Are Data Centres Part of the Energy Transition?

They are financing it and straining it at the same time. So which one wins?

By Joe Hetherington, Managing Partner, Green Executives

Ask anyone in this sector privately what they think about data centres and you will get a straight answer. Ask them publicly and you get a press release.

It’s a difference worth paying attention to, as it usually means the honest answer is uncomfortable for somebody's client list.

So here is mine.

The Power of Attraction 

My last piece argued that the line between infrastructure investors and private equity has stopped meaning anything. Plenty of people agreed, but there’s more to unpick.

If the categories have collapsed, what is all that capital actually buying?

Increasingly, the answer is power for computers.

The AES take-private ran to $33.4bn, and part of what the buyers were paying for was 11.8 GW of clean energy agreements signed with technology customers. EQT bought Copia Power from Carlyle and described it as an energy and AI infrastructure platform. 

The traffic runs both ways, too. Google has contracted a gigawatt of Texas solar with TotalEnergies and AES, the same AES now being taken private largely on the strength of contracts like that one.

The convergence trade everyone has been writing about is, to a significant degree, a compute trade. It is worth being honest about that before deciding how we feel about it.

Reversing Energy Demand; the positives

For twenty years electricity demand in developed markets was flat or falling.

That single fact made a lot of things uninvestable. If demand is not growing, nobody needs new transmission. Nobody needs long-duration storage. Nobody is building a nuclear plant.

Data centre demand reversed that, and it reversed it fast.

The tech giants are now paying for generation that utilities would never have risked building. The mechanism matters here: a twenty-year power contract from an investment-grade counterparty is what turns a project from an idea into something a bank will lend against. A regulated utility working under cost-recovery rules cannot offer that. Microsoft and Amazon can.

That is why nuclear restarts are back on the table. It is why storage projects that would have died in committee three years ago are being built.

Look at how Google does it. It does not, for the most part, own generation. It buys — long-term contracts across wind, solar, hydro, nuclear and, more recently, enhanced geothermal. And the collective weight here is not marginal. Amazon, Meta, Google and Microsoft between them accounted for 49% of every corporate clean power agreement signed anywhere in the world in 2025.

The geothermal example is the one worth studying, because it is the cleanest case of additionality anywhere in this sector. Enhanced geothermal was a laboratory proposition. It became a commercial one because somebody agreed to buy the output — Google contracting with Fervo Energy is a large part of why the first projects exist at all. Nobody can argue that generation was happening anyway. It was not.

That is what a data centre contract does at its best, and it is worth being precise about it, because it sets the standard everything else should be measured against.

Anyone who tells you data centres have contributed nothing to the transition is not paying attention.

Delaying the Transition: the negatives

Now the other side, which is equally real.

The transition means replacing fossil generation. This new demand is landing on systems that were already tight.

The consequences are documented. A Reuters analysis of company filings across PJM, the largest grid in the United States, found roughly 60% of oil, gas and coal plants scheduled for retirement have delayed or cancelled those plans. The EIA logged 4.8 GW of a planned 8.5 GW of retirements pushed back, with two coal plants abandoning closure altogether.

Dominion named projected data centre demand in its own resource plan as a reason to keep plant running, including a coal-fired peaker. Utilities across Virginia, Georgia and the Carolinas have proposed 20,000 MW of new gas by 2040.

Meanwhile clean energy projects sit in the PJM interconnection queue, waiting.

There is a second problem that gets less attention. Most corporate clean power claims are matched annually. A company buys enough renewable generation across the year to cover what it uses across the year, and the certificates balance. But a data centre runs at three in the morning in February, and at three in the morning in February the wind may not be blowing. What is actually keeping the servers on at that moment is whatever the grid is burning. The accounting is honest. It just describes a year rather than an hour.

That may not survive much longer. The GHG Protocol, the standard the whole of corporate carbon accounting rests on, is revising its Scope 2 rules, and the proposals under discussion would require hourly tracking and tighter geographic boundaries. Under an hourly regime, most 100% renewable claims become considerably harder to stand up.

The market has already noticed. BNEF tracked 5.8 GW of co-located and hybrid deals in 2025, structures built to deliver power at the hour it is used rather than across the year. Anyone underwriting a clean power claim today should assume the rules it was written under are about to move.

The question that settles it

Strip all of that back and one question decides which story is true for any given site.

Did the wind farm get built because the data centre signed a contract for its power? Then the data centre paid for clean energy that did not exist before. That is a real contribution.

Was it getting built anyway, and the data centre simply signed the contract first? Then nothing new was added. The clean power moved. Everyone else on that grid is now drawing from what is left, which may be dirtier than it was, while the data centre reports itself as running on renewables.

And where the grid is already tight, the effect goes further. A plant that was due to close stays open, because the demand it was going to stop serving has not gone away. It has grown.

Both of these are happening, at scale, right now.

Here is the part that should bother investors. Which one applies to a specific asset is a knowable fact. Someone in the room knows. It is almost never written down, and it is almost never asked for.

The question of water use

Water is the same problem, further back.

Carbon has one unit and three scopes, making disclosure comparable. Water has three different numbers that all get called "water use" — what you take, what you give back, and the difference between them. Reporting just one of these won’t reveal the full picture. 

Now look at which one gets measured.

The industry metric covers the water cooling the building. It does not cover the water used at the power station making the electricity. That is someone else's meter, in someone else's catchment, sometimes in another country.

Run the numbers on one megawatt of computing for a year. An efficient site uses around 1.1 million litres cooling itself. If its power comes from wet-cooled gas, that same megawatt takes another 12.6 million litres at the generating plant.

The water nobody counts is ten times the water everybody quotes.

In carbon we have a name for this. Emissions from bought-in electricity are Scope 2, and companies report them. In water there is no Scope 2. There is no name for it at all.

That is starting to change, slowly. The EU already makes operators above 500 kW report their water use annually. WRI, WWF and the CEO Water Mandate started building water scopes in April 2026, modelled directly on the carbon protocol, with data centres named as a priority sector — final version due late 2027. The UK requires nothing at all. Meanwhile Global Action Plan puts 84% of proposed UK data centres in areas already water stressed, or heading that way by 2040.

Four questions for an investment committee

If you are underwriting a data centre, or generation that serves one, or a platform holding both, these are the questions that separate the two stories.

Is this clean power new, or just moved? Did the project get built because of this contract, or was it happening anyway?

Does the clean power arrive when it is used, or just over the same year? Hourly matching and annual matching describe very different things.

Who was in the grid queue first? Connections are rationed. If this site got one, something else did not. What was it, and how long had it been waiting?

Is that water figure the building, or the power station too? Ask for both. Ask which catchment. Ask whether it is drinking water.

None of these are difficult questions. They are simply not being asked, and the answers are not being volunteered.

Where I land

Data centres are currently the largest single source of finance for new clean generation anywhere in the world. They are also the least examined claim in this sector.

Both are true, and they are true for the same reason: the disclosure that would settle the argument does not exist yet.

So my position is not that data centres are good for the transition, or bad for it. It is that we are taking an enormous amount on trust, and the industry has been given no particular reason to volunteer the answer.

Until that changes, treat clean power claims as unaudited. I’m not saying ‘dishonest’, there is a difference, and it is the difference between a claim you can price and a claim you cannot.

The sector spent twenty years learning to measure carbon. We are at the very beginning of that with water, and barely further along with additionality.

We should probably get on with it.

Joe Hetherington is Managing Partner at Green Executives, a B Corp-certified executive search and strategic advisory firm operating exclusively in the green economy.