On June 18, 2026, the Federal Energy Regulatory Commission issued six Section 206 show-cause orders directing every major regional grid operator except ERCOT to justify or reform their large-load interconnection rules within 60 days. The move confirms what hyperscalers have known for two years: the AI data center power grid bottleneck is no longer chips.
It is interconnection, generation, and transmission.
The FERC data center interconnection orders target facilities drawing 20 MW or more, the kind of load that can match the electricity of 15,000 to 20,000 American homes. Operators have until August 17, 2026 to propose reforms, and until July 18, 2026 to file informational reports on current practice.
The binding constraint on AI compute has moved from the fab to the substation, and regulators are now scrambling to reshape the grid around it.
TL;DR. FERC's June 2026 orders force six RTOs to fix large-load interconnection rules, but fast-tracking the queue does not create generation, build transmission, or settle who pays. Hyperscalers are responding with gigawatt PPAs, behind-the-meter fuel cells, nuclear commitments, and geographic flight to underutilized corridors. The grid bottleneck is real, contested in court, and unlikely to clear before the late 2020s.
Key takeaways
- FERC issued six Section 206 show-cause orders on June 18, 2026, covering PJM, MISO, SPP, CAISO, NYISO, and ISO-NE. ERCOT is excluded.
- Monitoring Analytics attributed roughly 40 percent of PJM's $16.4 billion 2027/2028 capacity auction to phantom data center load.
- LBNL projects U.S. Data center electricity rising from 176 TWh in 2023 to 325-580 TWh by 2028, or 6.7 to 12 percent of national load.
- PJM's interconnection queue has grown from under 2 years in 2008 to over 8 years today.
- Behind-the-meter fuel cells deploy in ~90 days and now account for $7.65 billion in signed Q4 2025-Q1 2026 deals.
- Eleven-plus state moratorium bills and active federal litigation threaten the fast-track premise.
What FERC actually ordered (and what it didn't)
The June 18 action is not a single rulemaking. It is six separate Section 206 proceedings, one per RTO/ISO, each under the Federal Power Act's "just and reasonable" standard:
| Docket | RTO/ISO | Status |
|---|---|---|
| EL26-67-000 | PJM Interconnection | Show-cause |
| EL26-68-000 | Southwest Power Pool | Show-cause |
| EL26-69-000 | New York ISO | Show-cause |
| EL26-70-000 | MISO | Show-cause |
| EL26-71-000 | CAISO | Show-cause |
| EL26-72-000 | ISO New England | Show-cause |
Commissioner Chang's concurrences and the Department of Energy's own April 9, 2026 directive confirm the six-order structure. The separate ANOPR at Docket RM26-4-000, opened October 27, 2025, remains the slower track that could eventually produce a generic interconnection reform framework. The June 2026 orders are enforcement-oriented and faster-moving.
The orders do not immediately change any interconnection rule. They direct each operator to explain why its current large-load provisions are just and reasonable, or to propose reforms. FERC's fact sheet frames the action as an effort to "Supercharge America's Grid for Efficiency, Reliability, and a Bold Energy Future."
Why ERCOT is excluded
ERCOT serves roughly 90 percent of Texas load and operates outside FERC's jurisdiction over interstate wholesale power sales. Texas handles its own resource adequacy and interconnection standards. But ERCOT is not idle.
On the same day as the FERC orders, June 18, 2026, ERCOT published a Trending Topics document describing a new Batch Connection Process for Large Electricity Users. Texas is managing the same pressure through state-level mechanisms rather than federal requirements.
How big is the data center electricity demand AI is creating?
The numbers are large enough to reshape national load forecasts.
Lawrence Berkeley National Laboratory puts U.S. Data center electricity at 176 TWh in 2023, 4.4 percent of national consumption, growing to between 325 and 580 TWh by 2028.
That upper bound is 12 percent of all U.S. Electricity. The International Energy Agency forecasts global data center electricity reaching roughly 945 TWh by 2030, a 165 percent increase from 2024 levels.
PJM's own 2025 Long-Term Forecast Report projects 32 GW of peak load growth between 2024 and 2030, with approximately 94 percent attributable to data centers. NERC's 2025 Long-Term Reliability Assessment projects summer peak demand growth of 224 GW over the next decade, 69 percent higher than its prior forecast, and explicitly warns that "projections for resource and transmission growth lag what is needed to support new data centers."
Northern Virginia's Data Center Alley alone is projected to grow from roughly 4,000 MW to 15,000 MW by 2030, which would be about half of Virginia's total electricity load.
The PJM capacity auction data centers broke
The clearest evidence that the data center grid bottleneck 2026 is a market problem, not just an engineering problem, is PJM's 2027/2028 Base Residual Auction.
The auction cleared at $333.44/MW-day, the administrative ceiling, for the second consecutive year, on 134,478 MW of capacity. Total capacity cost approximated $16.4 billion for the delivery year. Monitoring Analytics, PJM's independent market monitor, attributed roughly 40 percent of that cost to data center load, specifically the "phantom" load of projects that had secured transmission service agreements but had not yet entered commercial operation.
For the prior delivery year, the attribution was even starker: 63 percent of the year-over-year price increase was tied to data center demand growth. Capacity auctions clear two to three years ahead of delivery, so the 2027/2028 result reflects developers' expectations about 2027 load, not 2027 reality.
The 40 percent figure represents load share contribution to clearing prices, not a direct cost allocation. All capacity clears at the market-clearing price, which is why the cost-shifting fight is now in federal court.
What does fast-tracking interconnection actually fix?
The FERC orders target procedural delay. They do not target the three structural problems underneath it.
Queue length. PJM's queue timeline has stretched from under 2 years in 2008 to over 8 years today. A hyperscale facility is built in 18-24 months. That mismatch is the core complaint.
Cost allocation. Interconnection costs can be assigned to the data center developer, socialized across ratepayers, or shared. FERC's November 2024 co-location order rejected Talen Energy's attempt to expand its Amazon arrangement, ruling it would shift roughly $140 million in capacity costs to other PJM customers. The June 2026 orders do not settle this question.
Uniformity. Each RTO maintains distinct large-load provisions with different thresholds, study requirements, and timelines. A more uniform framework is the stated goal.
What the orders leave unsolved
The generation shortfall is the biggest gap. PJM's queue at end-2023 held 225 GW of projects against peak demand of 165 GW, yet only 10 GW entered service between January 2022 and September 2024, according to Kirkland & Ellis analysis. Queue position does not equal capacity.
Gas turbine components face 5-7+ year lead times. Large power transformers and EHV equipment face 3-7 year delivery lead times regardless of how fast the study process moves. You can streamline the paperwork and still wait years for the steel.
How hyperscalers are routing around the queue
The dominant 2026 strategy is four parallel tracks, run simultaneously.
1. Gigawatt-scale PPAs. The Brookfield-Microsoft 10.5 GW clean energy framework is the largest corporate clean energy deal ever executed. Meta's nuclear package totals 6.6 GW across Vistra, TerraPower, and Oklo. The AWS-Talen deal is 1,920 MW through 2042, structured front-of-meter to avoid FERC interconnection review entirely. PPA durations have stretched from 10-15 years to 20-25 years.
2. Behind-the-meter fuel cells and gas. Bloom Energy signed $7.65 billion in deals in 90 days across Q4 2025 and Q1 2026, including up to 2.8 GW for Oracle and a $5 billion Brookfield framework. Solid oxide fuel cells run at roughly 60 percent electrical efficiency versus 35-45 percent for gas turbines, and they deploy in about 90 days. Goldman Sachs projects behind-the-meter systems providing 25-33 percent of incremental data center power demand through 2030.
3. Nuclear commitments. Roughly 11 GW of hyperscaler nuclear commitments landed in 18 months. Nuclear PPAs align with 20-year license renewal cycles, giving operators 40 years of capacity visibility.
4. Geographic diversification. Siting now follows substation capacity, not fiber or tax incentives. The Texas I-35 corridor, Wyoming, New Mexico, and secondary PJM markets like Ohio are absorbing load that Northern Virginia cannot.
| Track | Deployment timeline | Capital intensity | Carbon profile |
|---|---|---|---|
| Grid interconnection | 5-8 years | Low (utility-funded upgrades) | Mix of grid resources |
| Gigawatt PPA | Contract now, power 2-4 years out | Low upfront | Renewable or nuclear |
| Behind-the-meter fuel cells | ~90 days | High | Natural gas |
| Nuclear co-location | 4-6+ years | Very high | Carbon-free |
The Stargate template
The Stargate campus in Abilene, Texas is the canonical integrated design: 1.2 GW across 8 buildings, with aeroderivative gas turbines, Tesla Megapack battery storage, planned Kairos Power SMRs, and Fervo geothermal. The pattern is grid-optional rather than grid-dependent. On-site generation and BESS provide NERC-compliant ride-through during grid disturbances while grid interconnection is still pending.
BESS is now mandatory, not optional. AI training GPU clusters exhibit power swings of 30-70 percent on sub-second timescales, roughly 15 times the variability of conventional data center loads. Wärtsilä and others cite BESS as necessary to keep GPU load swings from destabilizing the point of interconnection.
Is fast-tracking interconnection legal?
The most serious challenge is in federal court. Earthjustice filed suit on November 18, 2025 on behalf of Clean Wisconsin, NRDC, and Sierra Club, challenging FERC's approval of MISO's Expedited Resource Addition Study. The suit alleges ERAS "allows the fast-tracked projects to pass on significant upgrade costs to residential customers and to skip over clean energy projects that have been waiting for years to connect to the grid."
A D.C. Circuit opening brief was filed April 17, 2026.
Renewable developers have attacked PJM's parallel Reliability Resource Initiative as "a blatant attempt to perpetrate undue discrimination and preference for the benefit of a select few load-serving entities." Maryland's Office of People's Counsel filed a FERC complaint May 7, 2026 challenging PJM's Expedited Interconnection Track.
States are moving too. Florida Governor DeSantis signed an "AI Bill of Rights" on December 4, 2025 prohibiting utilities from passing hyperscale costs to residential ratepayers. Eleven or more state moratorium bills were filed in 2026. The political coalition is bipartisan and widening.
Will efficiency save the grid?
Probably not. The digital Jevons paradox is now documented in peer-reviewed work. A Nature Cities correspondence from July 2025 found that "algorithmic efficiency gains in metropolitan data centers may enlarge, and not shrink, the energy footprint of artificial intelligence."
Google achieved 33x efficiency gains per query over its history, yet its carbon footprint rose 48 percent over five years as query volume outpaced per-query gains. Nvidia's Rubin GPUs draw 1,950W per chip, multiples of prior Hopper parts, because FLOP/s scale with power at fixed silicon efficiency.
Liquid cooling gets PUE to 1.05-1.09 at best-in-class sites, but that margin cannot offset 160-165 percent demand growth.
Sam Altman put it bluntly in early 2026: "There's no way to get there without a breakthrough."
What this means for you
If you are siting, financing, or operating AI infrastructure, the practical playbook is concrete.
- Map substation capacity before you map fiber. Power availability is now the primary siting input. Underutilized corridors in Texas, Wyoming, New Mexico, and secondary PJM zones beat congested Northern Virginia on time-to-power.
- Plan a four-track power stack from day one. Grid interconnection, gigawatt PPA, behind-the-meter generation, and nuclear or geothermal commitments are complementary, not alternatives. The integrated campus model is the new default.
- Budget for BESS as a required component. GPU load swings of 30-70 percent on sub-second timescales make battery storage a reliability necessity, not a nice-to-have.
- Treat interconnection reform as contested. The Earthjustice litigation and state moratorium bills mean fast-track timelines could be reversed in court. Do not underwrite a project on the assumption that the queue clears in 24 months.
- Watch the cost-allocation fight. FERC has not decided who pays for grid upgrades driven by data centers. Florida has. The resolution will shape hyperscaler economics for the next decade.
- Date every version-specific assumption. FERC has been shipping interconnection policy roughly monthly since late 2025. Anchor capacity auction results, queue lengths, and PPA terms to a specific date, because the numbers rot fast.
The grid crunch is not a problem FERC can solve with one order. It is a multi-year rebuild of how AI compute gets power, and the operators winning right now are the ones who stopped waiting for the queue and started building their own.
Sources
- FERC fact sheet: Supercharge America's Grid
- FERC RM26-4 ANOPR on large-load interconnection
- Commissioner Chang concurrence, Docket EL26-67-000
- McGuireWoods: FERC Section 206 show-cause orders
- Berkeley Lab: data center electricity demand report
- PJM Inside Lines: FERC approves interconnection reform
- NERC 2025 Long-Term Reliability Assessment coverage
- Kirkland & Ellis: fast-tracking generator interconnection
- The Watt Report: transformer lead times as the post-queue problem
- NRDC: lawsuit challenging MISO ERAS fast-track
- Utility Dive: renewable developers urge PJM to drop fast-track plan
- World Nuclear News: Meta nuclear agreements
- OpenAI: five new Stargate sites
- Nature Cities: digital Jevons paradox in data centers
- ERCOT: Batch Connection Process for Large Electricity Users