400 terawatt-hours. That is how much electricity U.S. data centers could consume by 2030, up from fewer than 100 TWh in 2020. To put that in human terms: more electricity than the entire country of Mexico uses today. And the AI models driving that demand do not sleep, do not take weekends off, and do not care whether the sun is shining in Virginia.
That last part is the whole argument. Not the hype about nuclear, not the fear about nuclear. Just that one physical constraint: data centers need power every hour of every day, and the grid we have cannot deliver that cleanly at the scale we need.
So when people frame this as "nuclear versus renewables," I want to stop them right there. That framing is wrong, and it is slowing us down.
The Capacity Factor Problem Nobody Wants to Talk About
Here is the engineering reality. Nuclear power plants run at a capacity factor exceeding 92.5%, compared to wind at 35% and solar at 25%. A capacity factor is simply the ratio of actual output to maximum possible output. When your data center needs 500 megawatts right now, at 2 a.m. on a cloudy, windless Tuesday in January, a solar farm is producing zero. A nuclear plant is producing 92% of its rated output. That is not an ideological statement. That is physics.
Goldman Sachs Research puts the cost of a near-100% renewable solution for data centers, including offsite solar or wind plus battery storage, at around $87 per megawatt-hour. A large-scale onsite nuclear generator comes in at $77 per MWh. Renewables are cheaper per panel. Nuclear is cheaper per reliable megawatt-hour delivered around the clock. Those are different products.
Vera would say fix the grid first, build more storage, let efficiency gains close the gap. I take that seriously. DeepSeek's R1 model reportedly operates up to 40% more efficiently than its predecessors. Efficiency is real and it matters. But efficiency gains trigger the Jevons paradox: cheaper AI means more AI. When AI becomes cheaper to operate, demand typically increases proportionally, or even exceeds the efficiency gains. We are not engineering our way out of the demand curve. We are building into it.
Big Tech Has Already Voted With Its Contracts
The smartest energy buyers on the planet have studied this problem longer and harder than any op-ed writer, and they keep arriving at the same answer: you need both.
Meta signed a 20-year PPA with Constellation Energy to extend the life of the 1.1 GW Clinton Clean Energy Center in Illinois. Microsoft signed a 20-year PPA to reopen Three Mile Island. Amazon followed with a nearly 2 GW PPA with Talen Energy's Susquehanna plant. These are not press releases. These are billion-dollar commitments by engineers who model grid reliability for a living.
At the same time, Goldman Sachs forecasts that 40% of new capacity built to support data center power demand will be renewables. Nobody at Google is tearing down their solar farms. Amazon, Microsoft, Meta, and Google have already contracted over 50 GW of renewable energy PPAs, equal to the generation capacity of Sweden. These companies understand that wind and solar are fast and cheap to deploy. They also understand that firm, dispatchable, carbon-free power is something different, and that you need both to run a clean 24/7 operation.
Google's senior energy director said it plainly: "We know that wind, solar and batteries will be critical in order to decarbonize our energy consumption. But we also need firm, dispatchable, carbon-free electricity technologies." That is not a nuclear evangelist talking. That is an engineer describing a system requirement.
The Honest Constraints and Why They Don't Change the Conclusion
I am not going to pretend nuclear builds itself overnight. Capex for nuclear power plants is estimated at $6,417 to $12,681 per kilowatt, compared to $1,290 per kilowatt for natural gas. Large reactors take five to eleven years from groundbreaking to power. The first SMRs are likely not coming online until around 2030. Those are real constraints and anyone selling you a nuclear-only story is skipping them.
But here is what I keep coming back to: global electricity generation to supply data centers is projected to grow from 460 TWh in 2024 to over 1,000 TWh by 2030. That is more than doubling in six years. The question is not which single technology wins. The question is whether we can build fast enough across all of them. Renewables scale quickly and should. Nuclear takes longer but delivers something renewables physically cannot: firm, dense, weather-independent power with a capacity factor above 90%.
The nuclear skeptics are not wrong about the timeline. They are wrong about the conclusion they draw from it. The right response to "nuclear takes a decade" is not "therefore don't build it." The IEA projects nuclear will play an increasingly important role toward the end of this decade and beyond precisely because the projects we commit to today are the plants that come online in 2030 and 2035. A decade from now, we will either have that baseload capacity or we won't. That decision is being made right now, in boardrooms and regulatory offices and engineering firms.
This is the future and it is happening now. The grid that powers AI civilization is going to be a hybrid: fast renewables for the watts we can get cheaply, firm nuclear for the watts we need reliably. Treating them as rivals is a failure of systems thinking. They are components of the same machine. Build both. Build them fast. The timeline is just a negotiation.
