The answer to AI’s energy needs could be blowing in the wind

Data centers and offshore wind farms could prove to be a perfect pair as AI grows increasingly energy-hungry.

Many of the companies racing to roll out new generative AI tools have also made big commitments to rein in their greenhouse gas emissions. A rush of new AI data centers could throw those climate goals off — unless they find a clean source of energy.

Conveniently, some of the quirks with AI data centers make them a good match for wind farms at sea. And the offshore wind industry, which has already cozied up with Big Tech, could certainly use a boost as it looks ahead to more uncertain economic waters.

“Data centers and AI … We are excited about that opportunity,” Mads Nipper, CEO of leading offshore wind developer Ørsted, said in an earnings call last week. “We are excited about this both because it’s very high growth, in demand, but also because it [comes from] companies that generally are very committed to the transformation of their [carbon] footprint.”

Giants like Microsoft and Google are throwing everything they’ve got into developing the best AI tools and inserting them into every conceivable platform, from gaming to social media, work apps, and search. That all comes with steep energy costs.

By 2026, AI could gobble up 10 times as much electricity as it did in 2023, according to the International Energy Agency. Data centers used to train AI are even more energy-intensive than traditional data centers used to shepherd the world’s emails and cat videos. And AI data centers are taking up more real estate than ever. Data centers under construction in North America combined would need as much energy as all the households in the San Francisco metro area annually, according to a February report from real estate services company JLL.

“If we could do these data centers close to [offshore wind farms], this is something that can offer high loads of attractive power. So [it’s] something we will certainly dive into both in Europe and US to look at that opportunity,” Nipper said on the earnings call.

Flexibility is what might make data centers for AI particularly attractive for offshore wind developers. Unlike other data centers, those used for training AI models don’t have to worry as much about being near population centers to reduce latency. They can also theoretically ramp up and wind down operations more easily to match fluctuations with the weather, which can be a challenge with relying on wind energy. That means AI data centers could tap into surplus energy that might otherwise go to waste.

“That is actually very unique because if you look at the bigger picture, it’s not that many sectors that are flexible [with their energy use],” says Odin Foldvik Eikeland, an analyst at research firm Rystad Energy. “This can also open up some very cool collaborations … I don’t see any reason why not.”

For tech companies, offshore wind farms aren’t only attractive as an abundant source of carbon pollution-free energy. There are other potential benefits. Microsoft launched research project exploring the feasibility of subsea data centers powered by offshore wind farms called Project Natick. On land, data centers have provoked concerns about how much electricity and water they use for cooling systems. Storing servers on the seafloor could be a more sustainable way to keep them cool. Keeping servers underwater might also prevent corrosion from oxygen and humidity. After two years, Microsoft’s servers in Project Natick showed just one-eighth the failure rate of servers on land.

Tech companies have made a wave of climate pledges in recent years, becoming the biggest corporate buyers of renewable energy. Microsoft, which has invested more than $13 billion in OpenAI, last week made the largest corporate agreement to purchase renewable energy to date. Google, in its “Gemini era,” announced its biggest offshore wind deal yet in February to power data centers in Europe. Both companies have pledged to match 100 percent of their electricity use with carbon-free energy purchases around the clock by 2030.

That’s been a boost to offshore wind developers like Ørsted at a time when they face harsh economic headwinds. The offshore wind industry is still trying to take off, accounting for less than 1 percent of the world’s electricity mix. And supply chain disruptions and soaring inflation rates have bogged the industry down since the covid-19 pandemic, leading Ørsted to cancel two big projects in the US late last year.

Even so, utilities are running out of time to figure out how to meet skyrocketing electricity demand. Electrifying homes, buildings, and vehicles puts more pressure on power grids. The boom in new data centers used for crypto mining and AI is another huge challenge. And it’s all happening at once.

“Data center computing is super power hungry to begin with, and then AI is just a level of magnitude higher than that,” says Stephen Maldonado, research analyst for North America wind at Wood Mackenzie. “I think it’s a real possibility that we could see that sort of demand drive the utilities to pursue offshore wind as a way of achieving those decarbonization goals.” Offshore wind farms are especially attractive in densely populated coastal regions, he says, where there isn’t much space on land to erect skyscraper-sized turbines.

Whatever happens next is going to have to be a balancing act. The US, for example, hasn’t yet installed nearly enough renewable energy to meet its climate goals. And the surge of new AI data centers risks siphoning clean energy away from arguably more important things like keeping the lights on in peoples’ homes. But if data centers don’t run on clean energy, fossil fuels are the alternative. Gas pipeline operator TC Energy Corp said in an earnings call last week that it expects new data centers to lead to more demand for gas-fired electricity.

“It’s important for these data centers to ensure that the power they buy is coming from clean renewables such as solar and wind,” Eikeland says. “If we’re starting up fossil fuels again to accommodate them, we would be going in the wrong direction.”