Meta has signed new deals to draw computing power from Crusoe, the data-centre developer, adding to the vast and growing pile of infrastructure the company is assembling for its AI ambitions. The agreements cover two sites, in Childress, Texas, and Warrenton, Missouri, and will give Meta roughly 1.6 gigawatts of capacity between them, according to a Bloomberg report citing people familiar with the matter.
The figure is the part to dwell on. A gigawatt is the rough output of a large power station, and 1.6 of them dedicated to a single company’s computing needs is a measure of how the unit of AI ambition has shifted from chips to electricity. Meta is no longer just buying servers; it is contracting for power at the scale of national grid planning, and doing so across multiple developers at once.
Crusoe is the counterparty of the moment. Founded in 2018, the company specialises in building and running large data centres for AI work, and it already has deals with Oracle, Microsoft, and Alphabet’s Google for three other campuses.
Adding Meta to that roster places Crusoe among the developers the largest AI buyers turn to when they need capacity faster than they can build it themselves.
For Meta, the deals are one piece of a much larger and more expensive strategy. Like its Silicon Valley peers, the company has been racing to lock in computing power wherever it can find it, and its single biggest effort is a nearly 4,000-acre campus in Louisiana designed to provide up to five gigawatts of capacity.
Against that, the [Hyperion project](https://thenextweb.com/news/meta-200-billion-hyperion-data-center-louisiana), the Crusoe agreements are an incremental top-up to an already staggering build-out.
That build-out is straining the resources around it. Data centres of this size need power and water on a scale that has begun to draw scrutiny from grid operators and communities, and the competition for electricity is now a defining constraint on the industry, the same one visible in operators racing to secure gigawatts across multiple continents. Meta’s answer has been to spread its bets across developers and geographies rather than rely on any single site.
The spending behind all this is enormous and still climbing. The largest cloud and AI companies are collectively pouring hundreds of billions into infrastructure, and the demand has outrun even the most aggressive forecasts, with some operators renting capacity from unlikely sources to keep up.
Meta’s Crusoe deals are a small entry in that ledger, but they point in the same direction: secure the power now, at almost any cost, and worry about the returns later.
What Meta intends to run on the 1.6 gigawatts is the familiar list, training and serving the AI models and agents the company has reorganised itself around. The capacity is the precondition for the rest of the strategy; without it, the models and the products built on them cannot scale.
The shift toward leasing capacity from developers like Crusoe, rather than building every site in-house, is itself a strategic choice. Constructing gigawatt-scale data centres takes years and ties up enormous capital; contracting for capacity that someone else builds and operates lets Meta move faster and spread the construction risk, at the cost of paying a margin to the developer. With demand running ahead of supply, speed has become worth more than ownership, and Crusoe is selling exactly that.
Neither Meta nor Crusoe has commented in detail on the terms, and the figures come from people familiar with the arrangements rather than an official announcement. What the deals confirm is the trajectory: a company whose future depends on AI is buying the electricity to power it by the gigawatt, from whoever can deliver it fastest.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.