While everyone watches capex numbers in earnings calls, Meta just revealed it's locked into nearly $280 billion in lease commitments that won't hit the balance sheet until the facilities actually open.
The Summary
- Meta disclosed $279 billion in future lease obligations, mostly for AI data centers and network infrastructure, in its Q2 2026 securities filing — a 53% jump from $183 billion three months prior
- After the quarter closed, Meta signed another $68 billion in data center leases scheduled to begin in 2027-2028 with 18-20 year terms
- Meta and BlackRock formed a joint venture for a 1-gigawatt data center complex in Texas, part of $14 billion in infrastructure investment
- These obligations sit off-balance-sheet until facilities come online, scheduled between now and 2036 with lease terms ranging from one to 30 years
The Signal
Meta's $279 billion in future lease commitments represent a scale of infrastructure pre-commitment the tech industry has never seen. This isn't speculative R&D spending. These are signed contracts for physical data centers, colocations, and network infrastructure with terms stretching three decades. The company went from $183 billion in future obligations in Q1 to nearly $280 billion in Q2, then added another $68 billion in July alone.
The pace is accelerating, not stabilizing. That $68 billion in post-quarter leases translates to roughly $2.2 billion per day in new long-term infrastructure commitments during July. These aren't short-term bets either: 18-20 year lease terms mean Meta is locking in costs through 2048, betting that whatever AI architecture exists two decades from now will still need the physical infrastructure they're signing for today.
"Meta signed $68 billion in additional data center leases after the quarter ended, committing to infrastructure that won't come online until 2027-2028."
What makes this moment different from previous infrastructure buildouts:
- The capital is pre-committed before use cases are fully proven at scale
- Lease terms extend far beyond typical product lifecycles in tech
- BlackRock's involvement as a joint venture partner signals Wall Street treating AI infrastructure as an asset class, not just tech capex
- The 1-gigawatt Texas facility alone represents industrial-scale power consumption, equivalent to a small city
The off-balance-sheet structure matters. These obligations don't appear on Meta's books until facilities begin operations, which means the company's reported liabilities significantly understate its locked-in commitments. This is legal and standard accounting practice, but it means investors looking at balance sheet leverage ratios are missing nearly $280 billion in future obligations that will inevitably flow through as operating expenses.
The BlackRock partnership reveals another shift: traditional tech companies building their own infrastructure are now partnering with asset managers to fund and operate facilities. This changes the risk profile. Instead of pure build-to-own capex, Meta is essentially pre-leasing capacity from facilities that haven't been built yet, with BlackRock taking construction and operational risk in exchange for long-term contracted revenue.
The Implication
If you're building in the agent economy, watch where the big infrastructure money flows. Meta's quarter-trillion-dollar bet isn't just about training larger models. It's about persistent compute for billions of AI agents that need to run 24/7, not just during training runs. The 18-30 year lease terms suggest Meta expects agent workloads to be the steady-state future, not a temporary spike.
For anyone building AI infrastructure businesses, the message is clear: the hyperscalers are locking up capacity years in advance. If your business model depends on flexible access to compute at spot prices, that window is closing. The flip side: if you're selling infrastructure, data center operators, or power solutions, there's now a decade-plus runway of committed demand from companies that can't back out without eating massive losses.