What Happened
Meta has signed a fresh set of agreements with data-center developer Crusoe that will hand the company roughly 1.6 gigawatts of AI computing capacity, spread across two new U.S. sites. According to a Bloomberg report published on June 18, the capacity will come from Crusoe facilities in Childress, Texas, and Warrenton, Missouri. The spending figure and the delivery schedule were not disclosed.
To put 1.6 gigawatts in perspective, a single gigawatt is enough electricity to supply roughly 750,000 American homes. The combined capacity Meta is locking in from Crusoe is therefore comparable to the power draw of a mid-sized city, dedicated almost entirely to training and running advanced AI systems. The deal lands as one of the clearest signals yet that the competition among the largest technology companies has moved from a contest over models to a contest over raw computing supply.
The agreements slot into Meta's broader pledge to invest about $600 billion in U.S. infrastructure and jobs over the next three years, a commitment chief executive Mark Zuckerberg has tied directly to the company's bet on AI assistants, agents, recommendation systems, and large language models. As Tech Startups noted, securing capacity from a third-party developer like Crusoe lets Meta scale faster than it could by building every facility itself.
Why It Matters
The headline number is large, but the more telling detail is the unit of measurement. AI infrastructure is now negotiated in gigawatts rather than megawatts, and electricity, not silicon, has become the binding constraint. U.S. data-center power demand is projected to reach about 76 gigawatts in 2026, up from roughly 50 gigawatts in 2024, with forecasts pointing toward 176 gigawatts by 2035. In that environment, a 1.6-gigawatt contract is less a routine procurement and more a claim staked on a scarce national resource.
.jpg){kind=link}
The capital behind this build-out is staggering. The five largest hyperscalers are on track to spend close to $690 billion on infrastructure in 2026, a roughly 36 percent jump over the prior year, with nearly all of it directed at AI. Meta's own capital expenditure is expected to land somewhere in the $115 to $135 billion range, alongside separate projects such as a one-gigawatt site in Ohio and a Louisiana campus that could eventually scale to five gigawatts. Against that backdrop, leaning on a specialist developer is a way to convert balance-sheet ambition into operational capacity without waiting years for in-house construction to catch up.
For Crusoe, the deal is validation of a strategy that looks unusual among data-center operators. Rather than simply leasing space, the company tries to control nearly every layer of the stack, developing energy sources, building the facilities, deploying the computing hardware, and running cloud services on top. That vertical integration is precisely what lets it move quickly when demand for compute is outpacing supply.
Reaction
The agreement marks another conspicuous win for a startup that has become one of the most closely watched names in AI infrastructure. Crusoe was founded in 2018 by Chase Lochmiller and Cully Cavness, and it began life capturing stranded natural gas to power cryptocurrency mining. As the appetite for AI computing exploded, the company pivoted hard into large-scale infrastructure, a turn that has made it one of the fastest-growing players in the sector.
_02.png){kind=link}
Investors have rewarded the shift. Crusoe raised about $1.37 billion in a Series E round in late 2025 at a valuation above $10 billion, co-led by Valor Equity Partners and Mubadala Capital. The company now counts Microsoft, Oracle, and OpenAI among its customers, while its GPU-focused Crusoe Cloud platform serves a roster of AI firms. Earlier this month it said it held contracts for some 4.9 gigawatts of capacity, with a total project pipeline exceeding 40 gigawatts, figures that make the Meta agreement look like one step in a much larger expansion rather than a one-off.
Analysts watching the sector frame deals like this as a structural shift in how AI economics work. The competitive advantage is migrating toward whoever can assemble electricity, land, and hardware at speed, which helps explain why a developer most people had never heard of two years ago is now negotiating directly with one of the world's largest technology companies.
What's Next
The immediate question is timing. Bloomberg reported that it remains unclear when the Childress and Warrenton capacity will actually come online, and power availability is the variable most likely to dictate the answer. Grid interconnection queues, transmission upgrades, and local generation all sit on the critical path, and they are exactly the bottlenecks that have slowed even the best-funded projects across the industry.
Crusoe has shown it can move fast when the pieces line up. It recently brought the first phase of a 1.2-gigawatt campus in Abilene, Texas, online just a year after construction began, and a separate 1.8-gigawatt project in Wyoming is already underway, pushing its planned and operational capacity past 3.4 gigawatts. If the company can repeat that pace in Texas and Missouri, the Meta capacity could arrive faster than skeptics expect.
The wider pattern to watch is the wave of gigawatt-scale power agreements expected through 2026 and 2027, concentrated in high-demand grid regions such as Texas and the Southeast. Microsoft has already disclosed a multibillion-dollar backlog of cloud orders it cannot fulfill because of power constraints, a reminder that signing a capacity deal and energizing it are two very different milestones. How quickly developers and utilities can close that gap will shape the next phase of the AI race.
Closing Thoughts
Strip away the specifics and the Meta-Crusoe deal tells a simple story about where the AI industry has arrived. For several years the narrative centered on who had the smartest model. Increasingly, the decisive question is who can secure enough electricity, data-center space, and computing hardware to keep those models running at scale. Gigawatts have quietly become one of the most valuable currencies in technology.
That reframing has consequences well beyond any single contract. It elevates energy policy, grid capacity, and construction speed into core competitive variables for software companies, and it gives specialist infrastructure developers unexpected leverage in a market once dominated by a handful of cloud giants. Whether this build-out proves prescient or excessive will depend on how durable AI demand turns out to be, but for now the direction of travel is unmistakable, and the numbers keep getting larger.
한글 요약
메타가 데이터센터 개발사 크루소(Crusoe)와 새 계약을 맺고 약 1.6기가와트 규모의 AI 컴퓨팅 용량을 확보했다. 블룸버그 보도에 따르면 이 용량은 텍사스주 칠드레스와 미주리주 워런턴에 들어설 두 시설에서 공급되며, 구체적인 금액과 가동 시점은 공개되지 않았다. 1기가와트가 미국 가정 약 75만 가구에 전력을 공급할 수 있는 규모라는 점을 감안하면, 이번 계약은 AI 산업의 경쟁 축이 '모델'에서 '연산 용량 확보'로 옮겨가고 있음을 보여주는 상징적 사례다.
이번 계약은 메타가 향후 3년간 미국 인프라에 약 6,000억 달러를 투자하겠다고 밝힌 큰 그림의 일부다. 2026년 5대 하이퍼스케일러의 인프라 지출은 약 6,900억 달러에 이를 전망이며, 미국 데이터센터 전력 수요도 2024년 약 50기가와트에서 2026년 76기가와트 수준으로 늘 것으로 예상된다. 이제 병목은 반도체가 아니라 전력이며, 메타처럼 자체 건설을 기다리기보다 크루소 같은 전문 개발사의 용량을 사들이는 방식이 속도를 확보하는 전략으로 부상하고 있다.
2018년 설립된 크루소는 한때 버려지는 천연가스로 채굴 사업을 하다 AI 인프라로 방향을 틀어 가장 빠르게 성장하는 기업 중 하나가 됐다. 지난해 100억 달러 이상 기업가치로 약 13억 7,000만 달러를 조달했고, 마이크로소프트·오라클·오픈AI를 고객으로 두고 있다. 이미 4.9기가와트 규모 계약과 40기가와트 이상의 프로젝트 파이프라인을 확보한 만큼, 이번 메타 계약은 더 큰 확장의 한 단계로 볼 수 있다. 다만 전력망 연결과 송전 등 실제 가동까지의 과제가 남아 있어, 용량이 언제 현실화될지가 다음 관전 포인트다.
참고 / 출처: Bloomberg, Tech Startups, Reuters/Yahoo Finance