What Will Power the AI Revolution?

Here’s a law of the universe: Revolutionary inventions lead to massive new demands for energy.

Think about it: the steam engine led to a massive new demand for coal. The automobile led to a massive new demand for gasoline. The commercial jetliner led to a massive new demand for jet fuel. The light bulb – or the air conditioner? Massive new electricity demands.

Here’s a second law of the universe: When new energy demands arise, supply inevitably follows. 

Right now, we’re in the early days of another revolutionary invention: Artificial intelligence.

AI is a big deal, perhaps the biggest in history. That’s why Google, Meta, and dozens of other tech giants are spending hundreds of billions of dollars to deploy it. 

Skeptics wonder if it will pay off.

History, however, suggests that it will.

Like all earlier tech revolutions, AI not only creates new kinds of services, but dramatically increases productivity. 

This is important because, over the last two decades, America’s productivity growth has slowed. When that happens, real wages decrease, and the economy creates fewer jobs.  

“Wait a second,” I can hear you saying. “I heard that AI is going to trigger a jobs apocalypse — or worse. I’ve seen The Terminator and The Matrix. I know how this goes.”

That makes for good science fiction, but it’s not how tech revolutions play out. 

Just as the steam engine, automobile, and jetliner boosted economic growth, so will artificial intelligence. It’s hard to predict exactly how — which scares some people — but odds are high that history will repeat.

But where will the energy come from to power this latest energy-guzzling technology?

That question has come up with every revolutionary invention. 

It came up when coal use jumped more than 400 percent during the first half of the nineteenth century. Many feared coal supplies would run out as steam engines — the key feature of the Industrial Revolution — powered factories and ships. But the coal supply and the economic boom kept, well, booming.

More recently, the three decades following World War II saw oil use triple as more people bought automobiles, flew on airplanes, and air-conditioned their homes.

In 1977, President Jimmy Carter warned, “We could use up all the proven reserves of oil in the entire world by the end of the next decade.”

Today, thanks to technology, especially the fracking revolution, the world has plenty of oil. And natural gas, too.

Which brings us back to artificial intelligence.

AI lives on massive amounts of data, which requires massive datacenters, which require massive amounts of electricity.

There’s never been anything quite like it. 

To meet the lower end of the range of forecasts for the growth in datacenter power by 2030, we’ll need to build enough generating plants to power five New York Cities— more than powers than the entire nation of Sweden.

That explains all the headlines about AI using too much energy. That also explains why environmental activists suffer from what we could call fuel phobia. They’re not worried we’ll run out of energy. They worry we will actually be able to supply it.

They’re especially worried that we’ll use lots more natural gas and even coal. To them, that’s a Faustian Bargain, a deal with the global warming devil. If we need more energy for growth, they say, it will have to be wind and solar. 

That might sound appealing to environmentalists, but it won’t work, not only because AI’s energy appetite is voracious, but especially because AI runs every hour, every day, all year. It takes sources like fossil fuels and nuclear to provide that kind of reliable power. 

Tacitly acknowledging this, the no-growth, global-warming catastrophe crowd offers two other equally impractical solutions.

We should make — even require — far more energy efficient AI which, they say, will reduce energy demand. 

But it won’t.

Here I invoke Mills’ Law of Energy Reality. Improving efficiency always creates more demand for energy. 

Sound counter-intuitive? Consider this: Your smartphone is tens of thousands of times more powerful and millions of times more energy efficient than the room-sized mainframe computers that guided astronauts to the moon. But instead of reducing energy demand, smartphones increased it exponentially. Why? Because the few thousand 1960s mainframe computers have been replaced by billions of smart phones. 

So, no surprise, the no-growthers also want us to go on a digital diet.

Good luck trying to get people to restrain themselves when a transformational technology improves their lives. It didn’t work with steamships, automobiles, airplanes, or, for that matter, washing machines, refrigerators, laptops, or pharmaceuticals.

It won’t work with AI. 

Innovation leads to new energy demands. New energy demands lead to new supply. And efficiency democratizes the innovation. 

That’s just how our universe works. 

It’s called progress. 

I’m Mark Mills, author of The Cloud Revolution, and executive director of the National Center for Energy Analytics, for Prager University.