Meta’s Louisiana target exceeds 5GW of compute. The components that will decide whether it exists may never enter the server hall.

Power procurement has moved into the product

Meta’s additional $40 billion commitment in Louisiana arrived alongside a target for more than 5GW of compute. Google, independently, contracted the full initial 1.6GW output of the Steel River Energy Center solar project in Arkansas. One company is defining compute capacity in utility-scale terms; the other is reserving the output of a generation project.

Meta’s Louisiana compute target
Google’s contracted initial Arkansas solar output

Different strategies are converging because the constraint is shared.

At this scale, electricity cannot remain a back-office purchase made after the computing plan is settled. The power pathway determines whether the computing plan is credible in the first place. Site selection, generation and compute deployment become one design problem.

That changes the competitive unit. A conventional data-center plan can treat the building, servers and electricity contract as separable inputs. A multi-gigawatt campus cannot. Its usable capacity is the intersection of secured power, finance and permission to build. Without any one of them, planned compute stays on a slide deck.

The rack has acquired a capital structure

Meta has raised $62 billion of debt since 2022, with roughly half raised in 2025. It also moved $30 billion of debt used to build AI data centers off its balance sheet through special-purpose vehicles.

Hyperion makes the architecture more explicit. Meta is reportedly set to secure an almost $30 billion financing package for the Louisiana data center while retaining a 20% ownership stake in the project. Those terms are not merely a larger version of ordinary capital expenditure. They separate the operator’s use of infrastructure from full ownership of the financed asset.

That separation is what project finance is for: turning one enormous industrial build into an investable structure with its own capital stack. When compute ambitions exceed what companies will carry directly on their balance sheets, financial engineering becomes infrastructure engineering.

The competitive asset is no longer the data center. It is the assembled right to finance, power and operate the campus.

Financing did not follow a completed computing system. It evolved with the scale of the system being attempted.

Long lead times are not delivered electricity

Google’s contract should not be confused with near-term operating capacity. The Arkansas solar project is not expected to operate until 2029. A power contract is not an electron.

The 2029 date blocks a common shortcut: announced procurement is not available compute. Contracted capacity, financed capacity, permitted capacity and operating capacity are different states. Collapsing them produces impressive numbers with limited practical value.

The delay also reveals why advance procurement has become strategic. When the required power asset takes years to arrive, securing its future output must happen well before the corresponding electricity can be consumed. The long lead time pulls energy planning forward into the core AI build cycle.

xAI marks the other hard boundary: buildout speed collides with permitting and community constraints. Capital can accelerate equipment purchases; it cannot make local permission optional.

Utility-scale infrastructure inherits utility-scale clocks. Compute deployment is now governed by the slower institutions surrounding industrial power: financing, project construction, permits and community acceptance. Builder speed remains useful. It is simply no longer sovereign.

The system is assembled, not wholly owned

Meta’s reported 20% Hyperion stake and Google’s full-output contract point to a shared design principle: an operator can secure critical capacity without owning the entire asset. Project vehicles organize capital; procurement contracts reserve power. The operator’s task is to lock the interfaces together.

The repeated move is not “build a bigger data center.” It is to assemble the conditions under which one can exist. Computing, electricity procurement, real estate, structured finance and public approval are no longer adjacent functions; together they determine usable capacity.

Meta’s 5GW-plus target is labeled compute, but its decisive components sit outside the server hall. At that scale, the power contract, financing vehicle and permit are not supporting documents for the machine. They are the machine.