Meta’s AI Data Center Leases Are Reportedly Approaching $279 Billion. Almost None of It Is on the Balance Sheet.
Meta disclosed in its own SEC filing that data-center leases it has signed but not yet started paying on — real commitments that haven’t hit the balance sheet — reached $182.88 billion at the end of Q1 2026, up 76% in a single quarter. Bloomberg has since reported the figure climbed further, to roughly $279 billion, by the end of Q2 on June 30.
That $279 billion figure deserves a caveat: Bloomberg’s original reporting sits behind a paywall that blocked direct verification during our research, so we could not independently confirm the exact number from Bloomberg’s own text. What we can confirm — from Meta’s own filing and Bloomberg’s separately verified reporting on Microsoft — is enough to make the underlying story hold up regardless.
Microsoft’s own equivalent figure — fully confirmed — hit $329.1 billion for the quarter ended June 30, 2026, up from $196.6 billion three months earlier: a jump of more than $130 billion in a single quarter, on the record. Whatever Meta’s exact figure turns out to be, the pattern is real, accelerating, and largely invisible on the balance sheet investors actually read.
- $182.9B Meta's own disclosed value of data-center leases “not yet commenced” at the end of Q1 2026 — up 76% in three months — Meta 10-Q, SEC EDGAR
- $329.1B Microsoft's confirmed equivalent figure for the quarter ended June 30, 2026, up from $196.6 billion — a jump of more than $130 billion in one quarter — Bloomberg, July 29, 2026
- $662B off-balance-sheet future lease commitments Moody's found across the 5 largest hyperscalers combined — 113% of their own adjusted debt — Moody's Ratings / Fortune
Under standard lease accounting, a company doesn’t put a lease on its balance sheet until payments actually begin. For a data center, that can be years after signing — while the site is still being poured, wired, and fitted with racks of chips. That gap is where Meta, Microsoft, Oracle, Amazon, and Alphabet are parking hundreds of billions in committed future spending on the AI buildout: real obligations living in SEC-filing footnotes, not the balance-sheet number a headline earnings report is built around.
Confirmed: Meta’s SEC 10-Q for the quarter ended March 31, 2026 discloses $182.88 billion in “operating and finance leases that have not yet commenced” — Meta’s own words, in its own primary filing.
Confirmed: Microsoft’s equivalent figure for the quarter ended June 30, 2026 is $329.1 billion, up from $196.6 billion, per Bloomberg’s July 29, 2026 report — reporting this site could read and cite directly.
Reported, not independently verified: A Meta figure near $279 billion for the quarter ended June 30, 2026 has been reported and is consistent with the trend, but Bloomberg’s own original article sits behind a paywall this site could not get past to quote directly. We are naming that gap rather than treating the number as confirmed.
Bloomberg’s own account flagged the trend in real time on X back in June, ahead of either company’s most recent quarter closing:
Microsoft and Meta each committed nearly $50 billion in additional data center leases in their most recent quarters, underscoring an escalating bet the tech industry is making on artificial intelligence.
Meta’s and Microsoft’s numbers move together because the same forces drive both: a race to reserve every gigawatt of power and every rack of chips before a rival locks it up first. Benzinga’s non-paywalled coverage corroborates the trend where Bloomberg’s exact figures sit behind a subscription: it puts the industry-wide total for these not-yet-commenced leases above $850 billion as of late June 2026, naming Meta and Microsoft as the two biggest drivers. Bloomberg’s own recurring series on this metric found the industry aggregate at $700 billion in March and $850 billion in June — Meta’s and Microsoft’s disclosures are the two largest line items in both.
The comparison matters because it shows Meta isn’t an outlier — it’s running roughly the same playbook as its biggest rival, a notch behind on dollar terms. Microsoft’s jump from $196.6 billion to $329.1 billion in the three months ending June 30 is the largest confirmed quarterly increase either company has disclosed. If Meta’s reported $279 billion holds up, it means Meta added roughly $96 billion in new commitments in one quarter — smaller than Microsoft’s jump, on a smaller base, but moving at nearly the same speed.
Bloomberg’s own account posted the Microsoft figure the day its second-quarter numbers landed — the one piece of this story this site could verify word for word:
Microsoft reported more than $130 billion in new data center leases in the past quarter, signaling an accelerating pace of spending on AI infrastructure.
Meta and Microsoft don’t have the field to themselves. Oracle carries the largest single company-level commitment of any hyperscaler — snapshots this year put the figure between roughly $250 billion and $261 billion, much of it tied to its contract backing OpenAI’s Stargate project. Alphabet discloses a related but distinct metric: future data-center lease payments, which hit $42.6 billion in Q3 2025, up from $23.9 billion the prior quarter. Separately, Alphabet’s total purchase commitments — a broader category — jumped from $332.4 billion to $811 billion in one quarter, underscoring how fast every version of this number is moving industry-wide.
Amazon is the outlier in the other direction. It added only roughly $10 billion in new lease commitments in Q1 2026 — a fraction of what Meta or Microsoft each add in a single quarter. That gap reflects a real strategic difference: Amazon leans more heavily on capacity it owns and builds itself, so less of its AI buildout shows up in this accounting category at all.
Moody’s Ratings sized the full picture across the five largest hyperscalers — Amazon, Meta, Alphabet, Microsoft, and Oracle — combined: $969 billion in total future lease commitments as of year-end 2025, with $662 billion of that off the balance sheet. Moody’s analyst David Gonzales puts that figure at 113% of the five companies’ own adjusted debt — the hidden commitments now exceed the debt investors can already see, a gap he frames as real economic risk even though it doesn’t meet the threshold for balance-sheet recognition.
The Motley Fool went further for Meta specifically, adding its $182.88 billion in not-yet-commenced leases to $237.67 billion in broader non-cancelable commitments to arrive at what it called “$420 billion in hidden debt” — a July 22, 2026 piece that also cited a Nikkei Asia estimate putting the same five hyperscalers’ combined off-balance-sheet exposure at $1.65 trillion.
Meta raised its 2026 capex guidance to $130–145 billion on its July 29 earnings call, up from $125–145 billion previously. Q2 capex alone hit $31.1 billion, and free cash flow collapsed to $784 million — down roughly 91% year over year — as buildout spending consumed nearly all operating cash flow. Meta’s stock fell 7–10% after hours despite a revenue beat of $60.8 billion (up 28% year over year), because earnings per share missed at $6.18 against consensus above $7. The market was reacting to the spending trajectory, not weak sales.
“The industry has underbuilt historically for the wave of AI adoption, making existing capacity, including our own, extremely valuable.”
Susan Li · CFO, Meta · Q2 2026 earnings call, July 29, 2026
Li framed the near-term plan narrowly: “Our current plans are geared towards maximizing 2026 and 2027 capacity,” with longer-term strategy meant to “continue growing compute in 2028 and beyond.” Mark Zuckerberg defended the buildout in market terms: “We’re getting a lot of offers for compute at a significant premium over what we paid for it… It would be foolish to basically just sell all of the compute and take a short-term profit.” A day earlier, announcing a $14 billion joint venture with BlackRock for a 1-gigawatt El Paso, Texas data center — BlackRock 80%, Meta 20% — he shifted to mission framing: “Building the infrastructure for superintelligence is key to making sure the benefits of this technology are distributed to everyone.”
Meta’s Hyperion data-center project in Louisiana was financed through $27.3 billion in bonds via a joint venture led by Blue Owl Capital — a structure under which only $2.37 billion shows up as an investment on Meta’s own balance sheet. A roughly $28 billion residual value guarantee attached to the deal was deemed “not probable” and excluded from Meta’s liability recognition entirely. The bonds carry a 6.581% coupon, maturing in 2049 — one project, financed almost entirely off one company’s own books.
Wall Street’s read on where this goes next isn’t settled. Deutsche Bank analyst Benjamin Black projects Meta’s 2027 capex in the “low-to-mid $200 billion range” — roughly $210–215 billion — rising to about $265 billion by 2028, calling the trajectory “a key debate on the stock.” A CNBC Fast Money host described a recent Meta bond offering as pricing “just below like freaking junk,” and noted roughly 92% of hyperscaler operating profit industry-wide now goes straight to capex.
“History is littered with examples of industries that spent massively on infrastructure in anticipation of demand that arrived more slowly than expected. The most recent example is the fiber-optic networks of the late 1990s.”
Real Investment Advice
The fiber-optic comparison is a warning, not a prediction — those networks eventually filled up and became the modern internet’s backbone, years later than financiers expected, after a wave of bankruptcies wiped out the companies that borrowed to build them first. Whether Meta, Microsoft, Oracle, and the rest of the industry are building the AI era’s version of that backbone, or its version of the write-offs that preceded it, is what hundreds of billions in off-balance-sheet commitments now rides on.
Meta’s own SEC filing confirms $182.9 billion in AI data-center leases that haven’t started yet; Bloomberg reports that figure near $279 billion one quarter later, though this site couldn’t independently pull that exact number past Bloomberg’s paywall. What is fully confirmed — Microsoft’s $130 billion-plus one-quarter jump, Moody’s $662 billion off-balance-sheet estimate, and Meta’s own capex guidance climbing toward $145 billion — is enough on its own: the AI buildout’s biggest costs increasingly live in footnotes, not the balance sheet, and the bill comes due whether or not one headline number is exact.
