AI’s search for more energy is growing more urgent

As AI’s new direction is requiring even larger amounts of energy, tech companies have a problem.

James O'Donnellarchive page

December 17, 2024

Stephanie Arnett/MIT Technology Review | Envato

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.

If you drove by one of the 2,990 data centers in the United States, you’d probably think little more than “Huh, that’s a boring-looking building.” You might not even notice it at all. However, these facilities underpin our entire digital world, and they are responsible for tons of greenhouse-gas emissions. New research shows just how much those emissions have skyrocketed during the AI boom.

Since 2018, carbon emissions from data centers in the US have tripled, according to new research led by a team at the Harvard T.H. Chan School of Public Health. That puts data centers slightly below domestic commercial airlines as a source of this pollution.

That leaves a big problem for the world’s leading AI companies, which are caught between pressure to meet their own sustainability goals and the relentless competition in AI that’s leading them to build bigger models requiring tons of energy. The trend toward ever more energy-intensive new AI models, including video generators like OpenAI’s Sora, will only send those numbers higher.

A growing coalition of companies is looking toward nuclear energy as a way to power artificial intelligence. Meta announced on December 3 it was looking for nuclear partners, and Microsoft is working to restart the Three Mile Island nuclear plant by 2028. Amazon signed nuclear agreements in October.

However, nuclear plants take ages to come online. And though public support has increased in recent years, and president-elect Donald Trump has signaled support, only a slight majority of Americans say they favor more nuclear plants to generate electricity.

Though OpenAI CEO Sam Altman pitched the White House in September on an unprecedented effort to build more data centers, the AI industry is looking far beyond the United States. Countries in Southeast Asia, like Malaysia, Indonesia, Thailand, and Vietnam, are all courting AI companies, hoping to be their new data center hubs.

In the meantime, AI companies will continue to use up power from their current sources, which are far from renewable. Since so many data centers are located in coal-producing regions, like Virginia, the “carbon intensity” of the energy they use is 48% higher than the national average. The researchers found that 95% of data centers in the US are built in places with sources of electricity that are dirtier than the national average. Read more about the new research here.

Deeper Learning

We saw a demo of the new AI system powering Anduril’s vision for war

We’re living through the first drone wars, but AI is poised to change the future of warfare even more drastically. I saw that firsthand during a visit to a test site in Southern California run by Anduril, the maker of AI-powered drones, autonomous submarines, and missiles. Anduril has built a way for the military to command much of its hardware—from drones to radars to unmanned fighter jets—from a single computer screen.

Why it matters: Anduril, other companies in defense tech, and growing numbers of people within the Pentagon itself are increasingly adopting a new worldview: A future “great power” conflict—military jargon for a global war involving multiple countries—will not be won by the entity with the most advanced drones or firepower, or even the cheapest firepower. It will be won by whoever can sort through and share information the fastest. The Pentagon is betting lots of energy and money that AI—despite its flaws and risks—will be what puts the US and its allies ahead in that fight. Read more here.

Bits and Bytes

Bluesky has an impersonator problem

The platform’s rise has brought with it a surge of crypto scammers, as my colleague Melissa Heikkilä experienced firsthand. (MIT Technology Review)

Tech’s elite make large donations to Trump ahead of his inauguration

Leaders in Big Tech, who have been lambasted by Donald Trump, have made sizable donations to his inauguration committee. (The Washington Post)

Inside the premiere of the first commercially streaming AI-generated movies

The films, according to writer Jason Koebler, showed the telltale flaws of AI-generated video: dead eyes, vacant expressions, unnatural movements, and a reliance on voice-overs, since dialogue doesn’t work well. The company behind the films is confident viewers will stomach them anyway. (404 Media)

Meta asked California’s attorney general to stop OpenAI from becoming for-profit

Meta now joins Elon Musk in alleging that OpenAI has improperly enjoyed the benefits of nonprofit status while developing its technology. (Wall Street Journal)

How Silicon Valley is disrupting democracy

Two books explore the price we’ve paid for handing over unprecedented power to Big Tech—and explain why it’s imperative we start taking it back. (MIT Technology Review)