In 2020, Microsoft made a splashy pledge to be carbon negative by 2030. By purchasing carbon offsets and, crucially, cutting its own emissions, the tech giant committed to not just be a net positive for the climate but to “remove from the environment all the carbon the company has emitted either directly or by electrical consumption since it was founded in 1975.”
Then came AI.
AI is already consuming monstrous amounts of electricity, changing the equation for tech firms balancing the twin priorities of managing carbon emissions while staying competitive in the AI race. The data centers that store information and host the large language models that AI is trained on are stocked with high-powered computer chips that require lots of energy and lots of water to cool.
Read more: I sold a $1.4B big-data startup to IBM—then founded a nature sanctuary. Here are the dangers of AI energy consumption
Some analysts predict AI could consume as much as a quarter of all power in the U.S. by 2030. Microsoft disclosed last September that its water usage had spiked 34%. The power grid as it currently exists isn’t capable of handling such a massive surge in demand, so tech companies are taking matters into their own hands.
This morning, Microsoft announced a $10 billion deal with asset manager Brookfield to supply Microsoft with a whopping 10.5 gigawatts of renewable power capacity between 2026 and 2030. That’s by far the biggest such deal ever signed, roughly equivalent to the amount of electricity required to power 4 million homes.
“It’s a magnitude of difference,” Xavier Smith, director of research, energy, and industrials at research firm AlphaSense, told Fortune. “Energy firms … used to be excited about a four-megawatt deal. Then they turned into 600-megawatt deals. And then now we’re in the gigawatts. It’s growing at this exponential rate.”
Brookfield is a giant in the renewable energy space. It manages over 7,000 power generating facilities across five continents, from wind farms to solar to hydropower. But even Brookfield will likely need to build up more capacity to meet Microsoft’s demands.
That could come in the form of more solar projects, which could be a good fit given potential Sunbelt solar sites’ proximity to growing data-center hubs in Phoenix and Las Vegas. And nuclear has emerged as another possible option for the tech sector: Amazon bought a nuclear-powered data-center site in Pennsylvania last month, and OpenAI CEO Sam Altman has backed a startup that seeks to power data centers with small nuclear reactors.
“Nuclear is increasingly becoming an important part of this conversation,” Smith said. “Back in the fall of last year, [Microsoft] had a job posting for a nuclear technology manager. They want someone to be able to develop a strategy for implementing small modular reactors for them. And so increasingly, nuclear is what people are thinking in terms of this space going forward.”
Neither company specified what types of renewable power Brookfield would provide Microsoft with, or the precise financial terms of the agreement. The Financial Times estimated the deal was worth around $10.5 billion, based on current industry trends and pricing structures.
Renewable energy prices have fallen significantly in recent years and could continue to do so, but Smith said it still makes sense for companies such as Microsoft to lock in contracts now rather than waiting for the possibility of cheaper power in the future or trying to build their own energy infrastructure. Because AI is scaling up so quickly and energy projects can take years of construction and permitting to go from proposal to completion, tech companies have incentives to move quickly.
“I actually think they’re smart to lock this in now,” Smith said. “All the hyper-scalers are pretty much saying that by 2030, they’ll be operating their data centers on 100% green energy … The fastest way to do that is to work with someone who’s already built out this infrastructure, rather than trying to start from zero and build out yourself because of the lead times in the space.”
Microsoft manages more than 300 data centers around the world, which it uses to store information and train the large language models that power chatbots such as ChatGPT and AI applications including Copilot. Because of the distributed nature of AI infrastructure, some have proposed that off-grid, bespoke energy projects directly powering data centers could be a more efficient way of providing electricity for AI. Although it might be cheaper for Microsoft or its competitors to build their own off-grid solutions, Smith said that that’s a less attractive option.
“While [tech companies] want exposure to renewables, they don’t want to be running a power center,” Smith said. “Google doesn’t want to run a power [plant]. Microsoft doesn’t want to run a nuclear facility.”
