Tight. 3 of 4 warning signs are red, 1 is mixed. Confidence: high.
This page tracks two linked problems: getting connected (how long new data centers and new power plants wait to plug into the grid) and having enough power (whether the grid will have enough supply in the years those data centers come online).
In Texas, large data centers in the grid operator's first connection batch get their power allocated year by year over 2028–2032, and projects not in that batch can't apply until summer 2027. In PJM, the biggest US grid, power plants that started running in 2025 took 7+ years from request to operation. PJM's last two capacity auctions both fell short of the reliability target, at the price cap. The national grid watchdog rates both PJM and Texas "high risk" for supply shortfalls from 2029. Prices are the one mixed sign: high and capped, but not clearly rising.
Outlook: this is the slowest bottleneck to fix. New power plants need turbines that are booked to 2031, and power lines take years to permit.
New here? Skip to Background below for the basics. Coming back? Start with What changed.
This month
October 2026 · Updated Oct 4 · Next update after power producers' Q3 results
What changed since last month
First reading. From next month this section leads with what moved and why.
Scorecard
Warning sign | What we saw | Reading | Source |
Long waits (data centers getting connected) | ERCOT, the Texas grid operator, now connects large users in batches. Eligible projects in the first batch (Batch Zero) get power allocated year by year over 2028–2032. Projects not in that batch can't apply until the next batch opens in summer 2027. | 🔴 Yes | |
Demand above supply (PJM capacity auctions) | PJM's two most recent capacity auctions both bought less supply than its reliability requirement: 6,623 MW short for 2027/28 and 6,831 MW short for 2028/29. These are the first two auctions in PJM's history where the whole region fell short. PJM's CEO: "Demand for electricity continues to grow faster than electricity supply." | 🔴 Yes | |
Prices going up (capacity and power contracts) | No confirmed increase. PJM capacity prices cleared at the cap in both auctions (about $555 uncapped), but the cap itself fell 2.5%, so prices are pinned high, not rising. Long-term power contract prices for clean energy (PPAs) were mixed in Q2: solar down 4.8% from the prior quarter, wind up 5.5%. | ⚖️ Mixed | |
New capacity not enough (independent outlook) | Even counting planned new power plants, the North American Electric Reliability Corporation (NERC), the grid's reliability watchdog, rates PJM and ERCOT "high risk" of not having enough supply from 2029. MISO is high risk from winter 2028. NERC expects 10-year summer peak demand to grow 224 GW, driven mainly by data centers. New supply is slow to arrive: in PJM, power plants that started running in 2025 took 7+ years from connection request to operation. | 🔴 Yes |
Score: 3 red, 0 green, 1 mixed, 0 unknown = 🟠 Tight. Severe needs all 4 confirmed red, and a mixed sign isn't scored either way. Every sign has recent data, so confidence is high. The price sign turns red only if a representative measure confirms a rise, such as a higher clearing price in the next PJM auction or a broad power-contract price index rising across regions. A single data center deal isn't enough, because its contract terms or location can explain the price.
Dates to watch
Nov 4–9 (expected): Q3 results from Talen (Nov 4), NRG (Nov 5), Vistra (Nov 6) and Constellation (Nov 9). Listen for new data center power deals and prices.
Fall 2026: FERC decision on PJM's emergency "backstop" plan to buy more power.
By Dec 2026: Texas regulators' new rules for how large users like data centers connect.
Around year-end (expected): NERC's 2026 Long-Term Reliability Assessment. Could move "New capacity not enough".
Spring 2027: ERCOT tells Batch Zero data centers how much power they get, and when. Could move "Long waits".
"Expected" dates come from earnings calendars. The company hasn't announced them yet.
Status history
- Oct 2026: 🟠 Tight (3 red, 1 mixed), high confidence. First reading.
Background
Reference. Updated when the facts change, not every month.
What it is
A data center can't just plug into the wall. Two things have to be true:
- It has to get connected. The data center asks the local utility or grid operator for a connection. Then it waits while they study whether the grid can handle it and build any new lines or substations needed. Power plants go through their own, separate connection line before they can sell electricity. Both lines are called interconnection queues. They are measured differently and shouldn't be mixed up.
- There has to be enough power. Even once connected, the grid needs enough power plants available at the busiest hour of the year. If demand grows faster than new supply, the grid falls short of its safety margin.
Where it sits: Power plants (power-plant connection queue) → high-voltage grid (run by grid operators like PJM and ERCOT) → data-center connection approved and built → substation → switchgear → servers
A big AI campus can use as much power as a city. The grid has to have that much spare, and right now it often doesn't.
When something is short, two things happen. Whoever already has power to sell can charge more. Whoever needs power waits. Same shortage, opposite effect on the stock.
Why can't they just build more?
- New power plants take years. Gas turbines are booked to 2031. Nuclear restarts and new plants take longer still. See Gas turbines & on-site power.
- Power plant projects slip. In PJM, permits are the biggest single cause of delays for power plant projects in development, at 29% of milestone delays. Supply chain problems cause 23%.
- The equipment is short too. Substation transformers now take more than 160 weeks, up from about 50 in 2021. See Power transformers.
- The queue is clogged with maybes. Developers file requests in several places for the same project, which makes the line look longer and slows studies for everyone. ERCOT's switch to batch studies is meant to fix this for large users.
How big is it?
Grid | What's happening |
PJM (13 states, from Illinois to Virginia) | Short 6,831 MW of its reliability requirement for 2028/29, after 6,623 MW short for 2027/28. Capacity prices at the $325/MW-day cap (about $555 without the cap). These are payments for being available in 2028/29, not today's electricity price. |
ERCOT (Texas) | About 474 GW of large users seeking connection as of June 2026, about 90% data centers. That's up from 226 GW in late 2025. About 205 GW is preliminarily eligible for Batch Zero. Treat these totals as a ceiling: many requests are speculative or duplicates. |
Dominion (Virginia utility, inside PJM) | 53.8 GW of data center capacity under contract as of July 2026. Only 12.0 GW is at the firmest stage (signed electric service agreements). |
For scale, 1 GW is roughly the power used by a mid-size city.
Who wins, who waits
Company | Effect | Why |
Wins | Power producers with existing plants. Capacity prices pinned at the cap, and data centers willing to pay up for guaranteed power. | |
Wins | On-site power. Data centers that can't wait for the grid build their own with fuel cells, turbines or engines. | |
Wins | They build the power lines and substations the grid needs. Years of work ahead. | |
Wins | Already holds grid-connected sites. In a world where power is the queue, a site with power attached is the scarce asset. | |
Mixed | GPU cloud companies. Sites they already have gain value, but new capacity waits on power. | |
META, Microsoft, Google, Amazon | Waits | Their AI plans need gigawatts. Power, not chips, is increasingly what limits where and how fast they build. |
Households and businesses in PJM | Waits | High capacity payments flow into electricity bills, which is creating political pushback against data centers. |
The takeaway: grid access is turning into the scarcest resource in the AI build-out. Companies that already own power, or can make their own, hold the advantage.
When does it end, and what could replace it?
- Near term: PJM's emergency backstop purchase of extra power, new connection rules in Texas by December 2026, and ERCOT's first batch allocations in spring 2027.
- 2027–2031: new gas plants, limited by turbine supply. Nuclear restarts, some of which may wait into the next decade for line upgrades.
- Workarounds: on-site generation ("behind the meter"), building next to existing power plants, and moving to regions with spare power.
- Political risk: rising bills could bring rules that make data centers pay more or wait longer.
How to read a change
Good easing
Connection times shorten because new plants and lines came online, while data center demand stays strong.
Bad easing
Queues shrink because data center projects got cancelled or AI spending slowed. Power producers' pricing would weaken first.
What would change our view
- Gets tighter: the next PJM auction falls short again at the cap, ERCOT's Batch Zero allocations push most power into 2030+, or new rules slow large-user connections.
- Gets easier (good): a PJM auction meets its reliability requirement below the cap, NERC lowers PJM or ERCOT from "high risk", or faster permitting rules.
- Gets easier (bad): large data center requests are withdrawn from the queues.
Past scorecards
Each month the previous scorecard moves here, newest first.
- None yet. October 2026 is the first reading.
Evidence and sources
This month's scorecard
- Long waits: ERCOT batch process for large users: Batch Zero starts Jul 2026, allocations spring 2027, Batch 1 applications from summer 2027. ERCOT, Jun 18. Year-by-year allocation 2028–2032; 474 GW large-load requests (~90% data centers); ~205 GW preliminarily Batch Zero eligible. ERCOT Senate testimony, Jul 29
- New capacity not enough (supporting): PJM projects entering service in 2025 averaged 7+ years (3+ to interconnection service agreement, 4+ after). Permitting 29% and supply chain 23% of milestone change requests. Transformer lead times over 160 weeks. Data Center Knowledge, May 12. Note: the article calls these "AI infrastructure projects", but interconnection service agreements and milestone change requests belong to PJM's power-plant (generation) connection process, so this page treats the figure as a power-plant measure.
- Demand above supply / prices: 2028/29 auction (Jul 14, 2026): $325/MW-day at the cap, down 2.5% from the prior cap of $333.44; 6,831 MW short; the latest auctions are the first in PJM history where the whole region fell short. PJM, Jul 14. 2027/28 auction (Dec 17, 2025): $333.44/MW-day at the cap; 6,623 MW short. PJM, Dec 17 2025. Uncapped price ~$555 and $16.4B auction value. PA Environment Digest
- Prices: LevelTen North American PPA Price Index Q2 2026 (Jul 21): solar -4.8% QoQ, wind +5.5% QoQ and +17.5% YoY; data center buyers still procuring at scale. LevelTen
- New capacity not enough: NERC 2025 Long-Term Reliability Assessment, released Jan 29, 2026: PJM, ERCOT, WECC-Basin and WECC-Northwest high risk from 2029; MISO from winter 2028; 10-year summer peak growth of 224 GW. POWER Magazine · NERC report
Background
- Dominion contracted data center capacity 53.8 GW (12.0 GW at signed electric service agreements), Jul 2026. Investing.com, Q2 2026 slides
- ERCOT large-load queue 63 GW → 226 GW in 2025; Texas rules due Dec 2026. Latitude Media, Dec 3 2025
- GE Vernova taking turbine reservations for 2031. Utility Dive, Jul 23
- Berkeley Lab's power-plant queue dataset excludes large-user (load) connections, which is why the two queues are tracked separately here. Berkeley Lab, Queued Up 2026
- Where sources disagree: queue totals overstate real demand, because many requests are speculative or duplicates. That's why queue size isn't scored. Grid Strategies argues NERC overstates reliability risk. Utility Dive. PJM's auction shortfall, though, is measured against real auction results.
Words used on this page
- Interconnection queue: the waiting line to connect to the grid. Power plants (supply) and large users like data centers (load) have separate queues.
- Large load: a big electricity user, such as a data center, that needs its own grid study before connecting.
- Batch study: studying a group of connection requests together instead of one at a time, so each new request doesn't reset everyone else's study.
- Interconnection service agreement (ISA): the contract that lets a power plant connect to PJM's grid once studies are done.
- PJM: the grid operator for 13 eastern and midwestern states, the largest in the US.
- ERCOT (Electric Reliability Council of Texas): the grid operator for most of Texas.
- MISO (Midcontinent Independent System Operator): the grid operator for much of the central US.
- NERC (North American Electric Reliability Corporation): the watchdog that sets reliability rules and forecasts shortfall risk.
- Capacity auction: where PJM pays power plants to be available in future years. It's a payment for availability, not the price of electricity today.
- Price cap: the maximum price the regulator allows in PJM's capacity auction. Clearing at the cap means buyers would have paid more.
- Reliability requirement: the amount of supply, plus a safety margin, that a grid needs to keep the lights on in extreme weather.
- Resource adequacy: whether a grid has enough supply to meet demand at its peak.
- PPA (power purchase agreement): a long-term contract to buy electricity from a specific power plant at a set price.
- MW-day: the price unit for capacity, dollars per megawatt per day.
- GW / MW (gigawatt / megawatt): units of power. 1 GW = 1,000 MW.
- Behind the meter: power made on-site rather than taken from the grid.
- FERC (Federal Energy Regulatory Commission): the US regulator for interstate power markets.
Invest Deeper · Educational research, not investment advice.
Made with Bullet