Ben Parr: Why Big Tech Is Betting Billions on Nuclear to Power AI

From WWSG thought leader, Ben Parr. To host Ben for a speaking engagement, contact us.

AI is on a power binge, and the energy bills are staggering. By 2026, AI’s electricity usage will reach 1,000 terawatt-hours (TWh), more than double the 460 terawatt-hours it used in 2022. For context, one terawatt-hour is enough to power around 90,000 U.S. homes for a year. 1000 terawatt-hour is the electricity consumption of all of Japan in a year. Yes, an entire country.

But instead of cutting back, tech giants are doubling down—with nuclear reactors. Who could have imagined chatbots and image generators would now be driving a massive push toward nuclear energy? The AI revolution is now an all-out scramble for power.

I recently discussed this shift to nuclear power on NBC Bay Area, shedding light on why companies like Google and Amazon are rushing to lock down nuclear deals to fuel their AI ambitions. But a few minutes on TV isn’t enough time to dive into this fascinating topic.

Let’s dig into why this is happening — and whether it’s a good thing.

Why Are AI Giants Grabbing All the Nuclear Power They Can?

Google is hooking up with Kairos Power to develop small modular reactors (SMRs) that will create up to 7 million kilowatt-hours per day of clean energy per unit by 2035. Amazon just poured $500 million into X-energy for its own set of small modular reactors in Washington, potentially cranking out enough power to light up 770,000 homes. Even Microsoft’s jumping in, dusting off the Three Mile Island nuclear plant to power its data centers.

All these deals were announced in just the last 30 days. So why are these tech companies turning to nuclear now? It’s all about stability and scalability—something even the best renewables can struggle with.

Why Does AI Need This Much Power?

Let’s talk about the elephant in the server room: Transformers — the core technology behind large language models like GPT, Claude, Gemini and Perplexity. Unlike older models that processed data sequentially, transformers handle massive datasets in parallel, simultaneously juggling billions (sometimes trillions) of parameters. This exponentially increases the amount of number-crunching required for training these models, demanding significantly more GPUs and specialized chips, exponentially increasing their electricity use over previous models like recurrent neural networks (which struggle with larger datasets).

The power hogging doesn’t stop at training, though; these models use power every time they are prompted — a single ChatGPT prompt can use up to 10x the power of a conventional Google search query.

We have made an incredible leap in AI technology at the cost of increased power demand. But in the larger scope of humanity, I believe the trade-off is worth it — these new AI models are already helping us diagnose cancer, find clean geothermal energy sources, and boost our overall productivity. And we’re still in the first inning — we’ve barely scratched the surface of what AI can do for our society.

A Real Talk on Nuclear: The Good, the Bad, and the Radiant

AI is currently driving up our fossil fuel usage. Goldman Sachs predicts that 60% of AI’s power needs will have to be supplied by legacy sources by 2030 — specifically natural gas — if nothing changes.

Nuclear power has a knack for sending people into a bit of a panic. Chernobyl and Fukushima have left imprints in our collective consciousness. However, nuclear energy has a far smaller carbon footprint than coal or gas power — 1/20 to 1/50 the footprint. And for all the fears around nuclear power, most plants — including Chernobyl and Fukushima — were built in the 1970s, over 50 years ago, and nuclear technology is far more advanced today.

Small modular reactors are far safer, far smaller, require far less nuclear refueling (every 3-7 years instead of every 1-2), and can be built faster. They (along with other nuclear energy sources) are the best solution available to us now to decrease our reliance on fossil fuels and meet our energy needs. And unlike solar or wind power (which are also critical), small modular reactors generate a predictable, reliable flow of electricity.

New technologies exist now to reduce nuclear waste. Nuclear technology in 2024 is nothing like it was in 1974.

The cat is out of the bag regarding AI and our society’s power consumption. It’s only going to keep climbing. We’re at a crossroads, and nuclear is one of the few options on the table that can realistically power our future while minimizing climate change. It’s easy to fear nuclear power, but its pros outweigh the cons of fossil fuels.

The bottom line? We’re in big trouble if we don’t bet much more on nuclear. It’s time we stop letting the past hold us back.