The Trillion-Dollar Bet Against the Climate
The AI build-out that’s walking past every rule built to green it.
Grüezi!
I moderated a panel of thoughtful green finance people this week at SXSW London.
Whenever you moderate, it’s best to let the panellists do the talking.
But here’s what I was thinking.
1. What we were promised
When it came to tackling climate change, we were told the market would take care of it.
Give long-term investors the carbon on their books and the risk in their models, and their generational time horizons would do the rest: patient money, willing to hold assets over decades, would price in rising sea levels and melting glaciers and steer the world to a brighter future.
Mark Carney had a good phrase for it – solving the tragedy of the horizon. Larry Fink told the world’s chief executives that climate risk was investment risk.
And whilst Europe’s governments talked a good game they built LNG storage, kept faith with combustion engines, and left it to the European Union to spend the better part of a decade writing the investment rulebook that would get them off the hook – a whole set of lists and categories of what counted as green and disclosure rules to make it visible.
Brussels worked on the premise that information plus self-interest would bend capital the right way without anyone having to legislate the outcome.
The promise was decarbonising without politicking. You only had to let the smart, patient money see clearly, and it would make the right choices.
Then the largest pool of capital in a generation actually moved, and the theory was put to the test.
2. The test arrives
The test is AI data centre construction. The world’s five largest technology firms spent more than $400 billion building data centres in 2025, and they’ll spend around three-quarters of a trillion in 2026.
The IEA says that capital expenditure now exceeds the entire global investment in finding and producing oil.
This is the gold rush, and the patient money has proved just as fast to grab its shovelfuls of cash and throw them down the hole of the largest concentrated tech build-out of the century.
Among the funders are sovereign wealth funds that don’t answer to quarterly calls and can hold positions for generations, the very investors whose long horizons were supposed to make them climate sensitive, hustling alongside the private-credit funds that have moved in next to them.
Meta’s $29.5 billion campus in Louisiana barely belongs to Meta. It sits in a separate company four-fifths owned by an outside investment fund, financed almost entirely by debt, and kept off Meta’s balance sheet.
That is the new infrastructure model. Hyperscaler demands sitting on private-credit balance sheets with public consequences for grids and resources.
3. The money chose speed
The money chased the cheapest electricity, the fastest planning permission and the easiest water, and where the cleanest option was also the slowest it took the dirty one without hesitation.
The reason is time.
Gas turbines, fuel cells and reciprocating engines sited on a campus can deliver power in roughly 16 to 30 months; a grid connection that might bring renewables with it runs 36 to 84 months, and 84 in Columbus, Ohio.
So gas gets built.
Crusoe’s site for Stargate in Abilene and Meta’s Socrates facility in Ohio run on dedicated on-site gas, and American orders for large gas turbines have hit a twenty-year high. Globally gas turbine orders went up 70 per cent in 2025.
Brochures call this bridge power, a stop-gap until renewables kick in. But turbines have long life spans. And gas can stay cheap.
The patient capital that was meant to wait for the clean option turned out to have no patience at all.
It was patient about returns, not carbon – and that distinction, which the whole theory had quietly forgotten about, turned out to be the only one that mattered.
4. Sited beyond dissent
Set those choices side by side and a thread runs through them that cost alone does not explain. The spend went where the decision could be taken cleanly, without a contest it might lose.
In the Gulf it is speed and scale. Abu Dhabi’s Stargate campus is planned for five gigawatts on a blend of nuclear, solar and gas. Its electricity comes in at five to six cents a kilowatt-hour against nine to fifteen in the United States.
In the West it takes the form of diffusion: the cost is real, but it’s spread across people who cannot easily combine against it. In PJM, the largest electricity market in the United States, which runs the grid for 67 million people from New Jersey to Illinois, capacity prices rose from $29 per megawatt-day in 2024/25 to $329 in 2026/27. At least 40 per cent of that is down to data centres, according to the official market monitor.
The same thing goes for water. In Fayetteville, Georgia, residents complaining of low water pressure discovered that nearly 30 million gallons of unbilled water was going to a QTS data-centre development in the middle of a drought.
The bargains struck to support these centres are often done with utilities before the public understands the scale and impact. Then they’re asked to simply absorb them.
Centralised speed, diffused costs. You get the same result both ways – the decision is removed from any effective objection.
5. Why the market couldn’t
Green finance acts on an asset: it certifies the building and the company that owns it.
A data centre emits almost nothing on site. Its carbon emissions are decided one level up — by what powers the grid it draws on, and whether clean generation connects in time to serve it. That sits in the power system, upstream of anything green labels can reach.
And the one thing green finance can offer – cheaper capital – is not what the clean option was short of. It was short of time: a place near the front of an interconnection queue that no bond can buy.
So even the investor who wanted to choose clean could not make the numbers work against a gas plant standing in eighteen months.
Inside the market, greening the build-out was the slow, expensive, losing move. The theory had asked self-interest to choose against itself, and self-interest declined.
6. Why nations fail
If the market can’t deliver, the natural recourse is to politics and the state.
But the same qualities that drew the build-out to cheap energy and compliant planning are the ones that make it mobile, and that mobility is what turns governments against one another.
Tighten the rules in Ireland and the campus goes to Texas; tighten them in Brussels and it goes to the Gulf.
In a world of mobile capital and competing jurisdictions any attempt to impose national climate conditions risks being arbitraged out of existence the moment it is imposed.
The only politics that could bind this build-out is a politics that doesn’t stop at a border – coordinated across the jurisdictions the capital can move between.
That’s the joined-up politics the moment demands, and the only kind that could work.
7. The world we’re dismantling
That old rules-based order is what we are currently seeing fall apart in front of us. Larry Fink has retired “ESG” in the face of politics.
The European Union, the one body that came closest to a binding architecture, has just pulled an estimated 80 per cent of companies out of its reporting scope through the Omnibus package – narrowing the rules at the exact moment the build-out would have tested them.
Meanwhile, from Gulf capitals to American statehouses, governments are competing to host on the least demanding terms.
Everything points the same way: not towards the coordinated politics that could green the largest capital event of the era, but away from even the fragments of it that we already had.
So the lesson is that the one thing capable of greening this generational investment is the one we’re actively dismantling, right at the moment it is most needed.
And we’re dismantling it whilst our political leaders race to undercut one another on the very terms that might have made the difference.
The promise was that we could govern a transition this large without politics, by letting patient capital see clearly and choose well. It saw clearly. It chose gas.
What this build-out has shown is not that the market failed at the margins, but that the depoliticised theory of governing a transition fails exactly where the transition is largest – which is the only place governing it was ever the point.
Thanks for reading!
Bis bald,
Adrian