The AI gold rush has a power problem nobody wants to talk about. Data centers already consume about 2% of all U.S. electricity, according to the Department of Energy, and that number is accelerating fast. TerraPower’s Natrium reactor, a 345 MW nuclear plant that can surge to 500 MW on demand, is not a clean energy feel-good story. It is an AI infrastructure bet disguised as a utility project.
Why This Matters Right Now
In 2026, every major tech company is racing to build AI infrastructure at a scale that was unthinkable five years ago. Microsoft, Google, and Amazon have each committed over $50 billion to data center expansion this year alone. The bottleneck is not chips. It is not software. It is power.
Clean, reliable, always-on electricity is not keeping pace. Solar and wind are intermittent. Natural gas is politically complicated and getting more expensive. Coal is done.
TerraPower, backed by Bill Gates, is building its first Natrium reactor in Kemmerer, Wyoming, on the site of a retired coal plant. Construction is underway. According to TerraPower, the Natrium design adjusts its power output from 345 MW all the way to 500 MW within hours using a molten salt thermal storage system. That flexibility is something no other baseload energy source can match right now.
According to the International Energy Agency, global data center electricity consumption could double by 2026 compared to 2022 levels, reaching over 1,000 terawatt hours per year. That is more electricity than most entire countries consume.
The Angle Most Investors Are Missing
Everyone is pitching solar plus battery storage as the answer to AI power demand. I think that is the wrong bet, and here is why.
AI training runs are not predictable. You do not get to tell a GPU cluster to pause because the sun went down. Large language model training can run continuously for weeks without interruption. According to Goldman Sachs Research, training a single large AI model can consume as much electricity as 1,000 U.S. homes use in a full year. That kind of sustained load does not play well with intermittent renewables.
Nuclear power is different. It runs at over 90% capacity factor, according to the U.S. Energy Information Administration. Compare that to solar at around 25% and wind at around 35%. For a hyperscale data center that needs guaranteed uptime, nuclear is the only source that does not require enormous battery backup investment sitting idle most of the time.
The Natrium reactor’s molten salt storage system is the real story here. When the grid needs less power, the reactor stores thermal energy in molten salt tanks. When demand spikes, it releases that stored energy and pushes output to 500 MW. For an AI data center that needs to run a massive training job on short notice, that surge capacity is worth a serious premium.
Most investors are still pricing nuclear as a legacy energy story. I see it as an AI infrastructure play wrapped in a nuclear package.
The rich are already moving. Microsoft signed a 20-year deal to restart Three Mile Island Unit 1. Google signed agreements with Kairos Power for small modular reactor output. The smart money is not waiting for proof of concept. It is locking in capacity years before it comes online, because by the time the lights turn on, every megawatt will already be spoken for.
The poor mindset says nuclear is risky, slow, and expensive, so wait and see. The owner mindset says lock in long term power contracts before demand outstrips supply and prices explode. These are two very different positions, and only one of them builds wealth.
If you are running a business that touches AI infrastructure or energy contracting, having real time visibility into your capital spend matters more than most operators realize. Wallester’s business card platform lets you set spending limits by category and track every dollar across multiple vendor contracts, which is the kind of control that keeps a fast-moving infrastructure build from becoming a money leak.
What This Means for You
You are probably not building a nuclear reactor. But you might be building a company that depends on AI, or investing in one. Here is what I would do.
First, watch where the hyperscalers are signing power deals. When Microsoft or Google locks up nuclear capacity in a region, commercial real estate adjacent to that reliable power source appreciates. Data center rents follow power certainty.
Second, look at the companies supplying TerraPower and other advanced reactor builders. Sodium-cooled reactor construction requires specialized components and materials. The supply chain for advanced nuclear is thin and getting thinner. Smaller companies in this space could see serious upside as more reactors get funded and the race for parts accelerates.
Third, if you are hiring in the energy or AI space right now, get your payroll infrastructure right before you scale. I use Gusto for managing payroll across technical teams, especially when hiring engineers across multiple states. The compliance overhead alone makes a managed payroll solution worth it when you are growing fast and cannot afford HR mistakes.
Fourth, do not overlook the grid modernization angle. Every new nuclear plant connects to a grid built for a different era. Companies building software and hardware for grid management and power routing will see decades of tailwind from this transition.
The window to position ahead of this shift is not open indefinitely. Power deals get structured years before reactors come online. By the time TerraPower’s Natrium reactor is producing electricity, the capacity will already be spoken for by whoever moved first.
The Bottom Line
TerraPower is not a charity project. It is a bet that AI is going to need more reliable power than the current grid can deliver, and that nuclear is the only source that can do it at scale without a massive intermittency problem. The people who secure power contracts early will set the terms. Everyone else will pay whatever the market charges. That gap between those two groups is where fortunes get made and lost.
Frequently Asked Questions
What is TerraPower’s Natrium reactor?
The Natrium reactor is a 345 MW sodium-cooled nuclear reactor developed by TerraPower, backed by Bill Gates. It includes a molten salt thermal storage system that can boost output to 500 MW for short periods on demand. The first commercial unit is being built in Kemmerer, Wyoming, on the site of a retired coal plant.
Why is nuclear power better for AI data centers than solar or wind?
Nuclear power runs at over 90% capacity factor, generating electricity almost continuously regardless of weather conditions. AI training workloads run around the clock without interruption, which makes intermittent renewables a poor fit without massive and expensive battery storage investment.
When will TerraPower’s Natrium reactor come online?
According to TerraPower, the Kemmerer facility is expected to begin commercial operations in the late 2020s. Supply chain constraints for specialized reactor components have introduced some delays into the original construction schedule.
How are major tech companies securing nuclear power for AI infrastructure?
Microsoft, Google, and Amazon have each signed long term nuclear power purchase agreements. Microsoft restarted Three Mile Island Unit 1 under a 20-year deal. Google signed agreements with Kairos Power for small modular reactor output. These deals are being structured years before the plants produce a single watt of power.
Is TerraPower publicly traded?
TerraPower is a private company as of 2026. It is not publicly traded. Investors seeking exposure can look at public companies in the advanced nuclear supply chain or energy infrastructure funds that hold positions in private nuclear ventures.
