I think we can actually solve the climate crisis now.

June 23, 2026

Last week, Anthropic became the first pure-play AI company to join Frontier, the carbon-removal buyers’ club that Stripe, Google, and Shopify started in 2022. The new commitments doubled the pool to $1.8 billion. The headline badly undersells what happened. This is kind of the final chord in a three-chord progression build-up, where the chorus is going to be a cascade of technologies that repair the natural environment.

1. AI made the hyperscalers the biggest energy buyers on Earth.

The race for AI is, underneath, a race for power. Training a frontier model takes a staggering amount of electricity, and the hyperscalers are in an existential fight for compute. A single gigawatt-scale data center pulls more electricity than the entire state of Rhode Island. Construction has exploded in multiple states, and hit bottleneck after bottleneck that was overcomeable with money, until it hit one that wasn’t: energy.

So the hyperscalers have poured on the pressure, and tried to solve the bottleneck by buying energy in every tense at once. Whatever the utility can sell them today. Temporary gas turbines to bridge the gap. Their own on-site solar and geothermal. And the long-dated bets: small modular reactors, fusion, a restarted nuclear plant or two. Amazon, Google, Microsoft, Meta, and Oracle are on track to spend more than $600 billion on infrastructure in 2026 alone, over four times what the entire US energy sector spends drilling, refining, and shipping fossil fuels. Big Tech has become the primary financier of the energy transition, almost by accident.

2. They can’t unplug from the grid, and they’ve decided they don’t want to.

For a while it looked like the hyperscalers might just island themselves: build private power, skip the interconnection queue, move fast and apologize later. xAI’s Memphis site did exactly that, with unpermitted gas turbines, and got sued by its neighbors for it.

Then they changed course. As Jigar Shah, who ran the Department of Energy’s Loan Programs Office, has pointed out, Google, Meta, and AWS have quietly converged on the same conclusion: stay connected to the public grid. For Google it’s because of the carbon-free-energy commitments. For Meta it’s economics. The redundancy, logistics, and permitting of true off-grid generation eat the supposed advantages. For AWS it’s that the workload itself is shifting: they believe that inference, not training, will be the overwhelming majority of compute by 2030, and inference runs fine on smaller, distributed, grid-tied sites. Three competitors for three reasons arrived at one answer.

Staying on the grid changes the power dynamic. Electricity bills have become a kitchen-table political issue, and a data center that puts stress on a shared grid is an easy villain. The public, through its utilities and its governors, has a seat at this table now. The hyperscalers know it. The smart ones will show up with something to offer.

3. So they’ll start offering something back, and carbon removal is the perfect thing to offer.

Put yourself in the shoes of a company that needs social license and has more capital on hand than almost anyone in history. What do you offer the public in return for the right to keep building? Carbon removal is close to a perfect answer.

It’s the textbook public good. Everyone benefits from a cooler planet, which is exactly why no single buyer ever had a reason to pay for the cleanup, which is exactly why it has stalled. It needs a patron.

It’s acutely needed, and we are so early in its deployment that we could see measurable gains off achievable investments. Durable carbon removal delivered worldwide last year came to about five million tons. The job, by mid-century, is tens of billions of tons a year. We are at something like two-hundredths of one percent of scale.

And, it can run on energy a data center has and would otherwise go to waste. Pulling CO₂ out of the air is energy-hungry and time-flexible. Data centers that run on intermittent sources like solar and wind have to overbuild substantially and use storage, like batteries. Inevitably some of that energy must be curtailed. And, regardless of energy source, data centers create waste heat, pouring off server racks. A hyperscaler can fund, or even physically co-locate, removal without diverting a single electron from a training run. The goodwill is enormous; the cost in compute is close to zero.

That’s what makes Anthropic joining Frontier such a positive signal. Plus, Frontier said that going forward, it will fund fewer, larger projects, held to a higher bar: the ones with a credible path to removing a billion tons a year. Concentrated, durable, high-quality demand like this is the signal that drags a technology down its cost curve. It is the mechanism that turned solar from a luxury into the cheapest electricity in human history. Economists call it Wright’s law: every doubling of cumulative production knocks a predictable percentage off the price.

The trade

The data centers need to keep building, fast, and they need the public to let them. The public needs the carbon pulled back out of the sky and has never been able to find anyone to pay for it. Each side has problems, and they are the answer to each other.

The climate problem is starting to look less like a tragedy and more like a market that hasn’t finished forming yet. It does not matter whether Anthropic joined Frontier to save the world or to look good doing it. The carbon comes out either way.

I think we can actually do this now.

The full DAC breakdown (operators, durable storage, 45Q, and the supply-vs-demand gap) lives at moreenergycleanplanet.com/sectors/dac.

More Energy, Clean Planet is out January 19, 2027. Pre-order →