active users have generated more than 800 fixes that improved performance on a standard software-engineering benchmark, with corrections feeding training for an upcoming model. Badges on internal profiles reward contribution volume.

The strategic cost is credibility with the open-source developer base. This is a fully proprietary, cloud-only product from the company that spent three years arguing frontier weights should be free to download, self-host and fine-tune. That family had been downloaded roughly 1.2 billion times by early 2026, averaging about a million downloads daily. The pivot to closed and paid is a reversal.

For the equity, Muse Code matters because it is the first product where the $130-$145 billion of compute spend converts into a priced, sellable unit rather than an inferred improvement in ad targeting. Revenue at $0.30 to $4.25 per million tokens will not move a $240 billion top line this year. What it can move is the narrative — a company with a token-metered revenue line has an ROI story that a company with only an ad engine does not.

Separately, the underlying model reportedly compromised another company's systems during a cybersecurity test, and the company's AI assistant posted the largest jump in U.S. user share among its peers.