Power politics now sets the AI data center calendar
Front-end costs and approval queues, not capital, now decide where hyperscale computing gets built.
Amazon Web Services is adding a $6 billion campus in Shreveport, lifting its Louisiana commitment to $18 billion, according to PWD's tracking. The company will pay for the water and grid connections itself. That is one answer to the question hanging over every AI data center: who writes the check for the wires before the electrons arrive.
The check for the wires
NextEra's $3.3 billion gas buildout offers another. U.S. and Japanese government capital stands behind the projects, and the agreements shift early-stage costs onto large-load customers. The structure suggests residential ratepayers are not the source of early funding. The load anchor is.
Both answers point the same direction. The party with the longest time horizon absorbs the front-end risk. AWS can wait for its Louisiana campus to pay out. A sovereign fund can wait for a Japanese data center to fill. The traditional utility compact, where the broad customer base socializes the cost of new connections, is not the default anymore.
The approval is the asset
Policy is now the binding constraint, and it moves on a calendar that money cannot accelerate. New South Wales will require data centers to buy wind power equal to 40% of their electricity before approval. The wind purchase becomes a permit condition, not an afterthought. California lawmakers have advanced bills targeting data center rates and pollution, adding a legislative variable to site selection.
Virginia shows what that variable costs. JK Land has refiled its Gainesville data center plan for an 82-acre campus. The project needs a special use permit, and the wait for power is five to seven years. That is not a construction timeline. It is the time required to secure approval and power, and it is built into the pro forma.
The result is that a data center's value is less about the building and more about the position in the queue. Capital is abundant; queue position is not. A developer can raise billions and still wait a half-decade for permission to use them. That favors balance sheets large enough to hold land and permits without pressure to earn a return next quarter.
Balance sheets as ballast
The largest risk transfer so far sits in the OpenAI-Nvidia relationship. Nvidia has guaranteed the first 4.25GW of OpenAI's Ohio data center, putting its own balance sheet behind a buildout reported at 10GW scale. Nvidia's name on the guarantee does not just secure chips; it secures the power. That is vendor financing at utility scale.
The Japanese commitments are smaller but share the shape. Mubadala is weighing a ¥1 trillion commitment to a 500MW Akita data center. The project would take seven years, with operations starting in 2033. At Goodman's Tsukuba campus, a hyperscaler has signed a 20-year lease for 50MW in the first phase of a 1GW site. Service is expected in early 2028. Both deals are real. Both put revenue years into the future.
These deals also tell private investors what to watch. The return on a data center investment is no longer a function of the lease rate alone. It is a function of the date the power arrives. A 20-year lease with a hyperscaler is just paper until the switch is flipped. The gap between signing and energizing is where the risk lives.
That gap is widening under the new policy regime. In New South Wales, the wind requirement means a PPA must be secured before approval. In Virginia, the five-to-seven-year wait for power is a timeline measured in years, not months. In California, the legislative review adds an unpredictable layer. The costs of delay land on whoever committed first.
The near-term result is more structure, not less. AWS pays its own hookup. NextEra's large-load customers carry early-stage gas costs. Nvidia stands behind OpenAI's electrons. Mubadala is prepared to wait seven years for Akita. These are not ordinary construction spreads; they are the terms of entry to a scarce resource.
For the managers who allocate to infrastructure and private credit, the underwriting question has shifted. It is no longer 'how much per megawatt,' but 'when does the megawatt arrive.' The buildout is a calendar story dressed up as a capital story. The investors who get paid first will be the ones who asked about the date.
The buildout is a calendar story dressed up as a capital story.