As we stand on the brink of a new era in energy, a surprising shift is taking place in the tech world, one that is raising eyebrows: Hyperscalers are turning to nuclear power.
Yes, nuclear power. Use of this energy source, often associated with controversy and disasters, is being considered by Big Tech to meet its enormous AI-driven power needs while staying on track with net-zero goals. Major players like Google, AWS, and Microsoft are exploring nuclear as a way to support their ambitions in AI.
The rapidly rising energy needs of datacenters worldwide could surpass 1,000TWh by 2026 — a figure roughly equal to Japan’s total electricity use, according to the International Energy Agency (IEA). In Ireland, datacenters already strain the national grid, consuming around 21% of the country’s electricity.
As AI usage continues to expand, these energy demands are set to intensify, pushing hyperscalers to consider nuclear as a stable, high-capacity option.
In this way, the rise of AI is not just a technology trend — it’s a driving force in the energy transition, reshaping the power requirements of datacenters and challenging traditional energy sources.
But this AI-driven shift raises a critical question: Are hyperscalers truly prepared to handle the complexities and safety requirements of nuclear energy?
Balancing Sustainability with AI Power Needs
For hyperscalers, AI is creating an unprecedented demand for energy. Generative AI (GenAI), in particular, can use up to 33 times more energy than traditional software for a single task.
Given this surge in energy demand, hyperscalers face a major dilemma: how to secure a reliable power supply that aligns with their sustainability commitments.
Expanding grid connections to meet this demand is not a viable solution in many instances. In the U.S., for instance, about 1.5TW of generation capacity, mainly from low-carbon power sources such as solar and wind, is waiting for grid access. This backlog underscores the growing strain on the grid and the challenge of meeting rising energy demands in a sustainable manner.
In response to these challenges, hyperscalers are looking at restarting existing reactors already connected to the grid, as well as at the potential of off-grid small modular reactors (SMRs), which are faster to build and, according to proponents, safer.
However, a key question persists: Will nuclear power truly meet hyperscaler needs in a sustainable way — or will it cause more problems than it solves?
Why Nuclear?
Nuclear power offers reliable, low-carbon energy 24 x 7. A steady power supply is vital for datacenters, which need to operate continuously. Unlike solar or wind power, which depend on weather conditions, nuclear energy can provide power without interruptions.
For hyperscalers, reliability is crucial. A power failure at a datacenter could lead to major financial losses and service disruptions — making nuclear power’s dependability especially attractive.
Examples of hyperscaler investments in nuclear energy include:
- Google has partnered with Kairos Power to install SMRs, with a target of 500MW of capacity by 2035.
- AWS is working with Dominion Energy and X-energy on SMR projects that could provide up to 5GW by 2039.
Is nuclear energy as clean and safe as it needs to be? Opinions on nuclear safety are still divided. Our World in Data says nuclear is among the safest energy sources, with just 0.03 deaths per terawatt-hour, much lower than coal or oil. It’s also one of the cleanest, producing only six tons of CO2 per gigawatt-hour.
However, many members of the public continue to have serious concerns about nuclear safety, especially in countries like Germany and Japan, where memories of nuclear incidents remain fresh.
Risks and Complexities
Nuclear energy projects often face delays and budget overruns. SMRs promise lower up-front costs, but their economic viability is still unproven in practice. NuScale, the first U.S. company to gain SMR design approval, recently cancelled its first commercial project due to unexpected costs. With just two SMR designs in commercial operation so far, their ability to meet both cost and performance expectations remains largely untested.
There are also safety and security challenges. Relying on imported uranium (20–30% of which comes from Russia) may be risky geopolitically. Additionally, nuclear sites can be vulnerable to cyberattacks. A recent court case against the Sellafield nuclear waste site in the U.K., for instance, exposed cybersecurity weaknesses that could have had serious consequences.
For those uneasy about nuclear energy’s history and the associated security and safety concerns, the risks may be difficult to ignore. And with a history of budget overruns and regulatory obstacles, can nuclear realistically meet the short timelines hyperscalers need for their AI-driven power demand?
Hyperscalers as Energy Companies
The bottom line: As hyperscalers move toward nuclear power, they start to look more like energy suppliers than traditional tech firms. Building off-grid nuclear plants, investing in energy infrastructure, and complying with new regulatory requirements are pushing them into unfamiliar territory. But they may have little choice.
The Way Forward
Where does it all lead? Hyperscalers that want to move into nuclear energy face a tough decision. On the one hand, nuclear power might provide the energy they need to support AI’s growth without compromising low-carbon goals. On the other, such a step brings significant risks and challenges that go far beyond their core business.
The hyperscaler shift to nuclear could mark a new chapter in which Big Tech becomes deeply involved in energy transition policy and infrastructure. Whether this will lead to a more sustainable future is uncertain — but the decision could set a precedent that others will follow … or at least learn from.
Learn More
Curious about the energy transition? Discover IDC’s new Worldwide Energy Transition Strategies program, which builds on our utilities research to explore how this evolution impacts various industries.