Big Tech's Turn to Nuclear Power: Fueling the AI Boom with Small Modular Reactors

As artificial intelligence rapidly expands, so does its energy demand, pushing tech giants like Microsoft, Google, and Amazon toward nuclear power. These companies have secured multi-billion dollar nuclear deals to sustain data centers essential for AI, which require significantly more energy than traditional tech operations. However, this pivot to nuclear has sparked debate, with some urging a more measured approach.

Microsoft recently announced plans to purchase power from Pennsylvania's Three Mile Island plant, closed since 2019 after a historic nuclear accident in 1979. Google followed with a deal with Kairos Power for 500 megawatts from small modular reactors (SMRs), which is expected to go online by 2030. Amazon announced a similar partnership with X-Energy. SMRs are lauded for being safer due to their compact design, and they're also faster and cheaper to build, providing a steady renewable power source that can help meet the companies' 2030 emissions-free goals.

Generative AI's power needs are significant, with a single ChatGPT query consuming as much energy as a lightbulb left on for 20 minutes. AI's rapid demand has led to some missed net-zero targets, surprising companies that once set achievable sustainability goals. AI researcher Sasha Luccioni argues that companies should disclose AI's energy usage to help governments develop regulations. She also warns that tech's "move fast and break things" approach is risky when applied to nuclear energy, a sector that requires caution.

Microsoft, Google and Amazon turn to nuclear energy to fuel the AI boom | CBC Radio

Proponents highlight nuclear's potential to provide consistent, carbon-free energy. "The use of nuclear reactors is almost like an ideal energy source," explains MIT professor Jacopo Buongiorno, noting that SMRs are reliable for powering data centers. While wind and solar energy are renewable, they need battery storage for consistency, adding costs. Nuclear offers uninterrupted energy without carbon emissions.

Under the Biden administration, the U.S. government has also embraced nuclear energy. Nuclear power currently supplies 20% of the nation's electricity and is considered essential for climate goals. This represents a shift, as safety and economic concerns had previously deterred many Democrats. "Revitalizing America's nuclear sector is key," said Energy Secretary Jennifer M. Granholm, referencing AI, data centers, and manufacturing needs.

Tech companies have long invested in wind and solar energy, but nuclear's appeal is its 24/7 availability without requiring battery storage. This move toward nuclear reflects the heightened urgency to meet large-scale energy demands. Microsoft, Google, and Amazon pledged to use emissions-free power by 2030, though these commitments came before AI's rise. Microsoft co-founder Bill Gates and investor Warren Buffett have also backed TerraPower, a startup developing modular reactors for scalable, cost-effective energy.

Despite these advancements, nuclear's challenges persist. Critics argue that high costs, delays, and waste storage issues risk new reactors. Arnie Gundersen, chief engineer at Fairewinds Energy Education, noted that since 1960, over half of U.S. nuclear projects were canceled, with none completed on time or within budget. Although the industry remains unpredictable, some, like Rich Powell from the Clean Energy Buyers Association, believe mass production could lower costs over time.

Data centers, particularly for AI, are driving skyrocketing energy demand. While centers currently represent a small percentage of global energy use, they're expanding quickly, with regions like Northern Virginia already straining local grids. AI chips consume significantly more power than typical servers, making a consistent power supply critical. In the last quarter alone, Alphabet, Microsoft, and Amazon collectively spent $59 billion on infrastructure, a 63% year-over-year increase driven by AI demands.

SMRs offer scalability, cost-effectiveness, and advanced safety features. In North America, Canada's Ontario Power Generation is constructing the first SMR at its Darlington plant, expected to go live by 2029. Buongiorno believes this timeline is realistic if investment, demand, and regulatory support align.

The tech industry's nuclear power investments may play a key role in meeting AI's energy needs and supporting environmental goals. Yet, whether nuclear can overcome its longstanding challenges will shape the future of reliable, clean energy for companies like Microsoft, Google, and Amazon as they look to fulfill AI's energy needs.