AI data centers need power faster than the grid
IEA and Rystad say AI data centers need power faster than the grid. On-site gas and fuel cells fill the wait. The Information financing angle is a RUMOR.
The IEA and Rystad describe a grid delay. The Information’s “financing opportunity” remains a RUMOR.
On 30 September 2026, at 13:30 GMT (10:30 in Brasília), @theinformation posted on X a teaser for a paywalled Special Report: AI data centers need power faster than the grid can often provide. The public title is “How Investors Are Financing On-Site Power to Break AI Bottlenecks.” The body, the deals, and any financing figure were not verified. The International Energy Agency (IEA), in its April 2025 Energy and AI report, and the consultancy Rystad Energy, in a 25 June 2026 note, describe the same timing gap. The chips can be ready while the plug is not.
First: what on-site generation means here
On-site generation is a plant or modules on the data center’s land, so extra electricity does not wait in the utility queue. Picture a generator in the backyard instead of a new high-voltage line across the state. The kit is usually a gas turbine, which burns gas and spins a generator, or a fuel cell, which turns gas into electricity through a chemical reaction, in smaller blocks. The comparison fails if the gas or the transformer is also late. A grid hookup is still the usual path. The issue is the wait.
The Information’s public teaser
The public line says this creates a financing opportunity for investors backing gas plants, fuel cells and other on-site generation. RUMOR: the “financing opportunity” is The Information’s editorial angle. No funds, ticket sizes, or named projects were verified. This newsroom did not read the Special Report behind the paywall.
What the IEA measured
The IEA estimates data centers used about 415 TWh in 2024, roughly 1.5% of world electricity. A terawatt-hour (TWh) is 1 billion kilowatt-hours. In the agency’s base case, that use more than doubles, to about 945 TWh in 2030. World generation aimed at these buildings, a different tally, rises from 460 TWh in 2024 to more than 1,000 TWh in 2030. One figure is what the buildings draw; the other includes losses on the way.
Without easing grid risks, about 20% of planned data center projects may slip, the IEA says. New transmission lines take 4 to 8 years in advanced economies. Gas turbines for new plants already have lead times of several years, with a chance they come online after 2030. Natural gas and coal together are expected to meet more than 40% of the extra demand by 2030. In the United States, in the IEA’s 2025 cut, gas is already the largest source for data centers (more than 40%) and the largest extra slice through 2030 (more than 130 TWh a year). Worldwide, gas grows by 175 TWh for this demand. The IEA still has renewables covering about half of extra global growth. Gas shows up in the 2025-2030 window, especially in the U.S.
AI data centers, gas, fuel cells and the queue
Rystad Energy sells energy intelligence to clients. On 25 June 2026, the firm said developers are leaving congested grids for on-site fuel cells. According to the consultancy, market revenue would go from about $2.8 billion in 2025 to about $30 billion in 2030, with a contracted order book of about 9 GW and framework agreements involving Oracle, AEP, Equinix and Brookfield. The public note does not list megawatts per customer. U.S. interconnection waits (tying a large load to the grid) have tripled since 2015 and now run 3 to 6 years. The firm projects 10.4 GW of cumulative fuel-cell demand at data centers between 2026 and 2030 and models about 40% of U.S. data center capacity in 2030 as likely to use dedicated on-site generation. In that cut, North America is about 91% of global installed on-site generation capacity.
The note treats solid oxide fuel cells (SOFC) as the main technology for round-the-clock load and says Bloom Energy holds virtually every visible primary-load SOFC contract. Combined manufacturing capacity, operating plus planned, would go from 1.8 GW a year “today” (June 2026 in the note) to 4 GW a year in 2030. Rystad itself wrote: “Power availability has become one of the defining constraints on data center growth.”
Limits, and what we still do not know
On-site generation is a detour. The IEA still puts about 20% of projects at risk of delay, and gas turbines have a multi-year wait of their own. Rystad’s time edge for fuel cells is a commercial forecast, not a site-by-site test. A concentrated supplier (Bloom, in Rystad’s visible book) and near-term gas emissions remain. This newsroom did not audit 8-K filings or contracts.
This newsroom has not seen the Special Report’s body, and has no independent megawatt figures for those framework agreements. The capital The Information describes behind the megawatt bottleneck is still a RUMOR. The delay on the wires is not. Campuses want power now. A plant on the same land can shrink the wait in some cases. On the IEA’s own numbers, it does not remove the delay for the whole set of planned projects.
Sources
Official
Consultancy
Press (teaser; body not verified)