Data-Center Demand Is Growing 10x Faster Than the Grid — and Your Power Bill Is Where It Lands
The numbers behind the AI boom finally have a single, stark summary. New research from Howden, published November 11, finds U.S. data-center power demand has grown 19% a year since 2000 — ten times faster than the 2% annual pace at which grid capacity has expanded over the same period. Data centers have gone from 0.1% of U.S. grid capacity in 2000 to 5% in 2025, a 50-fold increase in share. The grid is not keeping up, and the gap is now showing up in build times, in capacity prices, and — for buyers in New Jersey — in electric bills.
This is not a story about whether AI demand is real. It is a story about what happens when the fastest-growing load in the history of the U.S. power system meets the slowest-moving part of it: wires, interconnection queues, and permitting.
The 10x Gap, in Plain Numbers
The Howden analysis, built on EIA data and 451 Research by S&P Global, puts the mismatch in one chart: data-center demand compounding at 19% a year for 25 years, grid capacity at 2%. The EIA's own data tells the same story from the demand side. U.S. electricity demand grew about 1.7% a year from 2020 to 2025, after averaging just 0.1% a year from 2005 to 2019 — and data centers are the driver. The EIA expects national load to rise 1.9% in 2026 and 2.5% in 2027, with the fastest growth in ERCOT (about 10% a year) and PJM (about 3%).
The scale is hard to overstate. A single 100 MW hyperscale facility uses roughly as much electricity as 80,000 U.S. households, according to the Congressional Research Service. Lawrence Berkeley National Laboratory puts U.S. data-center electricity use at 192 TWh in 2024 — 4.7% of national consumption — and projects a 9.5% to 15.3% share by 2030. McKinsey's estimate lands in the same range: AI data centers consuming 11% to 12% of U.S. electricity by 2030. However you cut it, this is no longer a niche load. It is a structural driver of the power market.
What the Delay Costs
The most immediate consequence is time. Power constraints are adding 24 to 72 months to the build times of the most constrained data-center projects, according to Howden — two to six years of delay on facilities whose top-end annual revenues approach $400 million. The
lost revenue on the largest, most power-constrained builds runs $800 million to $2.4 billion. New generation capacity itself typically takes 24 to 60 months to develop, so the queue is not a paperwork problem. It is a physical one.
Reliability is the second cost. Half of data-center facilities reported an impactful outage in the past three years, Howden finds; most of them were power and backup failures — uninterruptible power supply systems first, then cooling, network, and IT failures. Each outage costs an estimated $500,000 to $1 million, implying roughly $2 billion a year in global losses. Many of those losses fall outside traditional property and business-interruption policies, because there is no insured physical damage — an insurance gap the industry is only starting to price.
The Behind-the-Meter Pivot — and Why Gas Matters
Faced with grid connections measured in years, hyperscale operators are increasingly building their own power behind the meter. Howden's math shows why that shift is unavoidable: for data centers to hold their 5% share of grid capacity at current growth rates, they would need to source up to 58% of their power — 110 GW of capacity — from behind-the-meter sources by 2030. Howden calls that unlikely to be achievable, so the real outcome will be a mix: a bigger grid share, moderated growth at constrained sites, and behind-the-meter generation.
The behind-the-meter mix, per Howden, is natural gas, renewables, battery storage, and — over the longer term — nuclear. That ordering matters for energy buyers. Gas is the only option on that list that can be built at scale, on site, inside the 24-to-60-month window the
grid cannot meet. It is also the option that ties this story directly back to the fuel markets: every gigawatt of behind-the-meter gas generation is new, steady, baseload gas demand that does not disappear when the weather turns mild. For a gas market still debating whether winter demand will materialize, data-center load is the demand story that does not depend on a cold snap.
Why New Jersey Feels It First
For buyers in our service territory, this is not an abstract national trend — it is already in the capacity bill. PJM, the grid operator for New Jersey, has seen its capacity auction price go from $28.92 per MW-day for the 2024/25 delivery year to $269.92 for 2025/26, and then
to $329.17 for 2026/27, hitting the price cap regulators imposed. The 2028/29 auction cleared at $325.00 — still more than 11 times the 2024/25 price. PJM's independent market monitor, Monitoring Analytics, attributes about $9.3 billion of the 2025/26 capacity costs — roughly 63% — to data-center load, and about $23 billion across the last three auctions combined.
PJM's own 2026 long-term load forecast projects 32 GW of peak load growth from 2024 to 2030, with 94% of it attributed to data centers. In PSE&G territory alone, interconnection requests for data centers and other large-load projects jumped from 400 MW in 2024 to 4,700 MW in 2025. New Jersey has responded: Governor Mikie Sherrill signed the Data Center Fair Share Act on July 7, 2026, creating a separate ratepayer class for large data centers and requiring them to cover their own grid infrastructure costs, with the Board of Public Utilities given 12 months to set the tariff. FERC, meanwhile, rejected most of PJM's plan for connecting large new loads this week, keeping the interconnection bottleneck — and the upward pressure on capacity prices — in place.
Massachusetts Is a Different Story — and That Matters Too
Here is where zone-specific advice earns its keep. Massachusetts sits in ISO New England, not PJM, and the data-center story barely touches it — yet. ISO-NE's first large-load forecast, published in its 2026 CELT report, found just two qualifying projects in the formal study phase: a 200 MW data center in northeast Massachusetts and an 85 MW
electrification load in Connecticut, 285 MW combined. After discounts, they add about 110 MW to peak demand in the 2030s, with no effect on system demand before winter 2027/28. In ISO-NE's own words, "New England has not witnessed the scale of data center proposals seen in other regions of the country." Electrification — electric vehicles and heat pumps — is the region's growth story, not computing.
That does not make Massachusetts buyers safe. It makes their risk different. A New Jersey buyer's exposure runs through PJM capacity costs and data-center-driven load growth; a Massachusetts buyer's exposure runs through winter gas deliverability and the Algonquin
basis blowouts we covered last month. Same company, same winter, two different risk registers — and a single national headline about data centers does not describe either one correctly on its own.
The Counterweight: Not Every Megawatt Gets Built
A balanced read has to include the other side. PJM's independent market monitor calculates that 42% of the forecasted increase in data-center load between the 2026/27 and 2028/29 delivery years had already been cancelled or delayed as of September 21, 2026. PPL
Electric, FirstEnergy, and AES Ohio lowered their near-term large-load forecasts by 6.5 GW in filings to PJM's Load Analysis Subcommittee — the same revision that could shrink the suspended Reliability Backstop Procurement target to roughly 630 MW from about 6,800 MW. Forecasts built on interconnection requests overstate what gets built; some of
what is announced never pours concrete.
That does not erase the trend. Even after the cancellations, the direction is unambiguous: load growth is back after two decades of flat demand, and data centers are the reason. But it does mean buyers should not price the extreme case as a certainty — and should not let
a single forecast justify a panicked long-term position.
What Energy Buyers Should Watch
PJM capacity auction results and the Exelon FERC complaint (Docket EL27-11) — Exelon wants prompt auctions two months before delivery; PJM asked until December 21 to answer. The auction design decides how fast data-center costs reach your bill.
New Jersey's Data Center Fair Share Act tariff — the NJBPU has until mid-2027 to set it. Whether large loads pay their own way determines how much of the $9.3 billion lands on everyone else.
Behind-the-meter gas announcements — every hyperscale gas project is new baseload gas demand, and it tightens the same supply balance your winter contracts depend on.
Friday's CFTC positioning data and the EIA storage report — the gas market's short squeeze and the storage surplus still set the near-term price, whatever the long-term load story says.
ISO-NE's quarterly large-load survey — if a real data-center pipeline ever forms in New England, the CELT forecast will show it first.
Premier Energy Management Perspective
The grid was built for a world where demand barely grew. That world is over. Data-center load is compounding at 19% a year against a grid expanding at 2%, and no amount of wishing changes the arithmetic — the only question is who pays for the catch-up, and how fast it arrives. For commercial and industrial buyers, the practical lesson is the same
one we keep coming back to: your exposure is not the national average. A New Jersey facility's power costs are being repriced by PJM capacity auctions and a 4,700 MW interconnection queue; a Massachusetts facility's exposure is winter gas deliverability and basis. Both need a strategy built around their own zone, their own load shape, and
their own contract timing — not around a headline.
The behind-the-meter turn is the part to watch most closely. If hyperscalers build 110 GW of on-site generation, gas demand gets a new structural floor that has nothing to do with weather. If they cannot, the grid share — and the capacity bill — keeps rising. Either way,
energy is no longer a line item you can set and forget. It is a strategic input, and the buyers who treat it that way will be the ones who are not surprised by the next auction, the next interconnection delay, or the next winter.
Market Bottom Line
Data-center power demand is growing ten times faster than the grid, and the gap is already priced into PJM capacity auctions, multi-year build delays, and a behind-the-meter generation boom led by natural gas. New Jersey buyers feel it in the capacity market; Massachusetts buyers should watch it as a gas-demand story, not a load story. The forecasts will be revised — 42% of the near-term pipeline already has been — but the direction is set: plan for a grid that charges more for peak, and for gas demand that no longer waits on winter.
Source: Howden
Winter is almost here...
Would you like a review of your facility's Winter Energy strategy? We are here to help!
Comments