Developers have announced 190 gigawatts of hyperscale data-center capacity across 777 projects Bessemer Venture Partners. For an enterprise buyer waiting on compute, that sounds like relief; for investors reading utility filings, it looks more like an exceptionally large stack of option value.
Most of it is still vapor. Roughly 148 gigawatts remain planned, while about 21 gigawatts are under construction Bessemer Venture Partners. In practical terms, nearly four-fifths of the advertised capacity has yet to become a building drawing electricity. The GPU may appear in a purchase order long before the substation appears in a field.
On paper.
JLL expects roughly 100 gigawatts of new capacity to come online from 2026 through 2030, representing about $1.2 trillion of real-estate asset value JLL. That is enough property value to attract every infrastructure fund, pension manager and private-equity tourist with a spreadsheet. It is not proof that utilities can energize the sites on schedule.
That’s one way to look at it.
Why can’t capital simply solve the bottleneck? Because money can order turbines, transformers and switchgear, but it cannot instantly create transmission rights, utility studies, permits, skilled crews or local political consent.
The industry spent much of the AI boom discussing accelerator scarcity, then discovered that chips are portable and power systems are not. A rack can be redirected to another campus; an interconnection agreement is tied to a specific place, utility and queue position. And those queues were designed for a slower generation market, not clusters of large loads arriving with hyperscaler deadlines and executive-level urgency.
The demand curve is unpleasant. Global data-center electricity consumption was about 415 terawatt-hours in 2024 and is projected to exceed 945 terawatt-hours by 2030 International Energy Agency. That means the sector is trying to add electricity consumption comparable to that of a major industrial economy before many transmission projects can finish permitting.
A nasty mismatch.
In the United States, data centers consumed about 4.4% of electricity in 2023 and could reach between 6.7% and 12% by 2028 U.S. Department of Energy. At the upper end, more than one dollar in every ten spent generating national electricity would effectively serve server halls rather than homes, hospitals, factories and everything else plugged into the grid. Local effects will be sharper because data-center demand concentrates around fiber routes, available land and existing substations.
You can see where this is going.
Grid access is becoming a distinct asset class, though promoters will call it “strategic infrastructure” because apparently easements need branding. Powered land with a credible energization date should command a premium over land carrying only a developer’s load request. Existing campuses can gain scarcity value even when their cooling design or power density is less fashionable; electricity today often beats theoretical efficiency tomorrow.
But investors should separate four things that slide decks routinely blend together: control of land, a utility service agreement, an executed interconnection position and actual deliverable power. They are not interchangeable. A project can possess the first two while remaining exposed to transmission upgrades, transformer procurement, curtailment terms or a utility’s revised load forecast.
FERC has already pushed transmission providers toward cluster studies, firmer deadlines and readiness requirements intended to clear speculative projects from interconnection queues FERC Order No. 2023. The reforms matter, but they do not manufacture substations—or turn an economically dubious grid upgrade into an attractive one.
This is a mistake: valuing every announced megawatt as if it carries the same probability, timing and margin profile. A campus energized from surplus hydro, nuclear or gas-backed generation is not economically equivalent to one depending on several years of network construction (even if both render beautifully in PowerPoint). The discount rate should reflect electrical reality.
What deserves a premium? Sites with binding utility commitments, transparent upgrade obligations, credible equipment delivery schedules and enough generation diversity to avoid becoming hostage to a single constrained node. Behind-the-meter generation can help, as can long-duration nuclear or renewable contracts, but a power-purchase agreement is not a transmission line and carbon accounting is not dispatchability.
Credit where it’s due—the better operators have stopped treating energy procurement as a real-estate appendix. They are hiring utility veterans, reserving equipment earlier and designing campuses that can energize in phases rather than waiting for the entire load block. That engineering discipline will matter more than another ambitious capacity announcement.
The AI infrastructure race will not be won by whoever circles the most acreage in red. It will be won at the substation gate, where the marketing deck meets a locked fence and a date the utility is willing to sign.