Picture New York City on a sweltering summer night: every air conditioner straining, subway cars humming underground, towers blazing with light. Now add San Diego at the peak of a record-breaking heat wave, when demand shot past 5,000 megawatts and the grid nearly buckled.

That’s almost the scale of electricity that Sam Altman and his partners say will be devoured by their next wave of AI data centers—a single corporate project consuming more power, every single day, than two American cities pushed to their breaking point.

The announcement is a “seminal moment” that Andrew Chien, a professor of computer science at the University of Chicago, says he has been waiting a long time to see coming to fruition.

“I’ve been a computer scientist for 40 years, and for most of that time computing was the tiniest piece of our economy’s power use,” Chien told Fortune. “Now, it’s becoming a large share of what the whole economy consumes.”

He called the shift both exciting and alarming.

“It’s scary because … now [computing] could be 10% or 12% of the world’s power by 2030. We’re coming to some seminal moments for how we think about AI and its impact on society.”

This week, OpenAI announced a plan with Nvidia to build AI data centers consuming up to 10 gigawatts of power, with additional projects totaling 17 gigawatts already in motion. That’s roughly equivalent to powering New York City—which uses 10 gigawatts in the summer—and San Diego during the intense heat wave of 2024, when more than five gigawatts were used. Or, as one expert put it, it’s close to the total electricity demand of Switzerland and Portugal combined.

“It’s pretty amazing,” Chien said. “A year and a half ago they were talking about five gigawatts. Now they’ve upped the ante to 10, 15, even 17. There’s an ongoing escalation.”

Fengqi You, an energy-systems engineering professor at Cornell University, who also studies AI, agreed.

“Ten gigawatts is more than the peak power demand in Switzerland or Portugal,” he told Fortune. “Seventeen gigawatts is like powering both countries together.”

The Texas grid, where Altman broke ground on one of the projects this week, typically runs around 80 gigawatts.

“So you’re talking about an amount of power that’s comparable to 20% of the whole Texas grid,” Chien said. “That’s for all the other industries—refineries, factories, households. It’s a crazy large amount of power.”

Altman has framed the build-out as necessary to keep up with AI’s runaway demand.

“This is what it takes to deliver AI,” he said in Texas. Usage of ChatGPT, he noted, has jumped 10-fold in the past 18 months.

Which energy source does AI need?

Altman has made no secret of his favorite source: nuclear. He has backed both fission and fusion startups, betting that only reactors can provide the kind of steady, concentrated output needed to keep AI’s insatiable demand fed.

“Compute infrastructure will be the basis for the economy of the future,” he said, framing nuclear as the backbone of that future.

Chien, however, is blunt about the near-term limits.

“As far as I know, the amount of nuclear power that could be brought on the grid before 2030 is less than a gigawatt,” he said. “So when you hear 17 gigawatts, the numbers just don’t match up.”

With projects like OpenAI’s demanding 10 to 17 gigawatts, nuclear is “a ways off, and a slow ramp, even when you get there,” Chien said. Instead, he expects wind, solar, natural gas, and new storage technologies to dominate.

You, the energy-systems expert at Cornell, struck a middle ground. He said nuclear may be unavoidable in the long run if AI keeps expanding, but cautioned that “in the short term, there’s just not that much spare capacity”—whether fossil, renewable, or nuclear. “How can we expand this capacity in the short term? That’s not clear,” he said.

He also warned that timeline may be unrealistic.

“A typical nuclear plant takes years to permit and build,” he said. “In the short term, they’ll have to rely on renewables, natural gas, and maybe retrofitting older plants. Nuclear won’t arrive fast enough.”

Environmental costs

The environmental costs loom large for these experts, too.

“We have to face the reality that companies promised they’d be clean and net zero, and in the face of AI growth, they probably can’t be,” Chien said.

Ecosystems could come under stress, Cornell’s You said.

“If data centers consume all the local water or disrupt biodiversity, that creates unintended consequences,” he said.

The investment figures are staggering. Each OpenAI site is valued at roughly $50 billion, adding up to $850 billion in planned spending. Nvidia alone has pledged up to $100 billion to back the expansion, providing millions of its new Vera Rubin GPUs.

Chien added that we need a broader societal conversation about the looming environmental costs of using that much electricity for AI. Beyond carbon emissions, he pointed to hidden strains on water supplies, biodiversity, and local communities near massive data centers. Cooling alone, he noted, can consume vast amounts of fresh water in regions already facing scarcity. And because the hardware churns so quickly—with new Nvidia processors rolling out every year—old chips are constantly discarded, creating waste streams laced with toxic chemicals.

“They told us these data centers were going to be clean and green,” Chien said. “But in the face of AI growth, I don’t think they can be. Now is the time to hold their feet to the fire.”