Meta commits to multi-year datacenter capacity rental agreements reportedly valued at 960B+ in a single quarter, locking in infrastructure for 20 years to support AI/compute workloads.
Meta's latest regulatory filings reveal that as of June 30, the company's unfulfilled data center leasing commitments totaled $279 billion—a staggering 53% surge from the previous quarter. July alone brought an additional $68 billion in new commitments. The AI arms race has escalated from competing for GPUs to securing power supplies and infrastructure for the next two decades.
On July 29, Meta filed its latest regulatory document. While markets fixated on revenue and advertising, the truly alarming figure was buried in the footnotes: as of June 30, Meta's total unexecuted leasing commitments reached $279 billion. Just one quarter prior, this figure stood at $183 billion—a 53% increase in a single quarter.
To contextualize $279 billion: Meta's entire 2025 full-year revenue is projected at $201 billion. In other words, the company has committed more in data center rent for facilities that don't yet exist than its projected total annual revenue.
And the spending continues. July alone saw Meta sign $68 billion in additional leasing commitments, with contract execution scheduled to begin in 2027 and 2028.
This money doesn't yet appear on the balance sheet—since the leases haven't commenced and rent hasn't begun accruing, there's nothing to report. When execution begins in 2027, they'll transform into real cash outflows.
Why not build facilities outright? Meta does build, but strategically. The day before filing, Meta announced a textbook financial maneuver: a joint venture with BlackRock to develop a data center campus in El Paso, Texas. BlackRock contributes the bulk of capital for an 80% stake; Meta provides the land and construction in progress, receiving 20% ownership plus approximately $1 billion in immediate cash. Once operational, the entire campus will lease back to Meta on a flexible 20-year term. Meta provides a residual value guarantee of up to approximately $13 billion.
Through this sequence of transactions, massive construction expenditures stay off-balance-sheet while operational control remains intact. This explains the 204% annual surge in leasing commitments.
Zooming out, the table is crowded. Bloomberg data shows that Meta, Microsoft, Oracle, Amazon, CoreWeave and five other companies face more than $850 billion in future data center leasing obligations—up 204% year-over-year. Oracle leads at approximately $250 billion, with the bulk backing the $300 billion "Stargate" deal with OpenAI; Microsoft follows at roughly $197 billion. These leases lock in 15-19 year terms—the tech giants aren't racing for facilities, they're monopolizing power and land for the next two decades.
Start with burn rates. Meta's capital expenditures climbed from $39 billion in 2024 to $72 billion in 2025. Full-year 2026 guidance has been raised to $125-145 billion, with an April adjustment citing rising server and storage costs. JPMorgan is even more aggressive, directly forecasting 2027 capital expenditures could exceed $200 billion.
Now look at cash velocity. Meta's 2025 free cash flow sits at $43.6 billion; 2026 full-year expectations drop to just $1.8 billion. Wall Street is already trading the narrative that 2027 FCF could turn negative. In a single quarter, Meta committed to $107 billion in multi-year infrastructure while cutting 8,000 employees in May. Zuckerberg has been explicit: compute and labor are the two cost pillars, investing in one means cutting the other.
Risk signals exist. Oracle's 10-K filing contains a rare acknowledgment: if it overestimated customer demand or a key customer can't afford its commitments, the company becomes trapped in years of excess capacity with no exit option. Bloomberg openly names the "key customer" as OpenAI. Oracle's stock fell 35% in June, the worst performer among peers.
Bulls have ammunition too. Meta's Q1 revenue hit $56.3 billion, up 33% year-over-year, with ad impressions up 19% and prices up 12%. Its proprietary GEM advertising model delivered a 3.5% click-through improvement for Facebook in a single quarter alone—translating to tens of billions in incremental revenue on a $200 billion ad market. Management has signaled flexibility: if returns disappoint, investment pace can be adjusted.
Deeper still, this leasing arms race has fundamentally transformed the AI competition: from "who has the most chips" to "who dares wager all their cash flows for the next twenty years." Winners get computing dominance; losers carry the bill through 2040 and beyond. The paradox is that no one at the table dares stand up and walk away first.