Living report / May 2026 version

AI Power Demand 2026: The Grid Reality Check

How much AI load is actually hitting the system, where it is landing, and what filings, forecasts, and utility signals show versus the hype.

View regional table Get updates

ERCOT 2032 preliminary load forecast

367.8 GW

A queue-quality signal, not a guaranteed demand outcome.

Source: ERCOT long-term load forecast references

PJM new resource timing concern

4+ years

PJM says new gas capacity can take at least four years from investment decision to operation under optimistic assumptions.

Source: PJM market-design materials

Virginia commercial electricity sales growth

84M to 114M MWh

EIA says Virginia commercial sales rose sharply from 2019 to 2024, with data centers driving much of the increase.

Source: EIA Today in Energy

Typical U.S. household benchmark

~899 kWh/mo

A useful baseline for explaining how grid costs become visible to ordinary customers.

Source: EIA residential electricity data

Executive summary

The useful question is no longer whether AI needs power. It is which demand is real enough to plan around.

The strongest evidence is not a press release. It is a utility forecast revision, interconnection study, power contract, rate case, RTO queue signal, or electricity-sales data.

Announcements matter, but they are not all equal. A 5 GW ambition without power path is not the same thing as energized load.

Regional concentration matters more than national totals. A grid can absorb annual energy growth more easily than it can absorb a fast block of load at one constrained node.

The money question is shifting from chips to capacity: who pays for generation, transmission, substations, reserves, and risk when AI projects move faster than the grid buildout?

Regional reality table

Where AI power demand is becoming a grid-planning problem

AI Power Demand hub

Region / market	Current signal	Confidence	What to watch next
PJM	Large-load growth, capacity-market timing, and affordability	Medium-high	Track capacity auctions, expedited interconnection proposals, and utility queue updates.
ERCOT	Very large preliminary load forecast and data-center request stack	High but uncertain	Separate credible energized load from speculative queue requests.
Virginia / Dominion territory	Commercial electricity sales already show data-center pressure	High	Watch local transmission, substation, and rate-case pressure.
Oklahoma / Southwest Power Pool	Large-load contracts and cost-causation policy are becoming explicit	Medium	Track tariff design, Google/utility contracts, and ratepayer protections.
MISO	Queue reform, large-load treatment, and generation adequacy pressure	Medium	Watch zero-injection and large-load interconnection debates.

Announced vs. real

The AI load conversion funnel

This is the framework The Grid Report will use to judge whether a data-center power story is hype, early signal, or real grid demand.

Stage	What it means	Evidence strength	Why it matters
Announcement	Company says a campus, GW target, or AI capacity deal exists.	Low	Useful signal, but not enough for grid planning by itself.
Site control	Land, zoning, and local incentives start to appear.	Medium-low	Better than press language, but still not energized capacity.
Utility/RTO process	Interconnection, load study, tariff, or queue data appears.	Medium	The project begins to touch the power system.
Contract / construction	PPA, large-load contract, generator order, substation work, or construction activity.	Medium-high	This is where capital and timing risk become visible.
Energized load	Power is delivered and appears in electricity sales/load data.	High	The cleanest proof that announced demand became real grid demand.

Source watchlist

What we will monitor

Source	Watch for	Why it matters
EIA	Electricity sales, prices, generation, natural gas, STEO, Today in Energy	Turns hype into national and regional energy data.
PJM / ERCOT / MISO / SPP / CAISO	Load forecasts, queue updates, market design, large-load policy	Shows whether AI demand is becoming a reliability and capacity-market issue.
FERC / NERC	Large-load rules, transmission, reliability alerts, interconnection policy	Defines the regulatory and reliability frame around rapid load growth.
Utility IRPs and earnings calls	Forecast revisions, capex, rate cases, load-pipeline comments	Shows which utilities are actually planning around data-center demand.
Hyperscaler filings and energy reports	PPAs, capacity deals, emissions reports, capex language	Separates public AI ambition from power-readiness evidence.

What makes this report link-worthy

The dataset is the product.

A regional table readers can cite.

A clear methodology that separates announcements from real load.

Primary-source links, not vague market chatter.

A fixed refresh promise so the page becomes a reference asset.

Spinoff stories that keep pointing back to one canonical report.

Next refresh

The next version should add the downloadable table.

Version 2 should convert this framework into a small open dataset: region, utility/RTO, project or signal, reported MW/GW, status, source, date, confidence, and Grid Report note.