Abilene, Texas was mesquite shrubland not long ago. Today it is on its way to hosting 2.1 gigawatts of AI computing capacity, enough to power a mid-sized city, spread across what will be ten data center buildings on the same sprawling campus. OpenAI and Oracle built the first two. Crusoe, the data center company doing the construction, is finishing six more for them. And now, on adjacent land that OpenAI chose not to expand into, Microsoft is building two more buildings of its own, with a dedicated 900-megawatt power plant to run them. Microsoft still owns roughly 27% of OpenAI. They are also now its neighbor, building independently, on land OpenAI walked away from.

That detail is worth sitting with. When the Abilene campus was first announced as part of the Stargate initiative early last year, it was framed as a flagship moment for American AI investment. OpenAI, Oracle, SoftBank, and Microsoft all in the same orbit. Since then, OpenAI has said it will not expand the Abilene site further and is putting additional capacity in other locations. Microsoft moved into the gap. They are not leaving Abilene. They are just no longer going there together.

What Crusoe Is Actually Building

Crusoe started as a company that used stranded natural gas from oil fields to power Bitcoin mining. It has pivoted hard into AI infrastructure and is now one of the fastest-moving data center builders in the country. The first two buildings on its Abilene campus were constructed and powered in under a year, which is unusually fast for facilities at this scale. The new Microsoft campus, announced March 27, adds two more buildings plus a dedicated on-site power plant. That plant alone will generate 900 megawatts, more than double the existing 350-megawatt gas-fired plant serving the OpenAI and Oracle buildings next door.

The power plant detail matters. Most large data centers draw from the regional electricity grid and use on-site generation as backup. Crusoe is building Microsoft a facility with its own dedicated power supply built in from the start. That is a design choice driven by the reliability requirements of next-generation AI chips, which draw far more power per rack than anything the industry has previously had to plan around. The grid alone cannot guarantee the consistency these workloads need. So they are building the grid into the building.

The Microsoft-OpenAI Split, Made Physical

For most of the past few years, Microsoft and OpenAI's relationship looked like a tight partnership: Microsoft as the infrastructure provider, OpenAI as the model builder. That framing has been loosening. Microsoft has been building its own AI capabilities, its own models, and its own go-to-market strategy. OpenAI has been raising independent capital and pursuing its own cloud relationships. The Abilene situation makes the divergence concrete in a way that earnings calls and press releases do not.

They are on the same campus but building separately. OpenAI chose not to expand. Microsoft stepped in. Both decisions were made independently. That is not a partnership in the original sense. It is two companies with overlapping interests and increasingly separate strategies, operating in proximity because the infrastructure economics make West Texas a sensible place to build right now, not because they are coordinating.

Power Is the Ceiling for Everything Else

Microsoft's $37.5 billion capital expenditure for the quarter ended December 2025, with roughly two-thirds going to computing infrastructure, reflects a company that has decided the constraint on AI is not software or talent. It is power and physical capacity. You cannot run large AI workloads without somewhere to put the servers and enough electricity to run them steadily. The race to secure that capacity, in Texas, Wisconsin, and data center markets around the world, is moving faster than most enterprise technology organizations have registered.

Crusoe's chief executive framed it plainly in the announcement: the goal is to build the industrial foundation for American AI at a velocity the industry has never seen. The first two Abilene buildings went from cleared ground to operational in under a year. The question for every technology organization watching this is not whether AI infrastructure investment is accelerating. It is whether the companies controlling that infrastructure will also set the terms on which everyone else accesses it.