Electricity has replaced computing chips as the limiting factor in the AI arms race. While NVIDIA's dominance in AI processors created a $4 trillion company, the real bottleneck emerging now is far more fundamental: the power grid simply cannot keep up with the energy demands of artificial intelligence systems. A single ChatGPT query consumes roughly 10 times the energy of a Google search, and training next-generation large language models requires power equivalent to small cities. This infrastructure crisis is driving tech giants to pursue nuclear energy at an unprecedented scale.

The numbers tell the story. Goldman Sachs Research projects global data center power demand will surge up to 165% by 2030 compared to 2023 levels, while industry forecasts put AI data center capital expenditure at roughly $5.2 trillion between now and 2030. Yet the electrical grid was built for a world where demand grew predictably at 1 to 2% per year. Hyperscalers are now showing up at utility offices requesting hundreds of megawatts on three-year timelines, and the answer keeps coming back the same: the grid cannot deliver it. Berkeley Lab found that more than 70% of grid interconnection requests in the United States are ultimately withdrawn because the grid simply cannot accommodate them.

Why Is Nuclear Power Suddenly Central to AI's Future?

Unlike the chip shortage, which was an 18 to 24 month manufacturing problem, the power shortage is a 10-year infrastructure challenge with no shortcut. New nuclear plants take 10 to 15 years from approval to operation, while new transmission lines require 8 to 12 years to permit and build. These timelines do not compress, regardless of capital investment. Recognizing this reality, the world's largest technology companies are already making massive commitments to nuclear energy.

Microsoft signed a 20-year deal to restart the Three Mile Island nuclear plant, which has been offline since 2019, specifically to power its AI operations. Amazon paid $650 million for a data center campus directly co-located with the Susquehanna nuclear station in Pennsylvania. Google announced agreements with Kairos Power for small modular reactors, while Meta has been pursuing similar nuclear partnerships and recently issued a request for proposals seeking up to 4 gigawatts of new nuclear capacity. These are not speculative investments; they represent admissions that scarce, secured, low-carbon power is now the most important asset in the AI economy.

What Are Small Modular Reactors, and Why Do They Matter?

Small modular reactors (SMRs) represent a fundamentally different approach to nuclear power. Unlike traditional reactors that generate 1,000 megawatts or more, SMRs are designed to be smaller, factory-built, and deployable at scale. Canada just achieved a historic milestone by lowering a 953-tonne basemat into a 35-meter shaft in Ontario, marking the beginning of construction on the Western world's first grid-scale small modular reactor.

The reactor, a GE Vernova Hitachi BWRX-300, is designed to generate 300 megawatts of electricity, enough to power roughly 300,000 homes. The basemat, which weighs about 2.1 million pounds (equivalent to three fully loaded Airbus A380s), was welded together in one piece before being lowered into place on May 1, 2026. This moment marks the official start of construction under Canadian nuclear regulations. Ontario Power Generation, the provincial utility building the reactor, plans to have it generating electricity by the end of 2030, with three additional identical units planned for the same Darlington site.

The BWRX-300 is deliberately conventional in design, using natural circulation cooling instead of electric pumps and standard low-enriched uranium fuel. According to GE Vernova, the entire power block fits within two international soccer pitches, roughly 430 by 200 feet. The developer claims that build time will drop to 24 to 36 months for later, repeat units once the first one has worked out the kinks. This standardization is critical to the economic case for SMRs; the theory is that building identical machines repeatedly will drive down costs the way the Darlington refurbishment shaved 250 days off its second reactor compared with its first.

How Much Will This Cost, and What's the Timeline?

Ontario Power Generation's cost estimate puts the first reactor at CAD 6.1 billion (approximately USD 4.6 billion), plus another CAD 1.6 billion for shared infrastructure like roads, tunnels, and cooling-water lines. The full four-unit program is budgeted at CAD 20.9 billion (about USD 15 billion) in 2024 dollars, with interest and contingencies included. The company expects each later unit to cost less as the supply chain matures, dropping to roughly CAD 4.1 billion for the fourth reactor.

For that investment, the four reactors will deliver 1,200 megawatts of capacity, enough for about 1.2 million homes. The Conference Board of Canada estimates the program will add CAD 38.5 billion to the national economy over 65 years and sustain 18,000 jobs per year during the five-year construction phase, with more than 80% of spending going to Canadian companies. Ontario's grid operator estimated the power at about 14.9 cents per kilowatt-hour and judged a comparable build-out of wind, solar, and storage to be both pricier and riskier, though this calculation depends entirely on the first reactor coming in on budget.

What's Happening in the United States?

The Darlington project is not an isolated Canadian experiment. It is the reference unit for a much larger global expansion of SMR technology. In the United States, the Tennessee Valley Authority became the first U.S. utility to file a construction permit application for a BWRX-300, for a single reactor at its Clinch River site near Oak Ridge, Tennessee. The U.S. Nuclear Regulatory Commission docketed that application in July 2025 and expects to finish its review by the end of 2026.

Beyond Tennessee, a U.S. firm called Elementl Power unveiled plans for a 1.5 gigawatt nuclear plant in Southeast Ohio featuring multiple BWRX-300 reactors. The company recently submitted a formal application to PJM Interconnection, the regional grid operator, seeking permission to connect the first 600 megawatts of electricity directly into the transmission network. PJM is expected to provide an official response before the end of 2026. The proposed site covers nearly 700 acres in Letart Township, Meigs County, positioned along the Ohio River roughly 100 miles southeast of Columbus.

"Nuclear projects are substantial economic anchors for their communities, and with a proud industrial legacy, southeast Ohio brings the foundation and workforce needed for a project of this magnitude," said Chris Colbert, Chairman and CEO of Elementl.

Chris Colbert, Chairman and CEO of Elementl

Elementl finalized an Early Works Agreement with GE Vernova Hitachi Nuclear Energy to deploy the BWRX-300 reactors at the Ohio site, designating it as one of the earliest commercial SMR initiatives in the United States. Construction on the first unit is expected to begin in 2030, subject to final investment decision and regulatory approvals, with an anticipated completion date of 2034. Elementl intends to fund the entire facility independently through private capital, meaning the costs will not be recovered through local utility customer rates.

How to Understand the Global SMR Expansion

• International Momentum: Poland's Orlen Synthos Green Energy plans a fleet of about 24 BWRX-300 reactors, with its first unit targeted near Wloclawek by 2032. Utilities in Sweden, Estonia, Hungary, and the Canadian province of Saskatchewan are all pursuing similar paths, making SMRs a genuinely global strategy rather than a niche technology.
• Standardization as Cost Driver: The entire economic argument for SMRs rests on building the identical machine over and over until the price comes down. If Darlington succeeds in bringing the first reactor in on budget, the next two dozen become easier and cheaper to finance. If the budget blows, every utility watching will quietly revise its plans downward.
• Regulatory Validation: Russia and China already operate small modular reactors, and Argentina has a pilot under construction. However, Darlington is the first grid-scale SMR that a G7 country has built and connected to a major grid, making it the critical proof point for Western democracies considering this technology.

The basemat is now set in Ontario, cranes are lifting the reactor building out of the shaft, and Ontario Power Generation is leaning on roughly 7,000 lessons it logged refurbishing its existing Darlington reactors. The concrete is in the ground, which is further than anyone in the Western world has gotten with this technology before. Whether this represents the beginning of a nuclear renaissance or a cautionary tale about cost overruns will determine the trajectory of AI infrastructure investment for the next decade.

The parallel to NVIDIA's rise is striking. Investors who identified the chip bottleneck early made generational wealth. Now, the same dynamic is playing out with electricity. The companies that own power at scale, in the right jurisdictions, at the right cost, ready to serve AI workloads today, are positioning themselves as the next generation of infrastructure winners. Unlike the chip shortage, which lasted 18 to 24 months, the power shortage is a 10-year problem. That means the wealth transfer will be proportionally larger.

" }