The AI Buildout Has Entered Its Permission Era

Key points

• The International Energy Agency projects global data-center electricity use rising from about 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030, while warning that grid access, planning, and public acceptance can slow delivery. Sources: S1
• FERC reports that the average data center entering service grew from about 25 megawatts in 2020 to almost 80 megawatts in 2025; projects planned for 2030 average more than 400 megawatts, making accurate load forecasts and large-load tariffs more important. Sources: S2
• Texas ordered a state audit of proposed data centers after the governor’s office cited roughly 474 gigawatts of connection requests, about 90% tied to data centers. That is requested capacity, not a count of projects that will be built. Sources: S3
• CBRE reports record-low vacancy in major data-center markets even as power limits, long grid waits, and local opposition restrict new supply. Strong demand does not remove the need for permits, infrastructure, or community support. Sources: S4

Money is abundant; permission is scarce

The AI infrastructure race is often described as a contest over capital and chips. Those inputs still matter, but they do not guarantee a working data center. A large site also needs a believable power plan, a grid connection, land, permits, water arrangements where cooling requires them, and a community willing to accept the project’s costs and benefits.

The IEA expects global data-center electricity use to nearly double between 2025 and 2030. It also says planning rules, regulatory systems, grid connections, and social acceptance can become bottlenecks. This is a forecast, not a fixed outcome. It shows the size of the coordination problem if current projects move forward.

Sources: S1, S3

The grid needs projects it can believe

FERC’s market report shows how quickly individual requests are growing. The average data center entering service was about 25 megawatts in 2020 and almost 80 megawatts in 2025. Projects planned for 2030 average more than 400 megawatts. A few uncertain requests can therefore change a utility’s view of future demand.

A connection queue is not the same as a construction schedule. Developers may submit overlapping requests, change a project’s size, move it, or never build it. FERC says grid planners need better information about when a large load will arrive and how much power it will actually use. Special tariffs can also decide which costs belong to the developer and which might reach other customers.

Sources: S2

Texas turns the infrastructure question into a public test

Texas Governor Greg Abbott ordered the Public Utility Commission and ERCOT to audit proposed data centers before new projects move ahead. His office cited about 474 gigawatts of connection requests and said roughly 90% came from data centers. The order asks for project ownership, subsidies, power use, water use, and community effects.

That 474-gigawatt figure is a claim about requests in a queue. It is not operating demand, approved construction, or a forecast that every project will be built. The audit matters because it may separate serious projects from speculative ones and make their local tradeoffs easier to inspect before the grid is expanded around them.

Sources: S3

The next scorecard must measure delivery

CBRE describes a market with very low vacancy and strong demand, but also limited power, long development timelines, and local opposition. Northern Virginia and Atlanta reached vacancy rates near or below 1% in its 2026 market review. That supports the case for more capacity, but not for every proposed site in every location.

OMIKINA will track signed power agreements, interconnection milestones, tariff terms, permits, water plans, construction progress, and the share of capacity that reaches service. Capital announcements remain useful signals. The stronger proof is a project that can operate without hiding costs, overstating demand, or treating public consent as an afterthought.

Sources: S2, S3, S4

Why it matters

AI demand is moving from software forecasts into power systems, utility planning, water reviews, public budgets, and local politics. If load requests are weak or costs are shifted to other customers, communities and grids can absorb risk before a model serves one user. A credible buildout needs evidence that the project, power, financing, and public bargain are all real.

Sources: S1, S2, S3, S4

Sources

1. Key Questions on Energy and AI: Executive summary — International Energy Agency · 2026-04-16
2. 2025 State of the Markets Report — Federal Energy Regulatory Commission · 2026-03-24
3. Governor Abbott Directs Comprehensive Data Center Audit — Office of the Texas Governor · 2026-08-03
4. Global Data Center Trends 2026 — CBRE · 2026-06-17