Why is the AI industry shifting toward nuclear energy?
The rapid expansion of artificial intelligence has transformed electricity from a secondary operational concern into a primary strategic bottleneck. While previous industry strategies focused on improving performance per watt through more efficient chips and smarter cooling, these gains have failed to curb total energy consumption. Instead, increased efficiency has historically led to higher overall demand, a phenomenon known as the Jevons Paradox.
To secure the necessary power, cloud computing giants are now bypassing the search for efficiency in favor of direct energy procurement. In a single week, two of the world's largest cloud providers entered into 20-year nuclear energy agreements. Amazon has secured 690 megawatts from Constellation’s Calvert Cliffs plant in Maryland, with plans for an additional 190 megawatts between 2030 and 2032. Google has similarly signed an agreement intended to bring 890 megawatts of additional nuclear capacity into the United States.
Notably, these agreements do not focus on constructing new reactors, which can take over a decade to complete. Instead, they target the revitalization of existing infrastructure. Constellation is investing more than $4.3 billion to upgrade 11 reactors across Illinois, Pennsylvania, and New Jersey to maximize existing output.
How are existing nuclear plants being repurposed for AI?
The resurgence of nuclear power marks a significant reversal in energy policy, as the industry moves away from the post-Fukushima era of decommissioning toward a new era of reactivation. Rather than waiting for the long lead times of new builds, companies are betting on the restart of shuttered facilities.
One prominent example is the Three Mile Island reactor, which is being brought back into service under a 20-year contract with Microsoft. Furthermore, the US Department of Energy is providing up to $4.4 billion in loans to support the restart of three shuttered plants, aiming to restore roughly 2.25 gigawatts of capacity. The Palisades plant in Michigan, closed in 2022, is expected to be operational again by March 2027, potentially becoming the first US commercial nuclear plant to restart after entering the decommissioning phase.
The limitations of Small Modular Reactors (SMRs)
While the industry is looking toward Small Modular Reactors (SMRs) as a long-term solution for decentralized data center power, they are not a current fix. Most SMR projects are not expected to deliver electricity until the early 2030s, and very few have been successfully constructed globally. This creates a temporal mismatch: the high-speed capital expenditure cycle of AI hardware moves much faster than the decade-long timeline required for nuclear energy deployment.
What is the impact of data center demand on consumer electricity bills?
The costs of securing energy for AI are increasingly being socialized through electricity grid capacity charges. Much of this activity occurs within the PJM interconnection, the grid operator serving 67 million people across 13 US states and Washington, DC. PJM utilizes capacity auctions to ensure enough generation is available for future demand, and these costs are passed on to end-users.
Data center demand has become a massive driver of these costs. According to PJM’s independent market monitor, data center demand contributed approximately $29.4 billion in capacity costs across the four most recent auctions. In the most recent auction alone, roughly $6.3 billion of these costs were attributed to data center demand, including projections for facilities that are still in the planning stages.
This means households are effectively subsidizing the infrastructure required for AI. The Natural Resources Defense Council estimates that by 2028, a typical household within the PJM territory could see an increase of $70 per month in electricity costs if current trends persist. This financial pressure has triggered political reactions, including new electricity rate classes for data centers in Virginia and minimum billing requirements in Ohio.
How are regulators responding to the energy demand of Big Tech?
Regulators are caught between the need to facilitate technological growth and the necessity of protecting consumer interests. The current landscape is defined by a tension between mandatory requirements and voluntary industry commitments.
The Google agreement represents a model where large consumers aim to bring new capacity to the grid rather than simply competing for existing supply. However, such arrangements are currently largely voluntary. In October 2026, US federal regulators rejected most of a PJM proposal that would have required large new electricity consumers to provide their own generation capacity. This leaves the industry in a precarious position: if voluntary commitments fail to keep pace with demand, the regulatory environment may shift toward much more stringent, mandatory requirements that could disrupt the current AI investment cycle.
What can emerging AI hubs like the Gulf learn from this?
For regions such as the Gulf, which are aggressively expanding AI infrastructure, the US experience serves as a critical case study in energy-driven competition. Countries in this region are leveraging abundant energy resources to build a competitive advantage in the global computing market.
Current expansion plans in the Gulf are massive in scale:
- Qatar: A $20 billion joint venture with Brookfield targeting an initial 300 megawatts, with potential expansion to 1.5 gigawatts.
- Abu Dhabi: Pursuing plans for up to 5 gigawatts of AI data center capacity.
- Saudi Arabia: Targeting nearly 2 gigawatts through the HUMAIN initiative.
As AI computing becomes increasingly dependent on reliable, high-density electricity, the ability to provide power will be the primary determinant of where the next generation of digital infrastructure is located. The challenge for these nations will be managing the scale of these developments to ensure they remain economically and environmentally sustainable.
Frequently asked questions
Why is AI causing an increase in electricity demand?
AI requires massive computational power for training large models and running continuous background agents. Even as individual chips become more efficient, the scale of AI tasks—such as video generation and complex reasoning—is growing so rapidly that it outpaces any efficiency gains, leading to higher total energy consumption.
How does data center demand affect my monthly power bill?
Data centers increase the total demand on the electrical grid, requiring utilities to secure more "capacity" to prevent blackouts. These capacity costs are often passed to all customers through electricity bills. In some US regions, this could add significant monthly charges to typical household accounts.
Can nuclear power solve the AI energy problem?
Nuclear power provides the high-capacity, steady energy that data centers require. While restarting old plants and upgrading existing reactors offers a faster solution than building new ones, the timeline for significant nuclear expansion still lags behind the immediate energy needs of the AI industry.
What are Small Modular Reactors (SMRs)?
SMRs are a newer type of nuclear reactor designed to be smaller and easier to deploy than traditional large-scale plants. While they are a major focus for tech companies, they are not yet widely available and are not expected to contribute significantly to the grid until the early 2030s.
Is there any regulation preventing Big Tech from using all the power?
Currently, much of the responsibility for providing new energy capacity is voluntary. While some US states are implementing new billing rules for data centers, federal regulators have recently been hesitant to mandate that large energy consumers provide their own generation capacity.
Key takeaways
- The International Energy Agency predicts global data center electricity demand will exceed 945 terawatt-hours by 2030.
- Big Tech companies like Amazon and Google are securing long-term nuclear energy via existing reactor upgrades.
- Data center demand contributed $29.4 billion in capacity costs across four recent PJM auctions.
- Households in the PJM region could see electricity costs rise by $70 per month by 2028.
- The Gulf region is targeting massive AI capacity, including up to 5 gigawatts in Abu Dhabi.
The future of AI infrastructure depends on energy stability
The transition of the AI industry from a software-centric focus to an energy-centric one marks a fundamental shift in the global economy. As data centers become massive consumers of electricity, the intersection of technology, energy policy, and consumer rights will become increasingly volatile. The ability of the grid to absorb this demand without placing an undue burden on households will determine the social and political viability of the AI revolution. Whether through nuclear revitalization or new regional hubs, the winners of the AI race will be those who successfully navigate the complexities of the electricity bottleneck.
