By Derek Ma, Jobanpreet Natt & Mustafa Al-Shamaa | Illustrated by Angila Atif | Fall 2025 Issue | Technology
Introduction
On paper, they look like opposites. One is the defining technology of the mid-20th century: heavy, industrial, and regulated to the inch. The other is the prodigy of the 21st century: intangible, lightning-fast, and evolving faster than laws can keep up. But both share a history of being hailed as breakthroughs and feared as threats. To power the digital brain of the future, Silicon Valley is turning to the atomic brawn of the past. We are witnessing the start of a massive symbiotic relationship: Nuclear energy provides the baseload power that AI desperately craves, while AI offers the optimization needed to bring the nuclear industry out of stagnation.
An Insatiable 24/7 Appetite for Power
Today's hyperscale AI data centers consume as much electricity as a small city, and their energy use is still rising rapidly. Goldman Sachs estimates U.S. data centers will consume roughly 8% of all U.S. electricity by 2030, up from about 3% in 2020. Some analyses are even more stark and predict that thanks to power-hungry AI workloads, data centers could consume 12% of U.S. power demand by 2028. This isn't just because there are more servers but rather because AI computation is getting more intensive. Training GPT-4 used over 50 gigawatt-hours of electricity, which is 50 times more than what was required to train its predecessor, GPT-3. According to the International Energy Agency, every ChatGPT query uses approximately 10x the energy of a normal Google search. In short, AI's growth is increasingly constrained by the physical limits of energy infrastructure.
Crucially, AI's power needs aren't like a typical factory or office. These supercomputing clusters run day and night, meaning any outage or dip in supply can halt or disrupt services for millions. The "always on" reliability is a must-have. Meanwhile, companies are pressured to uphold bold climate commitments despite the intense energy usage. Google, for instance, aims to be carbon-free by 2030. We're facing a new energy problem: AI keeps growing fast, but it needs massive amounts of clean, reliable power the grid can't easily supply.
Outgrowing the Grid? AI Turns to Nuclear Power
Artificial intelligence's rapid ascent is no longer just a computing challenge; it is becoming an energy crisis. That tension is already reshaping how tech giants source electricity. In central Pennsylvania, one of America's most infamous nuclear reactors is getting a second chance thanks to this challenge. Microsoft recently agreed to buy 100% of the output from the previously shut-down Three Mile Island Unit 1. It's signing a 20-year deal, paving the way for the reactor's restart by 2028. The US $1.6 billion dollar bet on nuclear power is resurrecting a shuttered plant to keep Microsoft's cloud and AI servers running clean and constant. The deal is the first-ever restart of a closed U.S. reactor, underscoring a brewing energy crunch. Despite access to renewables and traditional grid power, Microsoft turned to nuclear to guarantee long-term, round-the-clock supply. As tech giants deploy machine learning at scale, they face AI's insatiable energy appetite — demands for reliable and round-the-clock power at massive scale. Few energy sources can meet all those needs, and nuclear is a leading candidate to close that gap.
Why Nuclear Checks the Boxes
Until recently, the default solution for data center operators was to buy more renewable energy. But intermittent renewables alone aren't keeping up with AI's 24/7 hunger. The fundamental issue is reliability. Nuclear plants by contrast operate at 90% capacity factors over the course of a year, meaning they run near full output almost all the time. That is far higher than solar or wind and lets them deliver steady electricity around the clock. A single reactor can churn out hundreds of megawatts day and night, which is exactly the uninterrupted supply that sensitive AI loads require. While renewables perform well during peak generation hours, maintaining round-the-clock, multi-day power for hyperscale data centers remains a major challenge.
Equally, nuclear energy has a clear advantage because a gigawatt-scale plant can supply a huge amount of power from one location and can be built close to major centers. This siting flexibility — meaning the ability to place reactors near where electricity is needed instead of in remote areas with strong sun or wind — is widely seen as a major benefit for data centers. This proximity helps reduce transmission losses and delays. By shortening the distance between power generation and consumption, nuclear power can offer greater reliability without waiting on major grid upgrades.
Perhaps the biggest selling point is carbon-free firm power. Tech firms cannot revert to coal or natural gas plants to keep their AI servers running without breaking their climate pledges. While companies can offset emissions through carbon credits, they cannot offset outages, interconnection delays, or supply constraints. As Amazon's AWS CEO, Matt Garman, put it, "Nuclear is a safe, carbon-free source that can scale to meet our growing demands." In short, nuclear uniquely hits the trifecta for AI: it is reliable, it is scalable, and it aligns with long-term sustainability goals.
The Infrastructure Wall Behind AI's Growth
Even if wind and solar could meet the same requirements, there is another practical constraint which is time. Building enough nuclear capacity is a challenge that can take more than a decade, and AI is booming now. Across the United States, new energy projects face long interconnection queues and delays. This lag has become a major obstacle for near-term clean power expansion. AI-focused data centers, however, are coming online within the next few years and not in the distant future. The mismatch between AI's rapid growth and the slow pace of grid upgrades has companies searching for alternatives.
To bridge this timing gap, companies are beginning to explore co-locating power generation directly beside data centers as a way to bypass the strained grid. A compact small modular reactor can be built on or near a data center campus, supplying power directly and avoiding many of the delays and complications found in the larger grid system.
A Critical Alliance Facing Major Challenges
Critics argue that pairing AI with nuclear is far from straightforward. Nuclear plants, especially new designs, face long construction timelines and complex licensing. Even with demand rising, moving a next-generation reactor from concept to the grid can take most of a decade in the current regulatory environment. That is why tech firms are hedging by supporting near-term options, such as restarting dormant reactors or keeping aging plants online, while also funding advanced designs that will not be ready until the 2030s. Yet despite these concerns, the flood of 20-year contracts and reactor investments shows that companies are betting the challenges can be overcome. AI needs the stability and strength that nuclear provides, and the nuclear sector is gaining a new lease on life by stepping in to support AI's rising energy demand.
AI: The Only Way to Break Nuclear's Curse
The nuclear industry has a habit of making hollow promises. The "atomic age" hype of the 1960s dissolved into a reality of regulatory gridlock, public fear, and catastrophic project management failures. Critics are right to be skeptical: if nuclear energy is so perfect, why does it remain so difficult to build? However, the variable that has fundamentally changed is not the reactor technology itself, but the emergence of Artificial Intelligence. AI has arrived not merely as a hungry consumer of power, but as the only force capable of solving the safety and efficiency bottlenecks that have strangled the industry for decades.
The Argument for Speed: Solving the "7-Year" Problem
The central argument against nuclear expansion has always been time. As Dr. Anthony Ciccone, VP of Global Nuclear Development at WSP, bluntly states, nuclear power plants still take 7 years to build — "Hyperscalers ask for 2." In an era where tech CEOs like Mark Zuckerberg are willing to "misspend hundreds of billions" to avoid being late to superintelligence, nuclear's traditional timelines are non-starters.
AI is the only mechanism capable of bridging this gap. The industry is infamous for schedule spillage where a single delay — pouring concrete or certifying a weld — ripples through thousands of subsequent tasks. Human project managers have historically failed to contain this chaos. In response, Westinghouse has deployed AI tools like HiVE™ and Bertha™, which analyze 75 years of historical project data to predict "choke points" before they manifest. By simulating thousands of schedule permutations to find the optimal path, these tools are doing what humans cannot: shaving months off construction time and preventing the billion-dollar overruns that kill projects.
Fixing a "Convoluted" Supply Chain
Dr. Ciccone further argues that the nuclear supply chain is a "complex convoluted mess that nuclear hasn't figured out." The delivery of millions of specialized parts — like valves, pipes, pumps — is prone to human error that leaves workers idle and budgets ruined. The argument for AI here is straightforward: copy the winners. Logistics giants like FedEx, UPS, Amazon, and Walmart already utilize AI logistics managers to track real-time data and predict shortages. By adopting these proven models, the nuclear sector can automatically adjust procurement to ensure zero downtime. Furthermore, AI solves physical coordination failures that humans miss. Manual reviews of 3D CAD models are slow and error-prone; AI-driven software like Autodesk's Revit can now scan digital twins to identify "clashes," such as a pipe running through a support beam, resolving them virtually before ground is ever broken.
Defeating the Paperwork Weight
The most damning critique of nuclear energy is often its bureaucracy. The adage that "a nuclear reactor is not finished until the paperwork weighs as much as the containment dome" is barely an exaggeration. The regulatory burden is the single largest non-hardware cost, with licensing reviews consuming 10–12 years of human effort.
This is where the argument for AI becomes undeniable. The Pacific Northwest National Laboratory (PNNL) is developing PermitAI to dismantle this barrier. By ingesting 120,000 documents (5 billion tokens of text), the system reduced data retrieval times from months to minutes — a 99% reduction. A PNNL pilot proved this efficacy by slashing environmental review times for small reactors from 3–6 years down to 6–24 months. AI isn't just "helping" with paperwork; it is the only viable path to clearing the regulatory hurdles that have historically frozen the industry.
Optimizing the "Million-Dollar Day"
Finally, the economics of nuclear power hinge on uptime. Every 18–24 months, a reactor shuts down for refueling, costing $1 million per day in lost revenue. This outage period is the most expensive phase of a reactor's life. Ontario Power Generation (OPG) proved that human planning is insufficient for these high-stakes windows. Their "Outage AI" tool analyzed eight years of data to predict logic ties for over 25,000 tasks, identifying dependencies humans overlooked. The result is plants getting back online hours or days sooner, saving tens of millions of dollars.
Furthermore, the hazardous nature of these facilities makes human maintenance expensive and dangerous. Deep learning models, like those researched at the Federal University of Santa Catarina, have achieved greater than 99% diagnostic accuracy in detecting failures in motors and pumps. Combined with autonomous robots using computer vision, AI can monitor radioactive sections prone to flow-accelerated corrosion that are otherwise costly for humans to access. The verdict is clear: Nuclear does not need more "hype." It needs the logistical, regulatory, and operational discipline that only Artificial Intelligence can provide.
Conclusion
The narrative of nuclear energy has long been one of stagnation: a technology frozen in amber, trapped by its own complexity and public apprehension. Conversely, the narrative of Artificial Intelligence has been one of unbridled velocity, a force accelerating faster than the infrastructure can support. The convergence of these two industries is not merely a transaction of power for optimization; it is the closing of a critical loop.
AI needs the stability of nuclear to grow without collapsing the grid or the climate. Nuclear needs the intelligence of the algorithm to shed the inefficiencies that have made it economically unviable. This is no longer a theoretical partnership. From Microsoft's resurrection of Three Mile Island to Westinghouse's deployment of predictive neural networks, the pact has already been signed in silicon and steel.
We are witnessing the end of the "atomic age" as a relic of the past and the beginning of the "atomic-digital age." In this new era, the electrons that train the next generation of superintelligence will be split from the atom, and the intelligence they create will, in turn, design the next generation of reactors. The future of clean energy is not just about building more power plants; it is about building smarter ones. And for the first time in history, we have the tools to do both.
