Meta Buys 1.6 Gigawatts It Will Never Own
Meta has signed new agreements with Crusoe for roughly 1.6 gigawatts of artificial intelligence computing capacity, split across two of the data center developer's campuses in Childress, Texas, and Warrenton, Missouri. Bloomberg first reported the deal on June 18, citing people familiar with the talks, and neither company has disclosed how much Meta will pay or when the capacity arrives. For an operator that has historically owned and built its own data centers, the structure matters: Meta is renting compute, not pouring concrete.
That is a meaningful tell about how the AI buildout is actually being financed. At 1.6 gigawatts, this single contract is large enough to power roughly 1.2 million American homes, and it lands on top of a five-year, up-to-$27 billion arrangement Meta struck with Nebius in March and a string of nuclear power purchase agreements. The company is assembling capacity wherever it can be brought online fastest, and increasingly that means leasing turnkey AI factories from specialists rather than waiting on its own construction pipeline.
Crusoe's Pivot From Flared Gas to Hyperscale Landlord
Crusoe was founded in 2018 by Chase Lochmiller and Cully Cavness on a narrow premise: capture flared natural gas at oil wells and burn it to power mobile data centers. That origin story is now almost irrelevant to the business. Crusoe has pivoted into building gigawatt-scale AI campuses, and its contracted pipeline sits near 4.9 gigawatts with a development pipeline the company says exceeds 40 gigawatts. The Abilene, Texas campus that anchors part of OpenAI's Stargate effort runs to 1.2 gigawatts and is already operational.
The Meta deal slots Crusoe alongside an enviable customer list that already includes Microsoft, Oracle, and OpenAI. In March, Crusoe signed a separate 900-megawatt agreement with Microsoft. A Series E round earlier reported at $1.37 billion pushed its valuation past $10 billion, and the company has raised roughly $7 billion in project debt to fund an OpenAI complex in Texas. Crusoe is no longer a clever energy startup. It is becoming one of the most important landlords in American AI infrastructure.
Why Meta Is Renting Instead of Building
Meta has pledged up to $600 billion in US infrastructure and jobs over three years, a figure CEO Mark Zuckerberg has tied directly to his AI ambitions across assistants, agents, recommendation systems, and frontier models. Given that war chest, leasing 1.6 gigawatts from a third party looks counterintuitive. The logic is speed and risk transfer. Building a gigawatt campus from raw land takes years and ties up capital in long-lived assets, while a developer like Crusoe absorbs the construction risk, the power procurement, and the supply chain headaches.
We read this as a deliberate hedge against the two scarcest commodities in the buildout: time and power. Crusoe has already secured interconnects and sites in Texas and Missouri that Meta would otherwise have to win on its own, often against local opposition and grid constraints. By renting, Meta converts a multi-year capital project into a contractual obligation that can flex as model economics shift. The trade-off is that Meta does not own the asset and pays a margin to the developer for that flexibility.
Texas and Missouri, Again, for a Reason
The choice of Childress, Texas, and Warrenton, Missouri, is not random. Both states have become magnets for AI capacity because they offer land, comparatively accommodating permitting, and access to power that more saturated markets like Northern Virginia can no longer guarantee on the timelines hyperscalers demand. Texas in particular has emerged as the center of gravity for the largest campuses, from Crusoe's Abilene site to the wider Stargate footprint, helped by the ERCOT grid's willingness to interconnect large new loads.
Missouri, meanwhile, is fast becoming a serious interior play. Amazon recently committed $10 billion to a Missouri data center campus, and Nebius is breaking ground on a gigawatt-scale AI factory in Independence. The migration inland reflects a hard truth: the coastal data center hubs are running out of headroom on power and water. The next wave of AI compute is being planted in the middle of the country, close to generation and far from the grid bottlenecks choking the established markets.
The Neocloud Credit Question Lurking Underneath
There is a wrinkle the headlines tend to skip. The same week brought a reminder that not every neocloud deal closes cleanly: Applied Digital's arrangements with CoreWeave hinge on a subsidiary achieving an investment-grade credit rating, a clause that exists precisely because counterparty credit risk now haunts these long-dated capacity contracts. Developers are committing tens of billions in debt against leases whose creditworthiness investors are scrutinizing harder than they did a year ago.
For Meta, that risk runs the other way. Meta is the investment-grade anchor tenant that makes a developer's project financing bankable, which is exactly why Crusoe wants its name on the contract. The 1.6-gigawatt agreement is as much a financing instrument as a compute purchase: it gives Crusoe a credit-worthy offtake to borrow against. As the AI buildout matures, the deals that get funded will increasingly be the ones with a hyperscaler's balance sheet standing behind them, and the developers without that backing will feel the squeeze.
What Enterprises Should Take From This
The strategic signal for enterprise leaders is that even the most capital-rich operator on earth has concluded it cannot build fast enough alone. Meta's willingness to lease 1.6 gigawatts validates a procurement model that mid-market and large enterprises can borrow: secure AI capacity through specialist developers and neoclouds rather than attempting to stand up scarce, power-hungry infrastructure in-house. The economics of speed favor renting when the asset risks obsolescence on a GPU refresh cycle.
The open questions are price and durability. Meta disclosed neither the contract value nor the delivery schedule, so it is impossible to judge whether this is a bargain or a premium paid for certainty. We would watch whether these leasing arrangements lock in favorable terms or expose tenants to renegotiation as power costs climb. For now, the message is unambiguous: the AI infrastructure market has split into those who build, those who rent, and a growing class of developers like Crusoe who profit by serving both.
