AI + energy
AI Energy Gridlock
Power availability is becoming a product constraint for AI companies.
The IEA projects global data center electricity consumption will exceed 1,000 TWh by 2028 — roughly Japan's total demand. U.S. interconnection queue wait times average 5+ years. AI companies are competing for megawatts, cooling capacity, and zoning permits, not just talent.
Pressure index by operating layer
Signal concentration
Capitalized attention split
Problem to company flow
What changed
Data center power demand is growing 20–25% annually. The IEA's Electricity 2026 report shows AI and crypto workloads will consume more electricity than many mid-sized countries within three years. Interconnection queues in the U.S. now average 5+ years, up from 3.7 in 2022. This is not a compute problem — it's an infrastructure planning crisis that requires policy intelligence, site selection analytics, and real-time grid negotiation.
What leaders should do
Map your power dependency chain: grid capacity, cooling requirements, local permitting timelines, and utility contract structures. Score each facility or region by power availability, cost trajectory, and political risk. Move planning cadence from quarterly to weekly. Build relationships with local governments before breaking ground — community opposition is the hidden constraint that kills 30%+ of new facilities.
What ZOAK wants to build
An AI infrastructure planning layer that scores regions, facilities, workloads, grids, and local policy in one operating dashboard. The product would integrate IEA energy forecasts, utility rate data, interconnection queue status, and zoning risk into a decision-support system for site selection teams.
Operating analysis
AI companies are competing for electrons, interconnection queues, cooling, and local political permission. The global data center market consumed roughly 460 TWh in 2022. The IEA expects this to more than double by 2028. In the U.S., Virginia's "Data Center Alley" already consumes more electricity than some states. New projects in Texas, Ohio, and Georgia face 4–7 year interconnection delays.
The near-term opportunity is not a prettier energy dashboard. It is an operating system that turns fragmented grid data, policy signals, and facility constraints into a repeatable site-selection and capacity-planning workflow.
| Signal | Why it matters | Action |
|---|---|---|
| Grid saturation | IEA projects data centers will consume 1,000+ TWh by 2028 — equal to Japan's total demand. | Score regions by available grid headroom and planned capacity additions. |
| Queue delays | U.S. interconnection wait times now average 5+ years, blocking new facility timelines. | Map queue positions and identify fast-track utility partnerships. |
| Community opposition | Local residents and governments are pushing back on noise, water use, and tax incentive deals. | Build a community impact assessment layer into site selection workflows. |
What would we build first?
A regional grid capacity scoring tool that integrates interconnection queue data, utility rate forecasts, and zoning risk into a single view. Start with the top 20 U.S. data center markets.
What makes this different from existing tools?
Existing solutions track energy prices or data center inventory. None integrate policy risk, community sentiment, and interconnection queue status into a single planning workflow. The gap is operational, not informational.
How would we measure success?
Time-to-decision on site selection should drop by 40–60%. False starts — projects abandoned after community opposition or queue delays — should decrease measurably within the first 12 months of adoption.
ZOAK_BUILD_THESIS = {
category: "AI + energy",
first_principle: "power is a product constraint, not an ops detail",
target_lift: "+64% site selection speed",
next_move: "prototype regional grid scoring tool"
}Sources: IEA Electricity 2026, IEA Key Questions on Energy and AI, Brookings Institution energy analysis
Related engagement
Facing infrastructure planning constraints?
Tell us about your capacity planning problem — we'll scope a diagnostic.
Start a conversation