IEA Sees Data Center Power Doubling by 2030 as Grid Delays Threaten One-Fifth of the Buildout
The International Energy Agency published a Nov. 19 scenario that doubles data center electricity to about 1,000 terawatt-hours by 2030, while its measured data show five-to-ten-year grid waits, so energy holds flat.
By Ryan Elliott Dennis · 7 sources · 8 min read

The International Energy Agency published its World Energy Outlook 2026 on Nov. 19, 2026, and one scenario in it doubles the electricity that data centers use, to about 1,000 terawatt-hours a year by 2030, CNBC reported 2. Data centers consumed 485 terawatt-hours in 2025, by the agency's own count from April 3. Grid congestion puts one-fifth of the global buildout at risk of delay, a February report from the same agency found 4.
A terawatt-hour, or TWh, is a billion kilowatt-hours. Carry one test through this piece: a terawatt-hour counts as supplied once a meter has recorded it. Queues count requests, and scenarios count conditions.
What did the IEA publish on Nov. 19?
Most of the IEA's release concerns the Hormuz crisis and oil 1. Data centers appear once, in a list of what drives electricity demand, beside electric motors, air conditioners, electric vehicles and heat pumps. The agency's website returned an access challenge to this desk on Nov. 19, so the scenario figures below come from CNBC's report and the director's words from the release as Mirage News reprinted it.
Natalie Feary of CNBC reported the data center numbers 2. In the Stated Policies Scenario, which rests on measures governments have proposed and have yet to adopt, data centers consume around 1,000 TWh a year by 2030. Peak electricity demand rises 40 to 50 percent by 2035. Grids add 26 million kilometers of new lines by that year, about 16 million kilometers of existing lines need replacement or refurbishment, and grid spending rises 75 percent over the next decade under the Current Policies Scenario, which uses rules already in force.
Two sentences in CNBC's account describe the present. "In several regions, the combined capacity of data centers waiting to connect exceeds current peak demand," CNBC wrote, citing the International Energy Agency. Supplies of copper, cables and transformers are already tight, CNBC added, citing the agency.
How should a reader hold a number as large as 1,000 TWh? Fatih Birol, the IEA's executive director, answered in the release. "By providing multiple scenarios - none of which are forecasts - and analysis of unparalleled scope, it can serve as a reference as countries consider their next steps," Birol said 1. Read the clause between the hyphens. Birol placed the limit inside the sentence that promotes the report, so the caveat travels with the claim. A scenario states what follows if its assumptions hold.

What the meters recorded in 2025
Measured figures come from an earlier document. The agency's special report "Key Questions on Energy and AI," typeset in April 2026 and directed by Laura Cozzi, the agency's director for sustainability, technology and outlooks, counted what data centers consumed 3. Global data center electricity demand grew 17 percent in 2025. Consumption reached 485 TWh, up from around 415 in 2024, a gain of 70 TWh by this desk's subtraction. AI-focused data centers grew faster, at 50 percent, and satellite tracking showed that data centers built for AI more than tripled in capacity in 18 months.
Seventy terawatt-hours is the proof on the supply side. Each one was generated, carried over a connection and metered at a site. Growth also matched what the agency had expected, since the report calls the 17 percent "in line with IEA projections."
April's report already projected a doubling: 485 TWh in 2025 to 950 in 2030, around 3 percent of global electricity demand. So the Nov. 19 figure of about 1,000 extends a line the agency drew seven months earlier. Does that path ask more of the grid than 2025 delivered? In rate terms it asks slightly less, about 15 percent a year compounded against 17, by this desk's arithmetic. Volume is the harder measure: 515 TWh across five years averages 103 a year, against the 70 delivered in 2025.
Data center electricity: two metered years against two 2030 projections
ComparedThe case for a step down: grids move slower than data centers
Time is the agency's own objection. Its Electricity 2026 report, from February, sets two clocks side by side: a data center takes one to three years to build, and new grid infrastructure takes five to 15 years to plan, permit and complete 4. One-fifth of the global data center buildout is at risk of delay from grid congestion, that report found. At least 150 gigawatts of queued data center projects sit in advanced stages worldwide.
Equipment adds to the wait 3. Transformers carry lead times of two to three years on average, and gas turbines can take around five. A grid connection can take five to ten years in many jurisdictions. Texas shows the scale: requests from large loads to join the grid run by the Electric Reliability Council of Texas, ERCOT, stood near 63 gigawatts in December 2024, 130 four months later and over 230 by January 2026, against a record peak demand of 85.
Contracts carry late dates too. Data center operators hold offtake agreements with small modular reactors, called SMRs, for 45 gigawatts, up from around 25 at the end of 2024, and the first projects are expected online around 2030. Developers constrained by slow grid connections have turned to on-site gas plants. Satellite images covering 50 gigawatts of those plants showed 75 percent still in pre-construction in February 2026.
Set those facts beside the ledger's question: is the power contracted, interconnected and generating? Forty-five gigawatts of reactor power are contracted and generate from around 2030. One-fifth of the buildout may wait on wires. Read alone, that record argues for a step down.

ERCOT's large-load queue against the Texas record peak
ComparedHow much of the queue is real?
Jonathan Koomey doubts the inputs. The researcher spoke at the Ohio Manufacturers' Energy Conference on Aug. 27, 2026, according to the Ohio Manufacturers' Association 5. "When you see a forecast, you think, well, that's going to happen. But it turns out forecasts are not reality," Koomey said. Developers approach several utilities about one site, he told the room, so a single prospective load can be counted more than once. He cited James Burke, chief executive of Vistra, who has said queues may be overstated by three to five times.
Ryan Augsburger, the association's president, drew the customer's conclusion in the same release. "If utilities and grid operators want to spend billions chasing projected demand, then prove the demand first," Augsburger said. Risk is his subject: customers get stuck with it, he said, when forecasts fall apart.
Birol and Koomey spoke 84 days apart and chose nearly the same sentence. One runs the agency that publishes the scenarios, and the other criticizes how utilities use such numbers.
February's IEA report sides with the doubters on queues 4. U.S. utility data indicate that around 20 percent of data center connection requests materialize in the short to medium term, the International Energy Agency wrote, because demand queues are "artificially expanded by duplicative or phantom projects which are not expected to materialise."
ERCOT measured its own version 6. In its 2025 long-term load forecast, the grid operator said average peak consumption per site was 49.8 percent of the megawatts requested, Data Center Knowledge reported on May 14. Ihab Osman, an independent strategist on data center infrastructure, gave that outlet a working rule. "A large share of the current AI/data center planned load should be treated as paper megawatts until it is validated through physical gates," Osman said. Alison Silverstein, a former senior adviser to the chairman of the Federal Energy Regulatory Commission, said "a large proportion" of projects in ERCOT's queue have already been canceled.
Utility Dive reported a matching national estimate in December 2025 7. Grid Strategies judged utility forecasts of 90 gigawatts of added U.S. data center load by 2030 likely overstated and put the figure closer to 65. Patricia Taylor, director of policy and research at the American Public Power Association, said developers commonly "shop around" one project across neighboring jurisdictions.
Here the step-down case loses weight. A 230-gigawatt queue against an 85-gigawatt peak looks like stranded demand. Applied to Texas as an illustration, the national rate of one in five would leave about 46 gigawatts of likely load, by this desk's arithmetic. A queue measures applications.
Fatih Birol and Jonathan Koomey on what a forecast is worth
Both sidesThe data2 rows · sources
| Side | Who | Claim | Source |
|---|---|---|---|
| For | Fatih Birol | By providing multiple scenarios - none of which are forecasts - and analysis of unparalleled scope, it can serve as a reference as countries consider their next steps. | [1] |
| Against | Jonathan Koomey | When you see a forecast, you think, well, that's going to happen. But it turns out forecasts are not reality. | [5] |
Why energy holds flat
Three kinds of numbers appear in this story. Metered: 485 TWh in 2025, 70 more than the year before. Queued: over 230 gigawatts in Texas and at least 150 in advanced stages worldwide. Projected: about 1,000 TWh by 2030. Only the first kind answers the ledger's question, and it is seven months old.
Nov. 19 added a scenario and restated constraints the agency had already measured 2. A new count of megawatts contracted, connected or generating has yet to appear on the pages this desk could open. Scenario evidence is inferred, a class that moves the reading 0.1 to 0.5 points when it points one way.
This evidence points two ways. Delivered growth of 17 percent says supply kept pace through 2025. Waits of five to ten years, a fifth of the buildout at risk and reactor power dated to 2030 say the next 515 TWh is harder. Energy holds flat, from a reading of 8.6.
Treat 1,000 TWh as a condition and 485 as a measurement. Utilities, lenders and AI developers that plan around the doubling should ask which of the 515 TWh already has a connection date.
The IEA's Nov. 19 outlook leaves energy flat
The moveThe data5 rows · sources
| Measure | Value |
|---|---|
| Reading before this day | 8.6 |
| This piece's move | 0.0 (Energy, inferred) |
| Band for inferred evidence | 0.1 to 0.5 |
| Reading after the day (with 2 other pieces that day) | 8.6 |
| Distance to 100 | 91.4 |
What the IEA metered, and what it projects or counts in queues
The recordThe data7 rows · sources
| Column | Item | Source |
|---|---|---|
| Metered or observed | Data centers consumed 485 terawatt-hours in 2025, up 17 percent | [3] |
| Metered or observed | AI-focused data centers used 50 percent more electricity in 2025 | [3] |
| Metered or observed | Satellite images show 75 percent of on-site gas capacity in pre-construction | [3] |
| Projected, queued or contracted | About 1,000 terawatt-hours a year by 2030, Stated Policies Scenario | [2] |
| Projected, queued or contracted | 26 million kilometers of new grid lines by 2035 | [2] |
| Projected, queued or contracted | Over 230 gigawatts of large loads in the Texas queue | [3] |
| Projected, queued or contracted | 45 gigawatts of reactor offtake deals, first power around 2030 | [3] |
By the numbers
- 485 TWh of electricity went to data centers in 2025, up 17 percent from around 415 3
- About 1,000 TWh a year by 2030 is the Stated Policies Scenario figure, as CNBC reported it 2
- One-fifth of the global data center buildout is at risk of delay from grid congestion 4
- Five to ten years is the wait for a grid connection in many jurisdictions 3
- Texas's large-load queue passed 230 gigawatts by January 2026, against a record peak of 85 3
- Around 20 percent of U.S. data center connection requests materialize in the short to medium term 4
- SMR offtake agreements reached 45 gigawatts, with first projects expected around 2030 3
- 26 million kilometers of new grid lines by 2035 sit in the Stated Policies Scenario 2
What to watch
The IEA's first measured figure for 2026 consumption settles the near question: about 560 TWh would match a 15 percent path, by this desk's arithmetic 3. ERCOT has said it may adjust its long-term forecast for actual realization rates, and that revision will show how much of 230 gigawatts survives 6. Transformer lead times under two years, or an SMR delivering to a data center before 2030, would argue a step up. A metered year of growth under 10 percent would argue a step down. Until a meter reports, this ledger counts 485.
Sources
- 1Duration of Unprecedented Shock to Shape Energy Shift, International Energy Agency (via Mirage News), International Energy Agency, Nov. 19, 2026
- 2Data center electricity use could double by 2030, IEA says, CNBC, Natalie Feary, Nov. 19, 2026
- 3Key Questions on Energy and AI: World Energy Outlook Special Report, International Energy Agency, Thomas Spencer and Siddharth Singh, April 16, 2026
- 4Electricity 2026: Analysis and forecast to 2030, International Energy Agency, Feb. 6, 2026
- 5OMA Says 'Enough' on Data Center Forecasts Nobody Can Prove, Ohio Manufacturers' Association, Dave O'Neil, Aug. 27, 2026
- 6Gridlock or Growth? ERCOT Warns Texas AI Power Boom May Not Materialize, Data Center Knowledge, Shane Snider, May 14, 2026
- 7U.S. data center power demand could reach 106 GW by 2035: BloombergNEF, Utility Dive, Brian Martucci, Dec. 3, 2025