The data centre boom is accelerating the energy transition, and the reason is impatience, not policy. A hyperscaler that has committed billions to a campus cannot wait for a gas turbine. It needs megawatts on a timeline measured in quarters, and in most of the places where it builds, the fastest megawatts available are solar panels and batteries. The intuition that AI demand means a gas revival gets the mechanism wrong. What matters is not which fuel is cheapest in a spreadsheet but which asset can be financed, permitted, delivered and connected before the servers arrive.
The strongest evidence for this comes from a Swiss research team, Heiss et al. (2026). They mapped US data centre construction from 2022 to the end of 2025 and overlaid it on new power plant construction. Their finding contradicts the dominant US narrative. In counties that gained a data centre, renewable and battery additions accelerated much more than in counties that did not. Gas additions also rose in data centre areas, but by less than in areas without one.
Where the Servers Went and What Followed
The geography is the first clue. The study shows the buildout concentrated in the regions with the lowest power costs: Texas, Virginia, Ohio and Nevada. Developers did not scatter capacity at random. They went where electricity was cheap and land and interconnection were obtainable.
Those regions are cheap for different reasons. Texas runs its own grid, ERCOT, with a fast connect-and-manage approach to interconnection and very large wind and solar resources. Nevada has some of the best solar irradiance in the country. Virginia's cluster in the Loudoun County corridor is older and owes more to fibre density than to generation mix. The common factor is not ideology. It is the ability to add capacity quickly at a low marginal cost.
The national picture supports this. The study's capacity-additions chart shows that most new US generating capacity since the launch of ChatGPT has been renewable, with gas a small fraction. That has held through a federal administration openly hostile to wind and solar. Developers keep building renewables because new solar and storage beat new gas on cost and on time-to-power in most of the markets where demand is growing.
Speed to Power Is the Binding Constraint
The mechanism linking a data centre to a solar farm runs through three constraints, and gas loses on two of them.
- Equipment lead times. Large gas turbines are made by a small group of manufacturers, mainly GE Vernova, Siemens Energy and Mitsubishi Power, and their order books have been widely reported as filled toward the end of the decade. A developer ordering a combined-cycle plant today is buying a slot years out. Solar modules and lithium-iron-phosphate battery systems are commoditised, globally oversupplied and delivered in months.
- Construction time. A utility-scale solar project can typically go from notice to proceed to energisation in roughly a year once permits and interconnection are in hand. A battery installation can be quicker. A greenfield combined-cycle plant usually takes considerably longer.
- Interconnection. This constraint is shared, and it may be the most important one. Grid connection queues are the real bottleneck across much of the US. Here batteries do something gas cannot easily match. They can sit alongside existing solar or at substations, absorb midday surplus and release it during the evening peak, which lets a data centre's load fit into grid headroom that already exists rather than waiting for a new transmission line.
The third point explains why the battery result in the study matters as much as the solar result. A data centre is a flat, round-the-clock load. Solar on its own does not serve that load. Solar paired with four-hour storage, firmed by the existing grid overnight, can serve enough of it to let the campus energise on schedule. The battery is the asset that turns cheap but intermittent energy into usable capacity on the timeline the tenant demands.
This also accounts for the result that looks oddest: slower gas growth in data centre counties than elsewhere. If data centres select regions where fast, cheap renewable capacity is already easy to add, then those counties are by construction the places where new gas is least competitive. Gas additions concentrate where the renewable route is harder, whether because of weaker resources, slower permitting for land-intensive projects or utility planning that favours thermal plants.
Europe Runs the Same Logic on a Tighter Grid
In Europe the link is more explicit. The UK and Spain are the continent's largest data centre hubs in this signal, and both are seeing renewables and battery investment rise alongside data centre demand.
Spain is the clearest case. It has abundant solar, solar output that frequently depresses midday wholesale prices, and since the Iberian blackout of April 28, 2025, a political and regulatory push toward grid stability and storage. A data centre operator in Spain faces cheap daytime energy and an increasingly valuable flexibility market. Batteries capture both. The UK lacks Spain's sunshine but has deep offshore wind and one of Europe's most active battery markets, built on balancing-service revenues that make storage financeable without long-term contracts.
In both countries the binding constraint again is grid access, not fuel. National grid connection reforms in the UK and congested nodes in Spain mean that the operator who arrives with a storage solution, one that reduces the strain its load places on the network, can move ahead in practice even when the formal queue does not change.
The Case Against the Clean Read
The thesis has weak points, and they deserve to be stated plainly.
Nameplate capacity overstates renewables' contribution. The study measures capacity additions, not energy delivered or firm capacity at peak. A gigawatt of solar runs at a capacity factor of perhaps a quarter. A gigawatt of combined-cycle gas can run at more than half. On an energy basis, and especially on a reliability basis, the renewable share of new supply is much smaller than the capacity share suggests. A data centre consuming power at 2 a.m. in January is drawing on the existing fleet, which in Texas and Ohio includes a lot of gas.
Timing flatters renewables. The study window ends at the close of 2025. Gas plants ordered in response to the AI boom largely have not been built yet because the turbines do not yet exist. A capacity series ending in 2025 records the assets that could be built fast. It cannot record those still in the manufacturing queue. It is possible that later data show a gas wave arriving once turbine slots come due.
Behind-the-meter generation may be missing. Several large campuses have announced on-site gas generation to avoid grid queues altogether. Whether such projects appear in standard capacity statistics depends on their size and how they are reported. If they are undercounted, the gas share in data centre counties would be understated.
Co-location is not causation. Data centres and renewables may both be responding to a third factor, such as cheap land, a fast-moving grid operator or favourable state permitting, rather than one driving the other. The study shows a strong association. The causal path through procurement contracts is plausible, and it fits what hyperscalers report about their power purchase agreements, but the charts do not establish it.
Policy is changing the economics. The 2025 US tax law accelerated the phase-out of production and investment credits for wind and solar. Projects already safe-harboured will be completed. The marginal project after the phase-out has to compete without subsidy. Storage kept more favourable treatment, which strengthens the battery side of the thesis while weakening the solar side.
This last point is where the US read is most exposed. The cost advantage of new renewables over new gas is real in the best resource regions, but part of the gap the study captured was policy-supported. Without the credits, the speed advantage remains while the cost advantage narrows.
What Would Confirm or Break the Read
The cleaner reading of the evidence is that AI demand rewards whichever capacity reaches the grid first, and through 2025 that was solar and batteries in the regions where data centres chose to build. This is a supply-chain argument, not an environmental one, and that makes it both more durable and more conditional than the headline implies.
It is more durable because it does not depend on corporate climate pledges. If net-zero commitments were dropped tomorrow, the turbine backlog would still exist. It is more conditional because it is tied to that backlog. The thesis loses force when turbine lead times shorten, either as manufacturers expand capacity or as orders are cancelled.
Two observable conditions decide it. First, whether gas capacity additions in data centre counties catch up with or overtake renewable additions in 2027 and 2028, when turbines ordered during the boom begin delivering. If they do, the 2022 to 2025 pattern was a timing artefact rather than a structural shift. Second, whether battery deployment keeps pace with load growth after the solar credit phase-out. Storage is the part of the mechanism that turns cheap energy into firm capacity. If storage additions keep rising in Texas and Spain as solar subsidies fall away, the transition is being driven by grid physics, and the data centre will have done more for it than most climate policy has.
