Meta Signed $825 Billion of Spending It Can Never Take Back

By Daniel Rao

Photo: Meta's data center in Luleå, Sweden, the model for the concrete-and-steel footprint Meta is now locked into worldwide. (Christopher Down / Wikimedia Commons, CC BY 4.0 — file photo)

The party lasted about four hours. Meta reported a thumping third quarter on Wednesday evening — $64.6 billion in revenue, up 26 percent, beating the Street — and the stock jumped five percent after hours to $748.80. Then somebody read the filing. By Thursday morning the shares were down six percent, roughly $100 billion of market value had evaporated overnight (ZeroHedge's arithmetic, and nobody is disputing it), and the cost of insuring Meta's debt against default had blown out to a record high. The reason was buried in a routine 10-Q: Meta has now signed about $825 billion of future spending commitments. Most of it, it turns out, cannot be cancelled.

This is the bill arriving. For two years the AI buildout has been a story about ambition — gigawatts, GPUs, frontier models. The 10-Q turned it into a story about obligation. And the market's reaction told you everything about how that story lands: Wall Street can price ambition. It is terrified of a commitment it cannot walk away from.

The $825 billion, in plain English

Sarah Frier's Bloomberg reporting laid the filing bare. The total breaks into two buckets. First, $355.7 billion in operating and finance leases that do not yet appear on Meta's balance sheet — data centers and energy infrastructure, with leases starting during the rest of this year and running through 2036. Second, $454.5 billion in non-cancellable short- and long-term contracts as of September 30, covering third-party cloud computing power, energy agreements for data centers, and consumer hardware for the Reality Labs division.

Read that second number again: non-cancellable. $454.5 billion of deals Meta has signed and cannot tear up. Then there is a third line, listed separately in the filing: an obligation to buy up to $14.7 billion of cloud capacity over five years, which can only shrink if the cloud provider manages to sell that capacity to someone else.

We re-ran the filing's two headline numbers ourselves — $355.7 billion in leases, $454.5 billion in contracts — and they tally to $810.2 billion, which tells you the much-quoted "$825 billion" is itself a rounded headline over line items the accountants are still nailing down. Precision is beside the point. The shape of the number is what matters: most of it is locked in, and it grew by roughly $125 billion in a single quarter, from almost $700 billion in July.

Wednesday's party, Thursday's hangover

Credit where it is due: the quarter was genuinely good. Ad impressions rose 10 percent, average price per ad climbed 14 percent, and Meta forecast fourth-quarter revenue of $73 billion to $76 billion — the midpoint ahead of what analysts expected. Muse, the consumer AI agent Meta launched in September, passed 10 million accounts in the US and Canada. Mark Zuckerberg told investors the next phase would be "more operational" — scaling the product, growing revenue, cutting compute costs.

But the call also carried the other shoe. Meta lifted the floor of its 2026 capital-spending guidance to $135 billion (from $130 billion, top end still $145 billion), spent $34.91 billion in capex in the quarter alone, and announced a new class of non-voting shares that can ultimately be used to pay for acquisitions and other expenses. New share classes are not what cash-rich companies do when the till is full; they are what companies do when they are staring at a decade of spending that dwarfs their current output.

And then Thursday's 10-Q turned the vague unease into hard paper. The revenue beat was gone in a morning. That whiplash — from $748.80 after hours to a six-percent plunge at the open — is the sound of a narrative breaking. The AI trade's earnings story, which has carried US stocks for two years, suddenly had a cost line attached to it.

The credit market's verdict

If equity traders had a bad Thursday, bondholders had a worse one. According to ICE Data Services, five-year credit default swaps on Meta's debt rose as much as 8.9 basis points to about 119.3 basis points — the highest intraday level since the contracts began trading actively about a year ago. Meta's bonds weakened markedly against Treasuries.

For the uninitiated: a credit default swap is insurance on a company's debt. When the price hits a record, it means the people paid to assess default risk are more nervous than they have ever been. Meta is not remotely close to default — it holds $82.5 billion in cash and marketable securities — but CDS prices direction, not destination, and the direction is what counts.

Ben Laidler, head of macro and equity strategy at Bradesco BBI, put it bluntly to Reuters: "There are lots of crosscurrents in markets at the moment, which are grappling for a narrative right now." And the one crossing hardest: "The big concern is that AI-driven earnings narrative, which has supported US earnings and spread to global equities, may be challenged here."

Here is the part that should worry everyone, not just Meta shareholders: the fear is not Meta-specific. Protection costs also edged up on debt from Oracle, Microsoft and Broadcom. The money is treating the borrowing wave behind the AI buildout as an industry-wide condition. Nebius just borrowed $3.5 billion in bridge financing despite sitting on $40 billion of contracts; the market now reads every such disclosure as another log on the same fire.

Photo: An aerial view of Meta's Menlo Park headquarters, where the company keeps raising the floor on its AI spending. (InvadingInvader / Wikimedia Commons, CC BY-SA 4.0 — file photo)

The cash question Zuckerberg can't dodge

The polite version of the concern is "capital intensity." The impolite version is on the cash-flow statement. Meta's free cash flow fell 91 percent year over year in the second quarter (per Crypto Briefing's reading of the numbers), and in the third quarter it went negative: minus $2.1 billion, against plus $10.6 billion a year earlier. The company spent $34.9 billion in capex in a single quarter — more than some countries spend on their militaries in a year, as Crypto Briefing framed it, and only the US, China and Russia outgun it.

One set of analysts, cited by MoneyCheck, does not expect free cash flow to turn positive again until 2029. That is a three-year desert. Meta will cross it with an advertising machine that is still superb — the 14 percent rise in ad prices proves the AI investment is at least improving the core business — and a legal docket that keeps billing. This was also the week a New Mexico judge handed Meta a $25 billion privacy penalty, and the quarter already absorbed a $9.89 billion charge to settle youth-harm claims with state attorneys general.

Morgan Stanley's Brian Nowak lifted his Meta price target to $820 and kept his Overweight rating — but even he, per MoneyCheck, urged investors to keep adoption rates in perspective. Ten million Muse accounts sounds like a movement; it is a rounding error against 3.64 billion daily users across Meta's apps. The AI spending is real, monetised, and enormous. The AI revenue, so far, is potential.

A bet or a trap? Here's our call

This column is required, by the reader's patience, to take a position. So: the $825 billion is not a bet. Bets have exits. You can be wrong about a bet and walk away poorer but free. $454.5 billion of this cannot be cancelled, and the rest stretches to 2036 — a decade over which the AI industry's current certainties about model architectures, chip vendors and compute pricing could be turned over three times.

None of this means Meta is in trouble. It means Meta has chosen a particular kind of trouble-proofing: it is spending so much, so fast, that the only way to lose is for the entire AI thesis to be wrong. That is not irrational. It is, in fact, exactly what a rational monopolist does when it believes compute is the new land. But let us not call a locked door a doorway. Zuckerberg has spent a quarter of a trillion dollars' worth of future cash on the assumption that more compute always wins, and the fine print says he cannot change his mind.

The botched part is not the ambition; it is the structure. Colossally ambitious spending with colossally rigid terms is how you turn a lead into a liability the moment the cycle turns. Wall Street understood that in about four hours on Wednesday night. It is worth understanding for longer than that.

What exactly is the $825 billion?

A disclosure, not a check. It is the sum of future obligations in Meta's 10-Q: $355.7 billion in off-balance-sheet leases for data centers and energy infrastructure (running from late 2026 through 2036) plus $454.5 billion in non-cancellable contracts for cloud compute, energy and hardware. Spread over a decade, it is roughly what UBS's estimate that big tech will burn $1 trillion on AI would predict for a company of Meta's weight class — except Meta's portion is already signed.

Is Meta in financial trouble?

No — and that is what makes the story interesting. Revenue is up 26 percent, the ad business is firing, and Meta holds $82.5 billion in cash and marketable securities. The issue is not solvency; it is flexibility. Negative free cash flow, a shrinking capex runway (analysts cited by MoneyCheck do not see FCF returning until 2029), a new non-voting share class for fundraising, and $454.5 billion that cannot be cancelled. Healthy companies can still make themselves hostages to fortune.

What is a credit default swap, and why does 119.3 basis points matter?

Think of it as fire insurance on corporate debt: you pay an annual premium, quoted in basis points, and if the company defaults you get paid out. At 119.3 basis points — a record for Meta, per ICE Data Services — insuring $10 million of Meta debt costs about $119,000 a year, the most it has ever cost. Nobody is predicting Meta defaults. The record tells you the credit market now prices AI-buildout leverage as a genuine, industry-wide risk premium, not a rounding error.

Who else is on the hook?

Effectively everyone building at hyperscale. CDS costs also rose for Oracle, Microsoft and Broadcom on the same day, and Morgan Stanley research cited this week shows record CDS across the AI complex — Oracle at 322 basis points (from 144 in January), Broadcom and Nvidia at records, CoreWeave above 1,000. AI-linked corporate debt issuance is running at an estimated $466 billion in 2026 versus $101 billion in the same stretch last year. Meta's filing is the loudest data point in a market-wide repricing, not a lone accident.

What should investors watch next?

Three things. First, the Q4 capex number: Meta has spent $85.8 billion in nine months, leaving $49–59 billion to go in the fourth quarter — the heaviest quarter of the year by a wide margin. Second, whether Muse's 10 million accounts turn into revenue or just compute bills; CFO Susan Li has already said total serving costs will rise with Muse's compute needs. Third, the 2027 budget — the earnings release conspicuously contained no 2027 capex figure, only a promise to "continue investing aggressively." After this week's filing, investors will want to know exactly what aggressively means.

Sources: Bloomberg reporting via Sarah Frier (Moneyweb, Oct 22); Meta Q3 10-Q filing (Oct 22) and Q3 8-K (Oct 21); ICE Data Services figures via BigGo Finance; TipRanks; Crypto Briefing; evidinvest 10-Q/8-K read-back; CNA/Reuters (Ben Laidler, Bradesco BBI); MoneyCheck (Morgan Stanley's Brian Nowak); BullBuzz; aibusiness.vc.

By Daniel Rao
Daniel Rao covers big tech and the money behind the AI buildout for TechPulse USA. He writes about what the filings say when nobody's cheering.

 

Update, Oct. 22: the session the filing is still falling through

The selloff deepened intraday (investinglive.com, Oct 22): Meta was down ~7.3% around midday to $651.61, off the $677.50 premarket level; Q3 capex hit $34.9 billion, up from $31 billion in Q2 and $19.4 billion a year earlier; this is Meta's first negative-free-cash-flow quarter since 2012; GAAP EPS was $3.83 vs $3.56 expected — and would have been $7.40 without the legal charge. The two commitment buckets went from ~$139 billion to $810 billion in a year, nearly 6x.

The 2027 wall (finance.biggo.com, Oct 22): about $45.85 billion of the non-cancellable contracts falls due in 2026 and $121.41 billion in 2027. Meta also moved $10.8 billion of money-market funds into restricted cash equivalents to satisfy escrow requirements under multiyear infrastructure procurement agreements, expected to be released between 2028 and 2030.

The comparison point (cryptobriefing, Oct 22): in Q2 the non-cancelable commitments stood at $349.3 billion and signed-but-not-started leases at $279 billion; Meta spent $72 billion on capex in all of 2025 versus the 2026 guide of $135–145 billion.

Class C in plain English (theamericanquorum.com, Oct 22): Meta will do a one-time dividend of one non-voting Class C share for each Class A and B share, effective January 19, 2027, and will seek a Nasdaq listing — a new corporate currency for acquisitions and fundraising that preserves Zuckerberg's voting control while potentially diluting economic interest.

The bull case, restated (marketwatch.com, Oct 22): Zuckerberg told the earnings call Muse could ultimately generate more revenue per user than the family of apps; Q3 capex of $34.91 billion came in slightly under the $41.12 billion consensus.