The Federal Energy Regulatory Commission voted unanimously on June 18, 2026, to order six of the country’s largest grid operators to explain, within 30 days, how they’ll connect a wave of new AI data centers to the power grid faster. If regulators didn’t like the answer, the operators had 60 days to rewrite their own connection rules, a window that closed on August 17, 2026, when their tariff filings came due; the operators sought more time, and at least some compliance deadlines now extend beyond that original window. The order landed in docket RM26-4-000, a proceeding FERC previewed in April when it said it intended to act by the end of June. It arrives as AWS, Microsoft Azure, and Google Cloud all race to plug new AI capacity into a grid that was never built for gigawatt-scale customers showing up several times a year.
For an industry used to measuring growth in server racks, the FERC order is a reminder that AI’s real bottleneck in 2026 isn’t chips. It’s electrons. This piece breaks down what FERC ordered, which grid operators are on the clock, how AWS, Microsoft, and Meta have already spent billions locking up nuclear power to get ahead of the queue, and what the ruling means for anyone building on cloud infrastructure for the rest of 2026.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What FERC Actually Ordered on June 18
FERC didn’t write new interconnection rules itself. Instead, it issued a show-cause order: a formal instruction telling six grid operators to either fix the problem themselves or explain, on the record, why the current process is fine. Forbes described the requirement as forcing operators to “explain or revise the guidelines that dictate how substantial new electricity consumers, such as AI data centers, connect to the power grid.”
The order asks each operator to address four things: how much spare generating capacity exists on their system, how they’ll manage interconnection queues so large loads don’t get stuck for years or jump ahead of smaller customers, how they’ll stop new substation and transmission costs from landing on residential bills, and how they’ll make data centers cover the full cost of the grid work tied to their own connection. That last point is what utilities and consumer advocates have fought over for two years, and FERC just made it a federal requirement rather than a state-by-state guessing game.
The timeline is tight by regulatory standards. Operators had roughly 30 days, landing around mid-July 2026, to file reliability and resource-adequacy information, and on July 20, 2026, they submitted informational reports concerning generation adequacy for existing and prospective large loads. They then had roughly 60 days to file revised tariffs or justify why existing rules already meet the bar, a deadline that passed on August 17, 2026. TechCrunch summarized the outcome bluntly: “FERC told grid operators on Thursday to fast-track interconnection requests from data centers and other large electricity users.”
The Six Grid Operators Under FERC’s 60-Day Order
The order names six regional transmission organizations and independent system operators, the entities that actually run the wholesale grid and decide who gets to connect and when. Between them, they cover most of the country outside Texas and the Southeast. Cloud Computing News reported that “FERC voted unanimously on Thursday to require the operators to show how they will handle requests from large electricity users in a faster and more orderly way,” a description that applies equally to all six.
| Grid Operator | Region Served | ~30-Day Filing Deadline | ~60-Day Filing Deadline |
|---|---|---|---|
| PJM Interconnection | Mid-Atlantic and parts of 13 states plus Washington, D.C. | ~July 18, 2026 | ~August 17, 2026 |
| MISO | 15 states across the central U.S. | ~July 18, 2026 | ~August 17, 2026 |
| SPP (Southwest Power Pool) | 14 states across the central and southern U.S. | ~July 18, 2026 | ~August 17, 2026 |
| CAISO | California | ~July 18, 2026 | ~August 17, 2026 |
| ISO New England | Six New England states | ~July 18, 2026 | ~August 17, 2026 |
| NYISO | New York State | ~July 18, 2026 | ~August 17, 2026 |
PJM is the one worth watching first. It’s the largest U.S. grid operator, and its queue of data center interconnection requests covers the Virginia data center corridor, one of the densest concentrations of cloud and colocation capacity in the world. The 60-day window from FERC’s June 18, 2026 orders closed on August 17, 2026, and the operators sought more time, so PJM’s large-load rules are still being reworked.
Why Texas and ERCOT Are Sitting This One Out
One grid is conspicuously absent from the list: ERCOT, which runs the Texas grid and serves roughly 90% of the state’s electricity load. Texas built ERCOT specifically to stay outside interstate power transmission, which is what keeps it beyond FERC’s direct jurisdiction in a case like this one. That’s a structural fact of how Texas set up its grid decades ago, not an oversight in the June 18 order.
That doesn’t mean Texas is immune to the same pressure. The state has become one of the fastest-growing markets for hyperscale data center construction in the country, and its own grid operators and legislature have been wrestling with large-load interconnection questions independently of Washington. Whatever changes happen there will run through the state legislature and the Public Utility Commission of Texas rather than FERC’s docket, which means AWS, Microsoft, and Google Cloud are navigating two separate regulatory tracks for the same problem, depending on which state a data center sits in.
The Data Center Power Numbers Driving the Order
FERC’s order didn’t come out of nowhere. It’s a response to a demand curve bending sharply upward, and not just in the United States. S&P Global’s 451 Research puts U.S. data center grid power at 64.4 gigawatts in 2025, up 25% year-over-year and nearly tripling since 2020, on a path that climbs to 75.8 gigawatts in 2026, then 108 gigawatts by 2028 and 134.4 gigawatts by 2030. Zoomed out to the whole planet, Gartner’s June 2026 forecast puts global data center power demand at 133 gigawatts for the year, up from 105 gigawatts in 2025, alongside worldwide data center electricity consumption climbing to 565 terawatt-hours in 2026, a 26% jump from 447 TWh the year before, a figure not far off the International Energy Agency’s own count of roughly 485 TWh of global data center electricity consumption in 2025. That’s more than double in five years for the U.S. alone, concentrated in regions that overlap almost exactly with the six grid operators named in the order.
Zoom out to the national picture and the numbers still stand out. The U.S. Energy Information Administration’s Short-Term Energy Outlook projects total U.S. electricity consumption rising from 4,193 billion kilowatt-hours in 2025 to 4,283 billion kilowatt-hours in 2026, a jump the agency attributes partly to data center load. The Electric Power Research Institute has modeled how much of that growth data centers alone will drive: from about 4% of total U.S. electricity load in 2023 to somewhere between 9% and 17% by 2030, depending on how aggressively AI buildout continues. Globally, the same pattern shows up in the International Data Center Authority’s May 2026 estimate that data centers now draw 2% of worldwide electricity, up from 1.9% in mid-2025 and 1.7% in 2024, a disparity the International Energy Agency quantified directly in its 2025 review: global data center electricity demand grew 17% that year, more than five times the 3% growth rate for total electricity demand worldwide. North America is already the center of gravity for this spending, accounting for 38.0% of worldwide data center power revenue in 2025, according to Grand View Research. The EIA’s Today in Energy series has tracked similar load growth tied directly to data center campuses coming online.
| Year | U.S. Data Center Power Demand | Source |
|---|---|---|
| 2023 | ~4% of total U.S. electricity load | EPRI |
| 2025 | 64.4 GW | S&P Global 451 Research |
| 2026 | 75.8 GW | S&P Global 451 Research |
| 2028 | 108 GW | S&P Global 451 Research |
| 2030 | 134.4 GW (9%-17% of total U.S. load) | S&P Global 451 Research / EPRI |
Put another way: data center power demand is on pace to grow roughly as fast between now and 2030 as it grew in the entire prior decade, and it’s happening in a country where building a new high-voltage transmission line can take longer than building the data center that needs it. That mismatch is the entire reason FERC felt compelled to intervene rather than wait for grid operators to fix it on their own timeline, and it echoes pressure building well beyond U.S. borders, including South Korea’s own power constraints on its $880 billion chip buildout.
Doing the Math: The Gigawatt Gap Behind the Urgency
Run the numbers FERC is staring at and the urgency behind the June 18 order gets easier to understand. S&P Global’s 451 Research trajectory shows U.S. data center power demand climbing from 75.8 gigawatts in 2026 to 134.4 gigawatts by 2030, a gap of roughly 58.6 gigawatts that has to be found somewhere in just four years, on a grid built around a much slower growth curve. S&P Global’s own analysts underscored how steep that curve looks during a June 2026 webinar, projecting that data center power demand will more than double by 2030 from 2026 levels, a pace that turns the six grid operators’ 60-day tariff rewrites from a bureaucratic formality into a race against a doubling clock. The global trend lines point the same direction: the International Energy Agency projects worldwide data center electricity consumption will nearly double by 2030, reaching roughly 950 TWh, a doubling timeline that closely tracks what S&P Global sees for U.S. gigawatt demand, while the IEA’s September 2026 outlook projects that AI data-center power use alone will triple by 2030.
That gap is what FERC’s 30-day reliability filings, due by mid-July 2026, were designed to surface: how much headroom already exists on each system versus how much still needs to be built. It’s also why the order leans so heavily on interconnection speed rather than generation alone. Even where new power plants get built, that capacity doesn’t help an AI campus sitting in a multi-year queue waiting for a study slot. Faster reviews are FERC’s way of unlocking usable capacity from a grid that can’t physically add 58.6 gigawatts of new generation in 48 months.
The regional concentration makes the math harder still. EPRI’s projection that data centers could reach 9% to 17% of total U.S. electricity load by 2030, up from about 4% in 2023, isn’t spread evenly across the country. It’s concentrated almost entirely inside the footprints of the six operators named in the June 18 order, which is exactly why PJM, covering the Virginia data center corridor, faces the tightest math of all six.
Who Pays: FERC’s Cost-Allocation Fight
Underneath the interconnection-speed headline is a messier fight over money. When a grid operator approves a new substation or transmission upgrade to serve a data center, someone has to pay for it, and historically that cost has sometimes been spread across an entire rate base rather than billed to the customer that triggered it. Consumer advocates in multiple states point to that dynamic as one driver of rising household electricity bills, arguing that ordinary ratepayers are subsidizing infrastructure built for a handful of trillion-dollar cloud companies.
FERC’s order tries to close that gap by requiring data centers to pay the full cost of grid upgrades tied to their own connections. In practice, that shifts a real expense onto AWS, Microsoft, Google Cloud, and every colocation operator building AI capacity, on top of what they already spend on chips, buildings, and cooling. It also gives grid operators a way to slow down requests that would otherwise force system-wide upgrades paid for by everyone else. Expect this cost-allocation language to be the most contested part of the tariff filings that came due on August 17, 2026, since it determines how much a new gigawatt-scale campus costs its developer versus its neighbors.
How AWS, Microsoft, and Meta Are Buying Their Way Around the Grid
None of this is news to the hyperscalers. Long before FERC’s order, AWS, Microsoft, and Meta had concluded that waiting in a public interconnection queue wasn’t a viable growth strategy, and started locking up dedicated power directly with nuclear plant operators. Microsoft moved first: in September 2024, it signed a 20-year power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1, renamed the Crane Clean Energy Center, securing 835 megawatts of carbon-free power. The restart cost roughly $1.6 billion, and Reuters later reported the project also picked up a $1 billion Department of Energy loan in November 2025.
AWS has been the most acquisitive of the three, stacking multiple deals with the same counterparty. It bought Talen Energy’s Cumulus data center campus for $650 million in March 2024, tied to up to 960 megawatts from the adjacent Susquehanna nuclear plant. Seven months later, AWS committed more than $500 million toward SMR development with Dominion Energy and Energy Northwest, whose first phase is 320 megawatts with room to expand to 960. By June 2025, AWS had returned to Talen Energy to enhance the deal into a 1,920-megawatt agreement through 2042, worth roughly $18 billion over 17 years according to PowerMag, though Reuters said Amazon didn’t disclose full terms.
Meta signed its own deal in June 2025: a 20-year power purchase agreement with Constellation Energy for 1,100 megawatts from the Clinton, Illinois nuclear plant, set to begin supplying power in June 2027. Google Cloud is the notable gap in this table. Nothing in the same window points to a nuclear or dedicated-generation deal from Google of comparable scale, which doesn’t rule out quieter arrangements in progress, but it does mean Google has been the least vocal of the major cloud providers about locking in its own power supply.
| Company | Energy Partner | Deal | Capacity | Date |
|---|---|---|---|---|
| Microsoft | Constellation Energy | Three Mile Island Unit 1 restart (Crane Clean Energy Center) | 835 MW | Sept. 20, 2024 |
| Amazon / AWS | Talen Energy | Cumulus campus purchase + Susquehanna nuclear ties | Up to 960 MW | March 2024 |
| Amazon / AWS | Dominion Energy / Energy Northwest | SMR development commitment | 320 MW (phase 1, up to 960 MW) | Oct. 16, 2024 |
| Amazon / AWS | Talen Energy | Enhanced nuclear PPA through 2042 | 1,920 MW | June 11, 2025 |
| Meta | Constellation Energy | Clinton, Illinois nuclear plant PPA | 1,100 MW | June 2025 |
| Google Cloud | – | No comparable deal publicly disclosed | – | – |
AWS vs. Microsoft Azure vs. Google Cloud: Who’s Most Exposed
Line the three biggest cloud providers up against the FERC order and their exposure looks different in each case, a divergence Bloom Energy’s mid-year 2026 report captured when it named power availability the factor sorting the industry into regional “winners and losers,” calling 2026 the inflection year. Microsoft got the earliest start with a single, high-profile nuclear restart, which buys it a fixed 835-megawatt block of carbon-free power but doesn’t scale further without a new deal. AWS took the opposite approach, signing three agreements with three partners in 15 months, betting that spreading commitments across multiple nuclear and SMR projects buys more flexibility. That fits AWS’s pattern of aggressive infrastructure bets, visible too in deals like Pinterest’s $4 billion AWS cloud commitment through 2031, and in the strain that scale creates, underscored by AWS’s fourth US-West-2 outage in four months.
Google Cloud presents the most interesting case precisely because of what’s missing. The division has posted some of the industry’s fastest growth, a trend Tech Insider covered when Google Cloud’s growth rate hit 82% ahead of AWS’s own earnings report, yet it hasn’t publicly matched its rivals’ dedicated power commitments. That leaves Google more dependent on whatever capacity PJM, CAISO, and MISO can free up through the FERC-mandated reforms, rather than power it has already locked in. If the 60-day tariff rewrites move slower than expected, Google Cloud has less of a buffer than its two biggest rivals.
Behind-the-Meter Deals and the Interconnection Queue They Bypass
There’s a reason AWS, Microsoft, and Meta didn’t wait for FERC to act before locking up nuclear power directly. Every megawatt secured through a bilateral power purchase agreement, like AWS’s 1,920-megawatt Talen Energy deal or Meta’s 1,100-megawatt Clinton, Illinois agreement, doesn’t have to move through the same interconnection queue slowing down everyone else. These arrangements sit adjacent to existing generation rather than requesting a brand-new grid connection from scratch, exactly the kind of workaround FERC’s order is trying to make less necessary by speeding up the standard process.
That’s also why Google Cloud’s absence from the nuclear dealmaking table matters more after June 18 than it did before. Every hyperscaler without a comparable dedicated-power arrangement is now betting more heavily on the standard interconnection process actually getting faster under the timeline FERC set. The 60-day deadline passed on August 17, 2026, and operators sought more time. If the tariff rewrites from PJM, CAISO, or MISO stall or draw prolonged legal challenges, the gap between cloud providers with dedicated power and those without widens further, turning what started as a procurement choice into a real competitive advantage.
From Server Rooms to Gigawatt Campuses: A Short History of Data Center Power
Grid interconnection rules were written for a different era of industrial customer, the kind that shows up once every few years: a steel mill, an aluminum smelter, maybe a large manufacturing plant. Those processes could take years to complete because the pace matched how often they were needed. Data centers used to fit inside that slow cadence too. A large facility a decade ago might draw 20 to 50 megawatts, a rounding error next to a coal or gas plant’s output. Most of that installed base still sits in Tier 3 facilities, which held 64.42% of the data center power market in 2025, according to Mordor Intelligence, but even that redundant, enterprise-grade design tier was never engineered for the gigawatt-scale draws AI campuses now request.
AI training and inference clusters broke that pattern. A single modern AI campus can request several hundred megawatts, and hyperscalers now file for that scale of connection multiple times a year, across multiple regions, simultaneously. EPRI’s trajectory from roughly 4% of U.S. electricity load in 2023 to a projected 9%-17% by 2030 captures how fast that shift happened. The equipment vendors supplying that buildout have scaled up right alongside it: Vertiv, the largest single supplier of data center power gear, held 8.4% of the global data center power market in 2025, per GMI Insights, a share built almost entirely on this new class of gigawatt-hungry customer. Newer entrants are chasing the same demand with different hardware: Sungrow launched a solid-state transformer for AI data centers in July 2026 that hits 98.5% peak efficiency on an 800V DC architecture, which Pandaily reported cuts a data center’s power distribution footprint by 50%, exactly the density gain gigawatt-scale campuses need to squeeze more compute behind the same interconnection point. Advanced Energy’s new 2026 M-CRPS PSUs, which deliver up to 3,600 W (October 2026), show the equipment race around these loads is still accelerating. The interconnection process wasn’t built for a customer class that behaves like a dozen new steel mills a year, wanting to connect within months, not years. That gap is what pushed FERC to act rather than wait for grid operators to fix it on their own schedule. The IEA’s September 2026 analysis found that five tech firms’ capex exceeded $400B in 2025, and the buildout behind it is also why data center lease commitments crossed $850 billion, with Meta and Microsoft leading that spending.
Market Impact: Utility Bills, Cloud Pricing, and Ratepayer Backlash
The most immediate market impact runs through utility rate cases, not cloud pricing pages. State public utility commissions across PJM and MISO territory have already fielded complaints from ratepayer groups arguing that data center-driven grid investment is showing up in monthly bills. FERC’s requirement that data centers cover the full cost of their own connection upgrades responds directly to that pressure, and it gives state regulators a clearer federal baseline when utilities propose new rate structures.
For cloud customers, the impact is more indirect but still real. If grid operators require hyperscalers to fund substations and transmission upgrades outright, that’s a new, recurring capital cost that wasn’t part of the calculus when cloud providers set regional pricing years ago. The market underneath that cost is expanding fast in its own right: North America data center power revenue hit $5.38 billion in 2025 and is projected to reach $5.73 billion in 2026, per Fortune Business Insights, with power solutions, hardware and software rather than services, accounting for 72.2% of that spend, according to Grand View Research. Estimates of the broader global data center power market vary by research house but point the same direction: Global Market Insights sized worldwide data center power market revenue at $77.5 billion for 2026 in its August 2026 update, Grand View Research separately put the market at $25.78 billion in 2026 (up from $22.77 billion in 2025), and Polaris Market Research pegged it at $26.13 billion in 2026 after $23.10 billion in 2025. That cost tends to surface in reserved-capacity pricing or new-region rollout timelines rather than sticker prices on existing instances, so architects should treat regional power availability as a pricing variable, not just a capacity one. The pattern isn’t unique to the U.S.: the UK’s decision to name four cloud giants as critical third parties shows regulators worldwide converging on the same idea.
What Industry Voices Are Saying
Coverage of the June 18 order has converged on a few themes: speed, fairness, and reliability. Writing for Forbes, contributor Ron Schmelzer framed the order as a direct mandate for grid operators to justify their existing processes: “The federal energy authority has specifically required these operators to explain or revise the guidelines that dictate how substantial new electricity consumers, such as AI data centers, connect to the power grid,” he wrote in Forbes.
Annie Palmer, reporting for TechCrunch, described the practical effect of the ruling for the companies building AI infrastructure: “FERC told grid operators on Thursday to fast-track interconnection requests from data centers and other large electricity users,” she reported for TechCrunch.
Industry outlet Cloud Computing News emphasized the unanimous nature of the vote, a signal that the push for faster interconnection has bipartisan support among FERC’s commissioners regardless of the underlying debate over who pays: “FERC voted unanimously on Thursday to require the operators to show how they will handle requests from large electricity users in a faster and more orderly way,” the outlet reported.
Inside a Show-Cause Order: What “Explain or Revise” Actually Requires
FERC’s choice of legal tool matters as much as the deadline itself. A show-cause order doesn’t hand grid operators a pre-written rulebook to adopt. It puts the burden of proof on them. Each of the six operators has to either demonstrate, with data, that its existing interconnection process already meets FERC’s goals, or propose a specific rewrite and defend it on the record. Silence or a vague response isn’t an option. The docket, RM26-4-000, gives FERC a paper trail it can point to if an operator’s filing looks like it’s dodging the question.
That structure explains why none of the six was likely to simply tell FERC their current rules are adequate. Interconnection queues at PJM, MISO, and several of the others have drawn public complaints for years, well before this order, from developers waiting on multi-year study timelines. Telling FERC that no changes were needed in a formal filing would have meant defending that track record directly against the commission’s own stated concerns, in writing, with the risk of a rejected filing and a second, less favorable round of forced rulemaking.
The more likely path is a set of tariff proposals that vary by region. PJM, CAISO, and NYISO each serve different generation mixes, queue backlogs, and state-level politics around cost allocation, so a one-size-fits-all fix was never realistic. The 60-day deadline passed on August 17, 2026, and at least some compliance deadlines now run later. What comes out of the process won’t be six identical rulebooks. It will be six regional experiments in interconnection reform that FERC can compare before deciding which approach becomes the model for the rest of the country, including, eventually, Texas.
The Compliance Countdown: What Happens by Mid-August
Two dates mattered more than any product launch on the cloud industry’s calendar. The first came in July: by around July 18, 2026, all six grid operators had to file reliability and resource-adequacy data showing FERC what spare capacity actually exists on their systems, and on July 20, 2026, they submitted informational reports concerning generation adequacy for existing and prospective large loads. That filing was revealing on its own, since it marked the first time some of these operators had to put a hard number on how much headroom is left before AI data center demand outstrips what the grid can deliver.
The second arrived on August 17, 2026, when the operators’ 60-day response deadline passed and each needed to have either filed revised interconnection tariffs or made the case to FERC that its existing rules already satisfy the order’s goals. Given how contentious cost allocation already is in states like Virginia, Ohio, and Illinois, few observers expected any of the six to simply tell FERC their current rules are fine. What followed was the kind of drawn-out process that question promised: the operators sought more time, and at least some compliance deadlines now extend beyond the original window, with further FERC review, comment periods, and probably litigation from parties on both sides of the cost question still ahead.
Where the PJM Filings Stand After the Deadline
PJM is the operator to follow once the deadline has passed. On September 29, 2026, FERC accepted but suspended PJM’s proposal to accelerate procurement of new capacity, and it identified unresolved concerns involving cost allocation, collateral, and exit rules. FERC also encouraged PJM to submit a revised filing that could shorten the suspension. On September 30, 2026, PJM suspended its proposed one-time auction for an estimated 6.8-gigawatt shortfall and said it would review FERC’s order and work on revisions.
On October 14, 2026, FERC blocked a PJM framework that would have allowed certain new data centers and other users requiring at least 50 MW to connect without matching new generation in exchange for possible curtailment during grid stress. Separately, a Department of Energy emergency order covering October 13 to 16, 2026 authorized PJM to dispatch backup-generation resources during a reliability emergency, and it was reported as effective only through October 16. PJM also began its annual review of proposed large-load adjustments for its 2027 Long-Term Load Forecast on September 29, 2026, and on October 19, 2026, twelve utilities presented proposed large-load adjustments to PJM.
For anyone planning capacity in PJM territory, the practical lesson is that the August deadline did not settle how large loads connect. Cost allocation and connection terms are still being worked out, so any assumed interconnection timeline there should be treated as provisional. Developers without dedicated power are also looking for other routes: on October 22, 2026, Google and Voltus announced a one gigawatt Bring Your Own Capacity agreement in PJM.
5 Predictions for the Rest of 2026
- PJM draws the most scrutiny. As the largest operator by population and data center backlog, expect PJM’s filings to face the toughest FERC questions and the most ratepayer pushback.
- More hyperscaler nuclear deals land before year-end. Given the pace Microsoft, AWS, and Meta set in 2024-2025, another large power purchase agreement is likely before December, and Google Cloud is the most probable candidate to finally announce one.
- Ratepayer advocates flood the comment windows. Consumer groups in PJM and MISO territory have already been vocal about cost-shifting, and FERC’s open docket gives them a formal channel to push for stricter cost-allocation language.
- Texas faces rising pressure to act independently. With ERCOT excluded from the federal order but facing the same demand growth, expect renewed legislative or Public Utility Commission of Texas activity on large-load interconnection.
- Regional capacity becomes a competitive differentiator. Operators that move fastest, likely PJM and CAISO given their AI data center density, will see hyperscalers prioritize new capacity there, widening the gap with slower regions.
What This Means for Developers, IT Buyers, and Cloud Architects
For most software teams, an FERC docket number will never show up in a sprint planning meeting, but the downstream effects of this order will. Regional power constraints are becoming a real factor in when new GPU instance types and AI-optimized regions come online, alongside the usual chip supply questions. Teams planning multi-year AI infrastructure commitments should ask cloud providers directly about power-sourcing plans for the capacity they’re buying, not just chip generations and pricing tiers.
It’s also worth connecting this to cost discipline more broadly. Cloud spend in power-constrained regions carries a different risk profile than spend in regions with headroom, and FinOps teams already fighting cloud waste that’s climbed to 29% as AI spend breaks budgets now have one more variable to track: whether a region’s grid operator is among the six that received FERC’s June 18 show-cause orders. Factoring regional power risk into capacity planning now means fewer surprise costs or delayed rollouts later in 2026 and into 2027.
Frequently Asked Questions
What did FERC order on June 18, 2026?
FERC voted unanimously to direct six regional grid operators to show, within roughly 30 days, how much spare capacity they have, then file revised interconnection tariffs, or justify why current rules are sufficient, within roughly 60 days. The order also requires data centers to pay the full cost of grid upgrades tied to their own connections.
Which grid operators does the FERC order apply to?
PJM Interconnection, MISO, SPP, CAISO, ISO New England, and NYISO. Together they cover most of the continental United States outside Texas and parts of the Southeast.
Why is ERCOT/Texas not included in the FERC order?
Texas built ERCOT to operate largely outside interstate power transmission, which keeps it beyond FERC’s direct jurisdiction in this proceeding. Any comparable reforms in Texas would have to come through the state legislature or the Public Utility Commission of Texas rather than this federal order.
How much power do U.S. data centers use in 2026?
S&P Global’s 451 Research estimates U.S. data center power demand at 75.8 gigawatts in 2026, up from 64.4 gigawatts in 2025 (itself a 25% year-over-year jump and nearly triple 2020 levels), and projects it will reach 134.4 gigawatts by 2030. Goldman Sachs Research’s own count runs lower but shows an even steeper climb: 31 gigawatts in 2025, rising to 41 gigawatts in 2026 and 66 gigawatts by 2027.
Who pays for new power grid connections for AI data centers?
Under FERC’s June 2026 order, data centers must pay the full cost of grid upgrades directly tied to their own interconnection, rather than having those costs spread across the broader ratepayer base.
Are AWS, Microsoft, and Google building their own power plants?
Not building from scratch, but all three have pursued dedicated power deals. Microsoft signed a 20-year deal to restart Three Mile Island Unit 1 (835 MW). AWS has signed three separate nuclear and SMR agreements with Talen Energy, Dominion Energy, and Energy Northwest since March 2024. Meta signed a 1,100 MW deal for Constellation’s Clinton, Illinois plant. Google Cloud has not disclosed a comparable large-scale deal in the same window.
When do grid operators have to comply with the FERC order?
Reliability and resource-adequacy reports from the operators were submitted on July 20, 2026, about 30 days after the June 18, 2026 order. Revised tariffs or a justification for existing rules were due on August 17, 2026, at the 60-day mark.
Has the 60-day FERC tariff deadline passed yet?
Yes. The 30-day reliability filings were due around mid-July 2026, and the operators submitted informational reports on July 20, 2026. The 60-day deadline for revised tariffs, or a filed justification for existing rules, passed on August 17, 2026, and the operators sought more time, so at least some compliance deadlines now extend beyond that original window.
Could the power crunch slow down cloud AI expansion?
It already has in specific regions with the longest interconnection queues, which is part of why FERC intervened. Whether it slows expansion nationally depends heavily on how the six grid operators followed up on the 60-day tariff deadline, which passed on August 17, 2026 with operators seeking more time, and how quickly hyperscalers can secure alternative power sources like nuclear PPAs in the meantime.
