You can ask the internet anything, but getting an answer via generative artificial intelligence consumes about 10 times more electricity than a traditional Google search. Consequently, the data centers where AI tools are trained and run are guzzling more and more power. Though their electricity consumption increased by only 6 percent from 2010 to 2018, these sprawling facilities are now on track to double their electricity demand in the next two years, thanks to AI. This energy- intensive boom poses an expensive challenge for big tech firms like Microsoft, Amazon, and Google, whose data centers run around the clock.

Some industry observers wonder if AI’s potential returns can outweigh these rising electric bills—and the cost to the climate from the widespread use of fossil fuels in the sector. But Bill Nussey (MBA 1996), a former tech CEO who is now a partner at the Atlanta- based venture capital firm Tech Square Ventures, sees a solution in the efficiency gains that widespread new technologies inevitably achieve and the tech sector’s embrace of renewable energy.

Experience thus far supports Nussey’s view. Similar alarms were raised about cloud computing more than a decade ago, but a study funded by the Department of Energy found that, over a five-year stretch, reality didn’t match predictions.

Already AI technology is making progress, Nussey says. He points to the July release of GPT-4o mini, a lightweight version of the technology that takes less electricity to run. He foresees AI model training becoming less energy intensive, too, with costs dropping as much as 90 percent in two to three years with the development of more efficient computer chips and training algorithms. “Just as the price of computing has dropped maybe a hundred million times—no exaggeration—in the last 40 years, that same technology curve will also reduce the cost of AI,” he explains.

And more AI processing will be done right on our smartphones and laptops as personal computing power increases. In recent months, Google and Apple announced new AI capabilities for their respective next generations of smartphones. Such a shift would decrease data centers’ electricity draws and help the sector’s bottom line by shifting the AI processing onto users’ personal devices.

To train AI models at data centers without blowing climate goals and damaging bottom lines, many tech firms are also investing in renewable energy, says Nussey, who helped launch several clean energy ventures and researched the energy industry extensively for his 2021 book, Freeing Energy: How Innovators Are Using Local-scale Solar and Batteries to Disrupt the Global Energy Industry from the Outside In.

Many data center owners have already cemented deals with utilities to get more of their electricity from sources like solar or wind. In May, Microsoft signed a $10 billion deal to increase the amount of renewable energy powering its data centers. Such “power-purchase agreements,” Nussey says, are popular with the tech firms because they can advance their internal climate goals but leave the energy infrastructure building and maintenance to others.

The resulting capacity won’t be available overnight, though. Nussey points out that energy from a brand-new solar farm might have nowhere to go because, in some parts of the United States—home to one-third of the world’s data centers—energy-grid transmission centers are “full” and can’t take in and direct power. “These choke points are a serious problem,” he says, noting that upgrades could take up to 10 years. Data centers could sidestep the issue by dramatically reducing their reliance on the grid, Nussey says, imagining, for instance, a rural four-acre data center powered directly by 50 acres of solar panels right next door. Batteries would store excess energy for nighttime and cloudy days. With the 24/7 demands of the internet, relying on local solar “feels risky to people,” he acknowledges, but new AI technology is going to demand new energy solutions.

“I’m looking at a startup right now that’s streamlining the ability to take data centers and power them directly with solar and batteries. There are so many ways to creatively address the concern that AI is going to be problematic on the grid and climate.”

Post a Comment