America Just Paid $1.2 Billion to Not Build Power Plants — During an AI Electricity Crunch

Paying to build a power plant is normal. Paying someone not to build one is a different genre entirely. The US government has agreed to hand German energy giant RWE roughly $1.2 billion to walk away from its American offshore wind projects — at the exact moment AI data centers are draining the grid faster than utilities can respond.

One caveat up front: this surfaced through policy and industry channels rather than as a fully public contract, so the specific terms could shift once the paperwork is out. The shape of the thing is clear enough.

Why you’d pay someone to stop

Offshore wind burns enormous capital long before a single turbine spins. Seabed surveys. Federal lease auctions. Permits. Turbine supply agreements. Cable-laying contracts. A developer the size of RWE had years and billions already sunk into the US Atlantic coast.

When a government reverses course and halts that, it isn’t a cancellation. It’s a breach. The developer can sue, and once international investor-protection provisions enter the picture, the exposure balloons. Hence a settlement. The calculus is simple: pay $1.2 billion now rather than litigate for five years and pay more.

That money bought zero electrons. It retroactively unbuilt spending that had already happened. On paper it’s a policy-transition cost. In practice it’s a bill for stranded capital.

Meanwhile, demand is going vertical

Here’s the contradiction. US electricity demand was essentially flat for close to two decades — efficiency gains kept pace with growth. That pattern shattered in the last few years.

The cause converges on one thing: AI. A single training cluster can pull as much power as a small city, and as inference services go mainstream, that load flattens into a 24-hour baseline. Individual data center campuses now request contracted capacity in the hundreds of megawatts. Which is why utility interconnection queues in PJM and ERCOT stretch out for years, and why “we can’t get you power until 2031” has become a normal sentence in commercial real estate.

Virginia, Ohio, and Texas — the places data centers actually cluster — are already dealing with interconnection backlogs and rate-increase pressure at the same time. Northern Virginia alone hosts the densest concentration of data centers on Earth. And the RWE projects that just got erased were aimed at East Coast load centers. The federal government spent money to delete supply in the region where demand is climbing fastest.

The case against offshore wind isn’t nonsense

To be fair, the arguments for pulling back aren’t invented. Three of them are real.

Money first. Rate hikes wrecked offshore wind economics. Projects priced at 2020 interest rates stopped penciling, developers demanded renegotiated power purchase agreements or walked entirely, and several high-profile East Coast projects collapsed on their own before any政策 intervention — Ørsted’s Ocean Wind write-off being the loudest example.

Intermittency is real too. Wind doesn’t blow on a schedule. AI data centers, by contrast, want flat, uninterrupted load around the clock. That mismatch is exactly the argument now boosting nuclear and gas: the AI era needs firm power, not variable power.

And then there’s politics. Offshore wind hardened into a partisan symbol somewhere around 2022. Add persistent opposition from commercial fishing interests and coastal towns, and it became the most convenient thing to cancel the moment the administration changed.

The trouble is that all three can be true and still not answer the only question that matters: what gets built instead, and when.

The replacements aren’t coming on time

The obvious retort is: fine, build gas. Except every major turbine manufacturer — GE Vernova, Siemens Energy, Mitsubishi — is sold out. New orders carry multi-year delivery lead times, and buyers are now reserving slots years ahead of having a site. Nuclear is worse. SMRs still have almost no commercial operating record, and conventional reactors are a decade-plus proposition from groundbreaking to grid.

Offshore wind was different in exactly one way that mattered: the leases were secured, the permits were through, the turbines were ordered. It was the shortest remaining path to new megawatts on the Eastern Interconnection. Deleting that pipeline means the power that would have arrived around 2030 simply won’t.

So where does the gap go? Three places. Existing fossil plants run harder. Retirement dates for aging coal and gas units get pushed back — something already happening across PJM. Or data center operators stop waiting and generate their own. That last one explains the parade of headlines about hyperscalers signing dedicated nuclear deals and dropping gas turbines directly on-site. When the grid can’t deliver, everyone builds their own grid.

The real cost isn’t $1.2 billion

The settlement figure is the least interesting number here. The signal is what matters.

Read it from the perspective of foreign capital considering large US energy infrastructure. You can win a federal lease, clear the permitting gauntlet, deploy billions — and still get stopped when the administration turns over. You will be compensated. But you wanted a power plant, not a check.

That lesson compounds. The next project demands a higher return to justify the political risk. Risk premium raises the levelized cost of electricity, and that cost gets split between ratepayers and the data center operators bidding for the same electrons. In a race where speed-to-power and cost-per-megawatt-hour have become the binding constraints on AI buildout, this is a tech industry story wearing an energy story’s clothes.

One government, two directions

One arm of the federal government warns that AI demand is outrunning supply. Another writes a check to erase generation capacity. Both decisions came from the same administration, in the same year.

Energy policy needs a twenty-year horizon. American politics resets every four. If the deciding factor in the AI race is shifting from model benchmarks to how fast and how cheaply you can secure electricity, leaving generation planning hostage to the election cycle starts to look less like ideology and more like a self-inflicted wound. Every country courting data center investment is staring at the same question right now — and most of them are watching how this one plays out before answering it.