Meta has signed two massive AI compute deals with Crusoe Energy totaling 1.6 gigawatts across Texas and Missouri. The agreements, announced this week, mark one of the largest direct energy procurements tied to artificial intelligence infrastructure. They underline the growing pressure on power suppliers to deliver specialized, scalable solutions for data centers running machine learning workloads.
Why the scale matters
The 1.6 GW combined capacity is enough to power roughly 1.2 million U.S. homes, though Meta will use it exclusively for AI training and inference. The company has been racing to secure reliable electricity for its expanding fleet of data centers, which require uninterrupted, high-density power. Traditional utilities often struggle to meet that demand quickly, especially in regions with grid constraints or long interconnection queues. Crusoe Energy, which focuses on what the deal description calls “specialized energy solutions,” offers modular, often gas-fired or low-carbon power plants that can be deployed faster than conventional grid infrastructure.
Texas and Missouri: the chosen sites
One of the deals covers a site in Texas, where ERCOT’s deregulated market allows large buyers to negotiate directly with power producers. The other is in Missouri, a state that has seen less data-center buildout than Texas or Virginia but offers cheap land and relatively low electricity rates. Neither Meta nor Crusoe disclosed exact locations or timelines, but the 1.6 GW total suggests multiple campuses will be built over the next three to five years. The deals also represent a bet by Meta that the AI boom is not a short-term spike. Unlike cloud providers that sell compute by the hour, Meta builds infrastructure for its own products, making long-term energy contracts a strategic necessity.
A shift from traditional utilities
The agreements highlight a broader trend: big tech companies are moving away from relying solely on regulated utilities. Instead, they’re signing direct power-purchase agreements with independent developers that can build generation and sometimes even transmission alongside data centers. Crusoe, which started as a bitcoin-mining firm using stranded natural gas, has pivoted heavily to AI compute. The company now operates or develops gas-fired power plants that can run on pipeline gas or, in some cases, flare gas. For Meta, the appeal is speed and predictability. A utility-scale solar farm can take four years to permit and build; a Crusoe-style gas plant can be operational in 18 months.
The deals also raise questions about emissions. Crusoe has promoted its use of otherwise wasted methane for bitcoin mining, but the Meta contracts involve building new gas capacity, which will produce carbon dioxide. Meta has pledged to reach net-zero emissions by 2030, and large gas-fired data centers could complicate that goal unless offset by carbon credits or paired with carbon capture. The company did not provide details on how it plans to balance the new deals with its climate targets.
What’s next
With 1.6 GW signed, Meta is likely to announce more such deals in other regions. The company has previously said it expects AI computing demand to grow by 10x or more over the next few years. Crusoe, for its part, is reportedly raising additional capital to accelerate its buildout. The next milestone to watch will be the start of construction at the Texas and Missouri sites, along with any environmental permits that could slow the timeline.
