Americas

  • United States
by Prasanth Aby Thomas

Google bets on nuclear power to drive AI expansion

News
Oct 15, 20244 mins
Data CenterPower Systems

Other tech giants, including Amazon and Microsoft, are pursuing similar strategies as AI-driven power demands are set to surge.

nuclear towers at three mile island
Credit: A. L. Spangler / Shutterstock

Google has signed its first corporate deal to purchase power from multiple small modular reactors (SMRs) to meet the energy needs of its AI systems, marking a key step as AI companies shift toward nuclear power.

In a blog post, Google announced an agreement with Kairos Power to source nuclear energy, aiming to bring the first SMR online by 2030, with more reactors planned by 2035.

“The grid needs new electricity sources to support AI technologies that are powering major scientific advances, improving services for businesses and customers, and driving national competitiveness and economic growth,” the company said in the post.

This move comes as other major tech companies are also exploring nuclear power as a sustainable energy source.

In March, Amazon acquired a nuclear-powered data center from Talen Energy. Last month, Microsoft and Constellation Energy agreed to revive a unit at Pennsylvania’s Three Mile Island plant, the site of the worst US nuclear accident.

Meanwhile, Oklo, a nuclear fission startup backed by OpenAI’s Sam Altman, is progressing toward its first SMR, aiming for completion by 2027. Altman has also invested in Helion Energy, a fusion power startup.

Benefits of using small modular reactors

Data center power usage is set to surge in the coming years, driven by cloud computing and AI. AI-related energy consumption alone is projected to increase by around 45% annually over the next three years.

“For the past 4-5 years, companies have actively attempted to reduce energy consumption or move to alternative sources,” said Suseel Menon, practice director at Everest Group. “Most recently Microsoft wound down an attempt to create underwater data centers. Given the global focus on sustainability, such companies have also explored renewable sources of energy for their operations. But these, although sustainable, can be disrupted due to natural phenomena.”

The high energy output of nuclear reactors is prompting hyperscale providers to turn to nuclear power to meet the growing demands of AI workloads, according to Rajiv Ranjan, associate research director at IDC.

“Nuclear power provides a consistent energy source, unlike renewables, which depend on the availability of sun or wind,” Ranjan said. “Although nuclear power sources are more expensive, their fuel efficiency is much higher than traditional sources.”

Google plans to purchase up to 500 megawatts of power from six to seven Kairos reactors, an initial commitment that analysts say may only cover a fraction of the company’s energy needs in the 2030s.

“The choice of Kairos Power is notable, as this startup in the SMR market has yet to launch its reactors,” said Hyoun Park, CEO and chief analyst at Amalgam Insights. “This could provide Google with an opportunity to co-develop solutions over the next six years, with reactor launches scheduled for 2030.”

It wouldn’t be surprising if Google explores ways to bundle these modular reactors with data center infrastructure, creating an integrated unit that could be replicated over time, Park added.

Challenges for AI companies


SMRs are being promoted as a clean energy solution, aligning with the net-zero carbon ambitions that most major tech companies have embraced.

However, the benefits are still largely theoretical, according to Sanchit Vir Gogia, chief analyst and CEO at Greyhound Research, who suggests that companies must consider nuclear waste disposal — an issue that will grow as SMRs scale up for data center use.

“There are also serious and practical considerations regarding where these SMRs must be located, along with safety and ongoing maintenance issues,” Gogia added. “If anything, the commissioning and use of these SMRs need to be tightly governed and periodically scrutinized, as there are still many unknowns.”

Managing and regulating nuclear power will also be a significant challenge, with some US states banning or restricting the construction of new nuclear plants. “Tech companies may use their considerable financial muscle to reduce permitting and governance challenges associated with nuclear energy,” Park said. “This could include supporting legislation that may allow tech companies specifically to build modular reactors for their own commercial use or with terms that would be difficult for standard utility companies to support in America’s privatized utility market.”