Data Center Power Demand and the FERC PJM Ruling Explained

Published: October 19, 2026 | Category: Technology | By Mahesh

On October 9, 2026, the Federal Energy Regulatory Commission rejected the core of PJM Interconnection's plan for connecting data centers to the largest power grid in the United States, three days after Google signed a 3,590 MW supply deal built around the same plan's bring-your-own-power idea.^[1][2] Data center power demand has moved from an engineering forecast to a regulatory fight over who builds the generation and who pays for it. The answer will shape where AI capacity gets built for the rest of the decade.

This is the first piece in a Depth Grid series on AI infrastructure and electricity, part of our wider technology coverage. It builds on two earlier reports that examined other bottlenecks: permission, in our story on San Francisco's pause on new data centers; credit, in our report on SpaceX's $40 billion chip financing. Here we follow the electricity, using the numbers from the PJM auction, the FERC order and the Google deal; we close with what it means for India.

What Is Driving Data Center Power Demand Through 2030?

Data center power demand is set to more than double from about 415 TWh in 2024 to about 945 TWh by 2030, according to the International Energy Agency, with the United States supplying nearly half of the growth. The strain shows up first as a capacity problem: in PJM, auction prices have hit the regulatory cap three years running.

The International Energy Agency put data center electricity use at about 415 TWh in 2024 (1.5% of global consumption) and projects roughly 945 TWh by 2030 in its base case. That is slightly more than Japan uses today. Its longer view reaches about 1,200 TWh by 2035.^[3] The jump works out to a compound growth rate near 15% a year, by Depth Grid arithmetic, which is several times faster than demand from other sectors.

The IEA's framing deserves care. Data centers account for around one-tenth of global electricity demand growth to 2030, less than industrial motors, home and office air conditioning or electric vehicles.^[3] The concentration is what creates the pressure. In the United States data centers account for nearly half of demand growth over the period; a single AI-focused facility consumes as much electricity as 100,000 households, with the largest under construction drawing about twenty times that.^[3] A load that large arrives at one substation at a time, which is why the fight plays out in regional grid rules and not in national averages.

How Big Is the Capacity Gap in PJM?

PJM, the nation's largest grid operator, runs an annual capacity auction that pays generators to be available at peak.^[4] The July 14, 2026 auction for the 2028/29 delivery year cleared 138,318 MW at the $325 per MW-day price cap and still fell 6,831 MW short of the 156,013 MW reliability requirement, leaving a reserve margin of 14.7% against a 20% target.^[4][5] It was the third straight auction at the cap; the shortfall grew from about 6,500 MW the year before.^[4] Forecast demand rose by roughly 2 GW, largely from data center development.^[4]

Measure	Figure	Basis	Reading
Reliability requirement	156,013 MW	PJM 2028/29 auction	Target for a 20% reserve margin
Capacity procured	138,318 MW	PJM 2028/29 auction	14.7% reserve margin
Shortfall	6,831 MW	Requirement minus procured	Largest on record
New or uprated supply cleared	About 525 MW	7.7% of the shortfall (Depth Grid arithmetic)	Price cap drew little new supply
Price versus 2024/25	$325 against $28.92 per MW-day	11.2x increase	Held down by a collar

The price cap hides how tight the market is. PJM's own simulation shows the auction would have cleared at about $555 per MW-day across the region and $777 in the ComEd zone in northern Illinois without the collar.^[6] The auction bill would have been $29.7 billion instead of the $16.4 billion that results from the cap, which means the collar held back about $13.3 billion of cost by our subtraction.^[4] The collar, a cap of $325 and a floor of $175 per MW-day, has been extended through the 2029/30 auction.^[5] A shortfall does not mean blackouts; PJM says as much, but it does mean less margin for heat waves, plant outages and sudden load changes.

Sudden load changes are no longer theoretical. PJM disclosed that about 3.8 GW of data center load tripped offline in Northern Virginia on July 22, the largest such event in its history; the grid operator is weighing ride-through standards for data centers and crypto facilities.^[15] A grid that is already short of capacity also has to handle loads that can vanish in seconds and return just as fast.

What Did FERC Decide on October 9?

FERC rejected PJM's Interim Resource Adequacy Service on October 9, 2026 because the plan appeared to place obligations on retail customers, dictate which of them are curtailed and set a retail rate, all matters for states. FERC kept only a 50 MW large load definition and a registry; PJM must file compliance within 30 days.

PJM filed its package on August 13 in docket ER26-3515.^[5] The centerpiece, the Interim Resource Adequacy Service, would have let new large loads entering service after June 1, 2027 connect before matching generation existed, as long as they accepted curtailment ahead of other customers during grid emergencies.^[1] A companion option, bring your own new capacity, let a load escape that default by supplying new generation of its own. PJM also asked to exclude eligible new loads from the capacity auction demand curve starting with 2029/30.^[5]

The Federal Energy Regulatory Commission accepted two pieces: the definition of a large load as 50 MW or more at one site and a registry recording each facility's location, ramp schedule and capacity supply.^[1] The registry took effect October 12; a compliance filing is due within 30 days of the order, which falls on November 8 by our count.^[5] It rejected the rest. Its reasoning was jurisdictional: provisions that place obligations on new large loads, which are mostly retail customers, that choose which retail customers get curtailed and that set what looks like a retail rate belong to state regulators.^[1] The commission also found the treatment of loads with and without new capacity may be discriminatory.

Commissioner David Rosner wrote a concurrence that points at a market-design risk. If eligible loads were removed from the demand curve, the capacity market could signal that PJM has more than enough supply while the region sheds thousands of megawatts of demand involuntarily, which would be inefficient and would keep supply and demand out of balance.^[1] He also said non-critical data centers should be shut off before homes, hospitals and schools in an emergency, with states setting the priorities.^[1] PJM said it is reviewing the order and wants to honor the principle that existing customers should not be unfairly burdened by new, energy-intensive ones.^[1]

What Does Google's Nuclear Deal Actually Add?

The announcement came on October 6 from Google and Constellation Energy. A 20-year power purchase agreement supports 890 MW of new capacity at 11 Constellation-owned nuclear units in Illinois, Pennsylvania and New Jersey, through equipment upgrades that raise output at reactors that are already running.^[2] Constellation expects to invest more than $4.3 billion, the first uprate is due by 2028 and one analysis reads the schedule as all 890 MW arriving before the end of 2032.^[9][16] A separate 15-year agreement covers 2,700 MW from Constellation's existing PJM fleet, which brings Google's contracted total to 3,590 MW.^[2]

Both companies framed the deal as a direct response to PJM's bring-your-own-power proposal, a model in which a large consumer adds supply alongside its load without state or federal mandates.^[2] Google's energy head, Amanda Peterson Corio, tied it to the White House Ratepayer Protection Pledge and to the goal of growing without passing costs to residential customers.^[8] The same analysis notes that Google can cut non-critical power use during high-stress grid events, which is the flexibility PJM wanted to require.^[9] Uprates are the cheapest route to new firm power because the reactor, the site and the grid connection already exist.

Three comparisons put the deal in proportion; all three are Depth Grid arithmetic from public figures. The 890 MW equals about 13% of PJM's 6,831 MW shortfall and about 1.7 times all the new or uprated generation that cleared the 2028/29 auction. The $4.3 billion works out to about $4.8 million per megawatt of added capacity. The 2,700 MW agreement is a supply arrangement from existing plants, so it secures Google's electricity without adding a megawatt to the grid. Counting the full 3,590 MW as a fix for the shortage would overstate what the deal does.

Timing also matters. The 2028/29 delivery year starts June 1, 2028; only the first uprate is expected by then, so the deal cannot fill that year's gap on its own. Its larger effect is on precedent. After FERC's order, bring-your-own-power is no longer a tariff option in PJM. It is a voluntary commercial structure, which means the buyers with the balance sheets to sign 20-year contracts can use it and smaller operators cannot. That asymmetry is the most important consequence of the week's two events taken together.

Who Wins and Who Pays After the Ruling?

Equity analysts at Jefferies read the ruling as positive for existing generators, because it could lead to higher capacity prices for longer in PJM. They named Constellation, NRG Energy, Talen Energy and Vistra as the likely beneficiaries and said they expect the proposal to return in some form, probably led by states, with Ohio and Pennsylvania the ones to watch.^[1] The logic is simple. Without a mechanism that forces new loads to bring supply or accept curtailment, demand stays in the capacity auction and prices stay pinned at the cap.

Consumer advocates read the same order from the other side. The Citizens Utility Board in Illinois argues that the decision rejected measures that would have held data centers accountable for their costs, conflicts with the Ratepayer Protection Pledge and leaves state action as the last protection for households.^[7] That is an advocacy position; PJM's stated goal is the same outcome the group wants, but it shows how the argument has moved. The question is no longer whether data centers should bear the cost of the generation they require. It is which level of government can impose that cost.

The collar adds a second tension. It protects customers from the $555 and $777 prices that PJM's simulation shows, yet it also caps the revenue signal that new generation needs. A PJM official said new plants need significantly more than $325 per MW-day to be financially viable; the 2028/29 auction drew only about 525 MW of new or uprated supply.^[4][5] A price limit that protects consumers today can lengthen the shortage that costs them tomorrow. For a data center developer the practical reading is that the market will not build supply on its own at this price; a company that wants firm power has to contract for it.

How Are States and Cities Responding?

FERC pushed responsibility toward the states; several have moved already. Pennsylvania's governor issued an executive order in August that offers preferential permitting to data centers over 25 MW that bring their own power and meet the state's infrastructure standards.^[17] In Virginia, home of the world's densest data center cluster, a stack of moratorium bills is building ahead of the legislative session in January.^[18] In California, San Francisco voted unanimously for a 45-day moratorium that can run to 22 months, a story we covered in detail in our report on the city's pause on new data centers.

The pattern across these actions is a split between carrots and sticks. Pennsylvania rewards operators who supply their own generation. Virginia's bills and San Francisco's pause restrict new construction until rules catch up. For an operator the two approaches create different risks: one rewards capital and long-term contracts; the other penalizes anyone who needs a permit before the study ends. Our October 8 technology digest carried the first report of the Google deal alongside the SpaceX financing story; the two belong together. Chips are being financed with debt; the electricity to run them is being locked in with 20-year contracts.

What Does This Mean for India's Data Center Buildout?

India is earlier in the same cycle. JLL puts the country's data center capacity at 1.6 GW in June 2026 and expects 6 GW by 2029, a roughly 3.75-fold rise that would need about $110 billion of capital.^[10] Savills counts 1.8 GW of operational IT capacity after 258 MW of additions in the first half of 2026, up 59.3% on a year earlier; projects more than 7 GW by 2030.^[13] Wood Mackenzie is the most bullish, forecasting growth from 2.2 GW in 2025 to 12 GW by 2030, a compound rate near 40%, with AI-dedicated capacity rising from 275 MW to 6,546 MW.^[11] The spread between 6 GW and 12 GW reflects different baselines and definitions, which is a caution for anyone building a model on a single forecast.

The electricity numbers carry the same message as the PJM story. BloombergNEF expects India's data center electricity demand to rise more than eightfold, from 11 TWh in 2025 to 91 TWh in 2035, which is about 23% a year by our arithmetic.^[12] The Ministry of Power projects demand from the sector rising from around 1 GW to 13.56 GW by 2031-32.^[14] Wood Mackenzie concludes that reliable, cost-competitive power has overtaken land and capital as the main constraint; that access to firm round-the-clock supply at the node level decides site selection and delivery dates.^[11] Adani and Reliance have together committed about $210 billion, 45% of announced investment; about 95% of announced spending targets large hyperscale AI campuses.^[12]

The nearest Indian counterpart to bring your own power is captive generation and long-term renewable power purchase agreements, which Wood Mackenzie reports are becoming the preferred procurement strategy, along with states that offer open access and competitive transmission charges.^[11] JLL's report adds small modular reactors by 2033 as a possible off-grid option for large campuses.^[10] The PJM episode offers a warning and a template. A regulator that tries to impose the obligation by rule can be overturned on jurisdiction. A buyer that signs a firm supply contract faces no such risk but needs scale to do it. Indian operators with large balance sheets will have that route; smaller colocation players may not.

For readers following Indian AI hardware, the power question connects to the supply chain. Our report on Volantis and its $88 million for light-based AI memory chips covers one startup attacking the energy cost of moving data inside AI systems; our India business coverage tracks the capital behind the buildout.

What Should Operators and Investors Watch Next?

The next six months will show whether the PJM rules return through states, through a revised federal filing or through private contracts. These are the dated and undated markers worth tracking.

1. November 8, 2026: PJM's compliance filing for the large load definition and registry is due; it will show how PJM reads the order and what it proposes next.
2. State rulemaking in Ohio and Pennsylvania, where Jefferies expects any revived curtailment framework to originate; Virginia's January session for moratorium bills.
3. The capacity auction for 2029/30, the last year covered by the price collar, which will test whether prices stay at the cap and whether new supply responds.
4. Delivery of Google's first uprate by 2028 and progress toward the full 890 MW, since schedules for nuclear upgrades often move.
5. More hyperscaler contracts of the same type. Each deal that brings new generation reduces pressure on the auction; each one that only contracts existing plants adds to it.
6. In India, the final shape of the draft National Data Centre Policy and the first large captive or open-access supply deals tied to hyperscale campuses.

A simple test applies to every announcement. Ask how many of the megawatts are new, when they arrive and who bears the cost if they do not. The Google deal passes the first test for 890 MW and the second only partly. The regulatory picture fails the third, because FERC's order left the cost-allocation question with the states. Until that question is settled, expect the capacity market to stay expensive and the largest buyers to keep signing their own contracts. This article is analysis and not investment advice.

Common Questions

Sources

1. Ethan Howland, "FERC rejects most of PJM's large load interconnection plan," Utility Dive, October 12, 2026, reporting on FERC's order in Docket ER26-3515. utilitydive.com
2. Constellation Energy and Google, "Landmark Agreement to Bring 890 MW of New Nuclear Capacity to PJM Grid," October 6, 2026. constellationenergy.com
3. International Energy Agency, "Energy and AI," executive summary. iea.org
4. Utility Dive, "PJM capacity prices hit price cap, reserve shortfall grows," July 2026. utilitydive.com
5. Energy News Beat, "FERC Rejects Most of PJM's Data Center Plan; the $29.4 Billion Question of Who Pays Is Still Open," drawing on PJM's auction report and the FERC order. energynewsbeat.co
6. Avalon Energy Services, "PJM 2028/2029 Base Residual Auction Clears at Price Cap," July 2026. avalonenergy.us
7. Citizens Utility Board, "FERC rejection of PJM data center plan makes state action last hope to protect consumers," October 18, 2026. citizensutilityboard.org
8. Google, "Google and Constellation Announce Landmark Agreement to Bring 890 MW of New Nuclear Capacity to PJM Grid," October 6, 2026. googlecloudpresscorner.com
9. Global Data Center Hub, "Google Signed for 3,590 MW of Nuclear," October 2026. globaldatacenterhub.com
10. IANS, reporting JLL, "India's data centre capacity expected to jump to 6 GW by 2029," September 2, 2026. ianslive.in
11. Energetica India, reporting Wood Mackenzie, "India Data Centre Capacity to Reach 12 GW by 2030," July 28, 2026. energetica-india.net
12. Energetica India, reporting BloombergNEF, "India's Data Centre Capacity Quadruples to 1.7 GW," August 21, 2026. energetica-india.net
13. The Realty Today, reporting Savills India, "India's Data Centre Infrastructure Reaches 1.8 GW." therealtytoday.com
14. CXOToday, reporting Rubix Data Sciences and the Ministry of Power, "Data Centre Boom: India's Capacity to Quadruple to 6.5 GW by 2030." cxotoday.com
15. Utility Dive, "PJM eyes data center, crypto reliability requirements after 3.8 GW of load trips offline," August 12, 2026. utilitydive.com
16. MLQ.ai, "Google Signs 20-Year Nuclear Power Deal With Constellation for 890 MW," October 7, 2026. mlq.ai
17. Utility Dive, "Pennsylvania dangles permitting carrot for data centers that bring their own power," August 19, 2026. utilitydive.com
18. Virginia Mercury, "Data center moratorium bills stack up ahead of Virginia legislative session in January," October 16, 2026. virginiamercury.com

Related Analysis

• San Francisco Hits Pause on New Data Centers
• SpaceX's $40B Nvidia Chip Financing: What We Know
• Tech News Today, Oct 8: GPT-6, SpaceX, Nuclear Clocks
• Volantis $88M Funding: Light-Based AI Chips Explained
• More Technology Analysis from Depth Grid

Article by Mahesh | Depth Grid