Amazon’s chief executive has said that power is the “single biggest constraint” in its cloud and artificial-intelligence business, and the scale of the company’s existing infrastructure reflects that. Aterio, a data-center intelligence firm, estimates that Amazon’s operating self-built data centers in the U.S. consume up to around 9 gigawatts — roughly the generation capacity of North Dakota. By comparison, Microsoft and Google each have self-built data centers using up to about 5 gigawatts, and Meta’s capacity is roughly 4 gigawatts.
However, the future capacity landscape is shifting. Aterio projects that Google will add data-center and power capacity in the U.S. at the fastest rate through 2030. When leased capacity from third-party data-center owners is included, Google will have significantly reduced Amazon’s lead by 2030, according to Aterio.
Sergio Toro, founder of Aterio, said Amazon has a “really good understanding of building capacity” and a methodical approach. Amazon plans to build most of its own capacity, while about a quarter of Google’s expected 2030 data-center capacity will come from leases. Self-built capacity is cheaper in the long term but slower.
The two companies also diverge on energy sources. Google stands out for “doing everything it can” to avoid building data centers reliant on fossil fuels, said Michael Thomas, founder of data firm Cleanview. Google is the only hyperscaler with an in-house renewable-energy-focused developer, having acquired Intersect Power earlier this year. The company was also the first major tech firm to study how its data centers could reduce power consumption during peak grid demand.
Google’s clean-energy speed is already observable. At least three of its planned Texas data centers will skip the long queue to connect to the grid by co-locating with solar and wind projects, according to a Cleanview report. In two cases, Google will build the solar and wind capacity through its Intersect Power subsidiary. Texas market rules permit faster grid connection for data centers co-located with new power sources. In all three cases, the renewable capacity well exceeds the data center’s needs.
But even a green-focused company must accept near-term trade-offs. Google is renting computing capacity from SpaceX, which built out its data centers in record time using off-grid natural-gas-fired power equipment; SpaceX subsidiaries were sued earlier this year over air pollution concerns.
Other hyperscalers are pursuing similar off-grid gas plans. Microsoft struck a 20-year agreement with Chevron to power an AI data center in Texas with an off-grid natural-gas plant. Data compiled by Cleanview indicates both Meta and Amazon have plans for such projects. Amazon has not publicly confirmed its involvement, but its data center in Fayette County, Ohio, is the only planned large power demand near a permitted off-grid gas project, according to Thomas.
All the hyperscalers have also placed bets on novel energy technologies, including small modular nuclear reactors, advanced geothermal, novel batteries, and even solar energy beamed from space. SpaceX and Google are exploring data centers in space.
The ultimate winner of the AI race will depend partly on political winds, which today are permissive of natural-gas power, and on which new energy technologies achieve breakthroughs first. For now, Amazon and Google have a clear lead.