Meta printed $64.6 billion in revenue for the quarter, up 26% year over year. Operating income landed at $12.99 billion — down 37%. And earnings per share came in at $3.83 against $1.05 in the same period a year earlier, a 265% jump that lit up every terminal on my desk before the bell.
Two of those three numbers cannot be simultaneously true in any world where accounting still means something.
I have spent four years translating filings into retail-readable threads on an exchange floor, and the first rule I give every junior analyst is blunt: when headline EPS runs in the opposite direction of operating income, you are not looking at growth. You are looking at a base effect, a one-off charge, or a bad transcription. This is a base effect, and it is enormous. The $1.05 print from a year ago was never a real earnings figure. It was what remained after roughly $15.9 billion in one-time, non-cash tax charges flattened the line.
Strip that out and last year's comparable EPS sits near the $7 mark. Which means this quarter's $3.83 is not up 265% — on a like-for-like basis it is down more than 40%.
Speed is the only currency that matters. The fastest thing moving through this report is the misinformation.
Meta stopped being a software company that happens to sell ads about two years ago. It is now an infrastructure company that happens to monetize attention, and the capex guidance is the confession: $135 billion to $145 billion for the year, with Reality Labs absorbing another $18–19 billion in losses. Crypto readers have a bad habit of treating legacy earnings as background noise, but this one matters for a structural reason. The capital deployed here sets the floor for every decentralized compute network, every GPU rental marketplace, every tokenized data-center play trading on the AI infrastructure thesis. When the largest single buyer of compute in the world doubles its order book, the ripples cross the fence whether we like it or not. From the front lines of the hype cycle, that is the part worth reading slowly.
There is a second reason to slow down. The report that reached me was a translated wire copy: seven data points, no fiscal-year label, no user metrics, no segment revenue, no free cash flow, no cost breakdown, no call guidance, no regulatory disclosure. That is not an accident of translation. It is the shape financial news takes as it degrades in transit. I rebuilt the prior-year base from the growth rates to confirm the figures were self-consistent, and they were — $64.6 billion divided by 1.26 lands at $51.27 billion, which matches the actual prior-year quarter. The math checks. The framing does not.
Here is the mechanical problem inside the report's own logic. A company that grows revenue 26% while operating income falls 37% is telling you its cost base expanded faster than its top line — roughly $7.6 billion of incremental operating profit vanished year over year. Capital expenditure does not hit operating income directly; it hits through depreciation, and depreciation on a fleet built over the last six quarters is now landing on the P&L in a wave. Same for AI research headcount, same for the data-center power contracts signed at 2024 and 2025 rates. The fall from a prior operating income base in the $206 billion range to $129.9 billion is not a one-quarter event. It is the first visible arc of a multi-year depreciation curve Meta has already committed to.
That brings me to the concept I keep circling — revenue duration. Advertising revenue has a duration of roughly one quarter. A buyer can pause spend tomorrow with no contractual penalty. Infrastructure capex has a duration of a decade or more. You cannot un-build a data center. Meta is financing a ten-year asset with a ninety-day revenue book, and the mismatch widens every quarter the capex guidance ratchets higher. In crypto we know this structure intimately. It is the same mismatch that broke every DeFi protocol that borrowed short and lent long — survivable while the funding rate stays positive, merciless the moment the curve inverts.
Run the cash math and the shape sharpens. Capex of $135–145 billion is now approaching, and on a weak quarter exceeding, full-year operating cash flow. That does not mean Meta is insolvent. It means free cash flow compresses toward zero, and a company with zero free cash flow that still wants to build has exactly three doors: debt, equity, or slowing down. Meta chose equity. The Class C issuance is not a rounding error in the capital structure — it is the disclosure itself. You do not dilute shareholders to fund a build you believe your own cash generation can carry. Equity-funded capex is an admission that the internal engine cannot fund the ambition.
Now the part that actually touches our markets. When a hyperscaler commits $140 billion, two things happen to decentralized compute simultaneously, and they pull in opposite directions. Near term, hardware demand and power contracts get bid up, and every DePIN network with a token and a rented GPU cluster catches a narrative tailwind it did nothing to earn. Medium term, that same capex floods the market with centralized capacity and rental pricing compresses. Distributed networks end up competing on price against a supplier with a lower cost of capital and a ten-year depreciation schedule it can absorb without flinching. I have sat through three compute-token launches this year that pitched "cheaper than the hyperscalers" as the core thesis, and every one of them was pricing against a spot market about to receive a lot more supply.
Then there is the open-weights track, which almost nobody in the earnings coverage connects to the capex line. Meta's model releases are not charity and not ideology. They are a distribution strategy aimed at making a particular inference stack the default every downstream developer builds on. Open weights buy ecosystem; closed advertising monetizes the consumer surface. Two tracks, one balance sheet, and the capex number is what funds the open half. For crypto builders that private-public split should look familiar — it is the same logic as a protocol that open-sources its client and monetizes its order flow.
What is missing from the report is as diagnostic as what is in it. There is not a single user metric in the wire copy — no daily actives, no family daily actives, no time spent. For a company whose entire asset base is attention, that omission removes the one variable telling you whether the capex is buying a moat or buying a hole. I pulled the translated copy apart twice looking for a DAU line and found nothing. When a summary drops the user dimension and keeps only the money dimension, the summary was written from a spreadsheet, not from the filing.
The consensus read on this report is an AI capex story. That is the comfortable read, and it is the one that misses the actual signal. The interesting thing is not that Meta is spending $140 billion on infrastructure — every hyperscaler is. The interesting thing is that it is spending it while operating income contracts at a rate that would have triggered a board crisis in any other cycle, and funding the gap with new shares rather than cash. Margin compression plus equity issuance plus an unchanged-or-rising capex guide is what a land grab looks like from the inside. Meta is not buying efficiency. It is buying the right to still be relevant when the compute layer consolidates, and it will compress earnings for three years to hold that seat.
The blind spot in our own market is worse. The crypto AI trade has been structurally long compute supply and structurally short the idea that demand consolidates. But demand is consolidating — into four or five balance sheets that can each outspend an entire token ecosystem in a single quarter. DePIN's real addressable market was never "cheaper than AWS." It was the workload categories hyperscalers structurally cannot serve: verifiable inference, censorship-resistant compute, sovereign and privacy-constrained deployments. If your network's pitch is price, the largest capex program in corporate history is about to price you out. Pivoting when the chart says pause is not weakness; it is survival.
Watch the next two prints, and watch one line specifically. If capex guidance is revised higher again while operating cash flow holds flat, free cash flow crosses zero and the story stops being about AI ambition and starts being about financing structure. That is the print that reprices everything downstream — including every compute token currently trading on the assumption that big tech will always be the buyer. Surviving the winter to plant for spring only works if you know which winter you are standing in.