Power First: Why AI Data Centers Now Go Where the Grid Is Ready

Childress, Texas, is set to become home to a gigawatt of AI computing. On July 15, Crusoe and Lancium signed a binding agreement to build a 1-gigawatt AI data center campus there, with construction expected to begin in the third quarter of 2026, according to Bloomberg Law. The division of labor is the telling part. Lancium will own the land, handle the power interconnection and manage energy for the site. Crusoe will design, build and operate the data center.

That split reflects a shift that has been building all year. The scarce resource in AI infrastructure is no longer only chips. It is land with power already attached, and the companies that control it are becoming central partners in the AI buildout. For enterprise leaders, this matters because it shapes where AI capacity will exist, who will operate it and what it will cost to reach.

Why are AI data centers being built where the power is?

A conventional enterprise data center could be sited near offices, fiber routes or customers, then connected to the local grid. AI campuses do not work that way. At hundreds of megawatts or a full gigawatt, they need the electrical demand of a small city. Grid connections at that scale can take years to secure, so developers increasingly start with sites where power is available and build the computing around it.

The Childress deal puts that logic into its contract structure. Lancium's role, owning land, interconnection and energy orchestration, is the part that is hardest to replicate. Financial terms were not disclosed, Bloomberg Law reported.

Physical density is part of the reason. "A thousand watt GPU needs a thousand watts of cooling," Steve Heibein, federal AI chief technologist at Hewlett Packard Enterprise, said during a AI Connect webinar on AI and high-performance computing. He noted that older data centers may allocate as little as 10 kilowatts per rack, far below what dense AI servers require. Retrofitting older buildings only goes so far, which pushes new capacity toward purpose-built campuses.

Who is building the new AI campuses?

Earlier deals this year show the same pattern: specialist developers, often with power secured first, signing long leases with AI companies and cloud providers.

• In May, Digi Power X signed a 10-year, $1.1 billion agreement with AI chipmaker Cerebras for a 40-megawatt campus in Columbiana, Alabama, according to Rallies. The report said the company owns the land, has completed an on-site substation with a grid connection and has a power delivery agreement with Alabama Power.
• Also in May, Hut 8 signed a lease worth about $9.8 billion with an unnamed technology client for 352 megawatts at its Beacon Point campus in Nueces County, Texas, according to a Silicon UK report. The site has up to 1 gigawatt of power provision.
• Later that month, Applied Digital signed a 15-year lease valued at about $7.5 billion with an unnamed investment-grade US hyperscaler for its Polaris Forge 3 campus, designed for 300 megawatts of critical IT load, InforCapital reported. The deal pushed the company's contracted AI capacity above 1 gigawatt.

These are long commitments, 10 to 15 years, and several involve tenants that are not named. That suggests the largest buyers are reserving capacity well ahead of need and prefer not to reveal where.

How does the supply chain fit in?

The rush for power-ready land is the last link in a chain that runs back to the chip factories. This week, TSMC raised its capital spending budget and announced another $100 billion for manufacturing in Arizona, as covered in our analysis of TSMC's results. ASML said memory makers are planning several new megafabs that will come online in phases, according to its July 15 investor call.

Each of those chips needs a powered, cooled building to run in. When chip output rises faster than grid connections can be approved, sites like Childress, where land, power and an operator are already lined up, become the bottleneck that sets the pace.

What does the power-first buildout mean for enterprises?

Few enterprises will ever lease a gigawatt campus. But the trend reaches them in several ways.

Capacity concentrates with a few operators. Much of the new AI capacity will be owned by specialist developers and leased to hyperscalers and AI labs on long terms. Enterprises will mostly reach it through cloud services, so their access depends on the commitments those providers make. Locking in reserved capacity for production AI workloads may matter more than chasing the lowest spot price.

Location becomes a governance question. AI computing is moving toward wherever power is available, which may not be where an enterprise expected its data to be processed. Organizations with data residency rules, or regulated workloads, should ask providers where inference actually runs. The same concerns drive interest in sovereign AI, where control over infrastructure and data location is the point.

On-premises plans need facilities planning. Companies considering their own AI infrastructure, such as the AI factory model described by NVIDIA and HPE, have to check whether their buildings can supply the power and cooling. Colocation at a power-ready campus may be the faster path.

Energy costs show up in AI pricing. Developers that control cheap or flexible power have a cost advantage, and over time that will show up in what providers charge. Enterprises negotiating multi-year AI contracts should ask how energy costs are passed through.

What to watch next

The Childress project is scheduled to break ground this quarter. The questions now are how fast power-first developers can turn announced gigawatts into energized buildings, and who signs on as tenants. For enterprise leaders, the practical lesson is that AI capacity is being allocated years in advance, in places chosen for their power supply. Planning for AI at scale now includes knowing where that capacity sits and who controls it.

Frequently asked questions

What is the Crusoe and Lancium Childress project?

A binding deal to build a 1-gigawatt AI data center campus in Childress, Texas, with construction expected to start in Q3 2026, according to Bloomberg Law.

Why are AI data centers built where power is available?

Large AI campuses need city-scale electricity, and grid connections can take years, so developers start with sites where power is already secured.

How does this affect enterprises?

Most enterprises reach this capacity through cloud providers, so reserved capacity, data residency and energy costs in contracts become more important.

Filed under: Data & AI Infrastructure, AI Strategy & ROI