What happens to clean energy when the AI boom cools?

The companies building artificial intelligence have become the largest private buyers of clean energy. They sign long contracts for wind and solar power, and their deals are helping to restart closed nuclear plants in the United States. Microsoft alone has placed most of the publicly reported orders for carbon removal. In 2025, Amazon, Meta, Google and Microsoft signed contracts for about half of the clean power that companies bought worldwide, while many other corporate buyers held back.

For clean energy, this has been good news. In a hostile political environment, the AI companies kept buying, and they became the first large customers for several new technologies. I would argue, though, that much of the private demand for clean energy now rises and falls with spending on AI. While these companies build data centres at speed, demand for clean power grows fast. When their spending slows, a good part of that demand will slow too. The boom is also a mixed blessing for the climate, since these buyers take power in whatever form comes fastest, gas included. Investors, other corporate buyers and the AI companies themselves should use the boom to build assets and contracts that will still find buyers when AI spending slows.

Carbon removal shows what this dependence looks like. The young companies that take carbon dioxide out of the atmosphere and store it for the long term built their plans around one buyer. By the spring, Microsoft had ordered nearly four-fifths of all the tonnes in publicly reported deals. It was also the only buyer to have signed a deal for more than a million tonnes. Its appetite for removal is itself tied to AI. Microsoft has pledged to be carbon negative by 2030, and building data centres has pushed its emissions up, so it needs more removal to keep that pledge. In April, Microsoft reportedly told suppliers that it was pausing new purchases. The company denied an indefinite pause, and it has since signed smaller deals. Whatever its reasons, a change of plan at one company left a whole industry looking for buyers, most of whom order in far smaller amounts.

The deeper problem is a mismatch in timing. A wind farm, a nuclear plant or a carbon removal facility takes years to build and decades to pay back. The AI companies run on a much shorter cycle. Their spending on data centres can rise or fall within a year, depending on how investors judge the returns. This year the four largest plan to spend about $700 billion between them, and analysts already expect that spending to grow more slowly next year. A supplier whose main customer is an AI company takes on that company’s cycle as its own risk.

The telecoms boom around 2000 shows how such a cycle can end. Carriers laid far more fibre-optic cable than demand could fill, and many went bankrupt when the boom ended. The cable they buried outlived them, and over the following decade broadband and cloud computing put it to use. The assets that kept their value were the ones other companies could use. A wind farm or nuclear plant connected to the grid can sell its power to any buyer. A plant built beside a single data centre, or a removal project with one customer, keeps its value only while that customer stays.

The pattern reaches well beyond AI. In recent years, sustainability has often advanced by riding a bigger agenda. Industrial policy paid companies to build battery and solar factories, and some of those factories were put on hold once subsidies and demand fell away. Today, much of the case for grids, storage and nuclear power rests on energy security. AI is the latest and largest of these agendas. Each brings money and speed. Each also leaves sustainability exposed when the governments or companies behind it change their priorities. For boards, the lesson is to use that momentum while it lasts, and to build assets and commitments that can stand on their own once it fades.

Some will object that the AI companies are the strongest customers any supplier could want. They are among the richest firms in the world, and the International Energy Agency expects data centres’ electricity use to double by 2030. I accept that they are rich and that their need for power will grow. A lasting need for power, however, differs from a lasting need for clean power, since the same firms are also turning to gas. Suppliers can count on these buyers to honour the contracts they have signed. They can count far less on the next round of orders, which young industries need in order to grow. Others will add that companies buy only a small share of the world’s clean power. That share, though, includes the first customers for young technologies such as carbon removal and advanced nuclear.

Three groups of businesses can use the boom to build assets that outlast it. Investors and lenders should treat dependence on one customer as a risk to be priced, as they would in any other industry. They should favour projects on the shared grid, which can sell to many buyers. Other companies with climate targets should see an opening. They can buy alongside the AI companies, alone or through buying groups such as Frontier for carbon removal, and secure supply on good terms. Their orders also give young markets more than one large customer. The AI companies’ own boards have an interest in this too. They will need clean power and carbon removal for decades, and a supplier base that depends on them alone is fragile. Buying steadily over time, and signing contracts that suppliers can pass to other buyers, would help those suppliers survive a slowdown.

The AI companies became clean energy’s most powerful customers just as other buyers were holding back. Carbon removal suppliers have already seen what happens when the largest of those customers slows its orders. When the wider boom cools, grid-connected wind farms and nuclear plants are likely to find other buyers, much as the fibre did. The task for business now is to make sure that more of what the boom builds can do the same.

illuminem Voices is a democratic space presenting the opinions of leading Sustainability Thought Leaders, their views do not necessarily represent those of illuminem.

The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.