Amazon Web Services (AWS) is investing millions of dollars into nuclear power projects to meet data center energy demands that have skyrocketed due to the growing use of artificial intelligence (AI) while at the same time maintaining a goal of net-zero carbon emissions across its operations by 2024, a move analysts’ note is a growing option for hyperscalers.

The AWS move will see the hyperscaler “support” the development of three nuclear energy projects, including “enabling the construction of several new small modular reactors (SMRs).” These SMRs are described as nuclear reactors with a smaller footprint that allows them to be constructed quicker and closer to the energy grid.

The agreements are with Energy Northwest in Washington for a facility that can provide up to 960 megawatts (MW) of power; an investment in X-Energy that is developing the SMR for the Energy Northwest deployment; and with Virginia’s Dominion Energy to work on development of an SMR project near an existing power generation facility. AWS had also previously signed a deal to build a data center next to Talon Energy’s nuclear facility in Pennsylvania.

Ed Anderson, a distinguished VP analyst at Gartner, explained in an interview with SDxCentral that AWS was running into a pair of problems: how to power these data centers and how do they stay on track with their sustainability goals.

“The answer to the problem has to be ‘we're going to invest in new sources of energy, but it has to be clean energy,’” Anderson said. “That could have been, ‘we're going to build solar farms. We're going to build wind farms, or hydroelectric, or geothermal, or something like that, or, in this case, nuclear. The announcement is interesting because it indicates that Amazon is adding that nuclear option.”

The AWS move came just days after Google Cloud signed a deal to purchase nuclear-generated power from multiple SMRs being produced by Kairos Power. This includes the eventual acquisition of up to 500 MW of nuclear-based energy.

“I think it's also notable that a large company like Amazon is putting their stamp on this direction, saying this is the way we're going to help solve this problem,” Anderson said. “I do think it's going to bring more visibility, more credibility to this approach. The thing to watch is what comes next now that Amazon and some of Amazon's peers are taking a stance like this.”

What are the power requirements? MTN Consulting in a report found that webscale energy consumption has doubled since 2019, growing 15% per year over the past two years.

“More important, since Covid, the webscale sector has become more energy intensive, not less,” MTN Consulting Chief Analyst Matt Walker wrote. “In 2021, 59 MWh [megawatt-hours] of energy was consumed per [$1 million] in revenues. That figure grew to 65 in 2022, and 70 in 2023. This is notable as webscalers are supposed to be able to exploit their size in order to get efficiencies from scale; it’s going the opposite direction with data center power consumption.”

Microsoft in a recent blog post discussed disaggregated power systems as a way to help tackle this growing data center power demand. The cloud giant noted that compute and storage systems for cloud typically have power density below 20 kilowatts (kW), but that “AI systems has driven power densities to hundreds of kW.”

To counter this, Microsoft said it was working with Meta on its Mt. Diablo disaggregated rack design that uses a 400 high-voltage direct current (VDC) unit that can scale from hundreds of kW up to 1 MW “enabling 15% to 35% more AI accelerators in each server rack.”

“This modular approach allows for power adjustments in the disaggregated power rack to meet the changing demands of different inferencing and training SKUs,” Microsoft’s team explained.

AI driving alternative energy demands One driver for this surging energy demand is the growing use of AI and generative artificial intelligence (genAI), which requires more powerful and plentiful data centers.

Dell’Oro Group found that worldwide data center capex surged 38% during the first half of this year compared to 2023, which it tagged to spending on accelerated servers that are powering genAI use cases. This included servers using “Nvidia Hopper GPUs and custom accelerators, such as Google’s TPU and Amazon’s Trainium” that “gained traction among hyperscale cloud service providers.”

This insight affirmed previous reports that found hyperscalers were driving significant new investment into the data center space.

Synergy Research Group (SRG) noted that hyperscalers now control more than 1,000 total large data centers around the globe, which accounts for 41% of the worldwide capacity of all data centers. Just over half of that hyperscaler capacity is from own-build, owned data centers with the remaining portion from leased facilities.

SRG chief analyst John Dinsdale also noted that enterprises are also planting more of their gear in colocation facilities, which is reducing their overall on-premises data center capacity and is a practice expected to continue to increase as enterprises tap more into AI services.

“The rise of generative AI technology and services will only exacerbate those trends over the next few years, as hyperscale operators are better positioned to run AI operations than most enterprises,” Dinsdale wrote.

Research firm ISG earlier this year also noted that the average large enterprise is planning to nearly double their number of AI-enabled applications by the end of this year. This will see that average grow from 250 applications that were AI-enabled at the end of 2023, to 488 AI-enabled applications by the end of 2024.

Dell’Oro Group pegs full-year data center capex as growing 35% compared to 2023, hitting $400 billion in total segment revenues. This will be led by those large-scale cloud providers like AWS, Microsoft, and Google Cloud.

“We are – and we’ve talked about now for quite a few quarters – we are constrained on AI capacity,” Microsoft CFO Amy Hood said during a Microsoft’s second-quarter earnings call. “And because of that … we’ve … signed up with third parties to help us as we are behind with some leases on AI capacity. We’ve done that with partners who are happy to help us extend the Azure platform, to be able to serve this Azure AI demand. And you do see us investing quite a bit as we’ve talked about in builds so that we can get back in a more balanced place.”

Gartner’s Anderson added that this expansion will drive unique energy options from the hyperscalers.

“They're all a little bit different, but they're all kind of in the same vein of, ‘we need new energy and we want to adhere to our sustainability goals,’” Anderson said. “I don't think we’re going to see hundreds of companies making these announcements as it takes the mega-providers, those who are operating these massive data centers, they're the ones who are likely to get directly involved.”