Latest / Minds, Bodies, and Terawatts / Google's AI Buildout Drove a Record 37% Jump in Its Electricity Use as Emissions Climb

Transcript

Google's AI Buildout Drove a Record 37% Jump in Its Electricity Use as Emissions Climb · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, where the smartest machines on earth keep running up somebody's power bill. Ars Technica reported on July 2nd that Google's electricity use jumped 37 percent last year, the biggest single-year increase in the company's history.
• 0:16
  [Guest] And that's from Google's own sustainability report, not a critic's estimate. Its data centers alone burned more than 42 million megawatt-hours in 2025.
• 0:25
  [Host] Companies use more power every year though. Why is that the number that stops you?
• 0:31
  [Guest] Because of the scale. Google's data centers now draw about as much electricity as an entire country, roughly what New Zealand or Denmark uses in a year.
• 0:42
  [Host] One company, pulling down a whole nation's worth of power.
• 0:45
  [Guest] And the driver is one thing. Not search, not YouTube. The AI buildout. Google says its climb won't be linear because, its words, the buildout is accelerating faster than the grid is decarbonizing.
• 0:59
  [Host] But Google's supposed to be the green one. They've been buying wind and solar for years.
• 1:05
  [Guest] They have, and they still matched all that power with clean energy. Their total emissions rose 18 percent anyway, the biggest jump they've ever reported. Buying green electricity doesn't cover the carbon baked into building the chips and servers themselves.
• 1:21
  [Host] So even the cleanest version of this still pushes emissions up. And it's not just Google?
• 1:27
  [Guest] Amazon reported record emissions the same week, same cause, data centers for AI. The whole industry's demand is outrunning the clean supply.
• 1:36
  [Host] Here's my honest worry. When one company needs a nation's worth of power, who builds the grid to deliver it, and who pays for it?
• 1:46
  [Guest] That's the fight nobody put on a ballot. When one campus wants a gigawatt on a two-year timeline, either the utility builds fast and spreads the cost across everybody's bill, or the tech company builds its own power plant.
• 2:01
  [Host] And Google's picking the second one?
• 2:04
  [Guest] More and more. On-site gas, batteries, deals for nuclear and fusion later. When your demand grows exponentially and the grid grows a few percent a year, you stop waiting in line.
• 2:16
  [Host] That gap between exponential and a few percent, somebody said it out loud, didn't they?
• 2:22
  [Guest] Elon Musk, on stage at Davos in January, told BlackRock's Larry Fink the bottleneck plainly. AI chips are scaling exponentially, but power generation crawls up only a few percent a year. The chips aren't the ceiling anymore. The watts are.
• 2:38
  [Host] So the cloud that feels weightless is actually one of the heaviest things on the grid.
• 2:44
  [Guest] That's the illusion. We talk about intelligence like it's pure thought, but every answer is electrons by the billion. You can double a model overnight. Transmission lines and substations take years.
• 2:56
  [Host] Which is why everyone's suddenly betting on nuclear and fusion.
• 3:00
  [Guest] It's the only supply answer big enough. Google already signed to buy power from a future fusion plant. But fusion shows up in the 2030s, and the demand is here now.
• 3:12
  [Host] So for the rest of this decade, we close the gap with gas turbines and hope.
• 3:19
  [Guest] Pretty much. The abundance of intelligence is real. The energy to run it is finite, at least until something like fusion matures.
• 3:28
  [Host] It's strange. We built the most advanced tool in history, and the thing capping it is what runs your refrigerator.
• 3:35
  [Guest] And whoever fixes that inherits the whole AI era.
• 3:39
  [Host] Patrick Deglon calls this the Electron Gap in Unscarcity, energy, not chips, as the real ceiling on how far intelligence goes.
• 3:47
  [Guest] Digital abundance looks unlimited, but it's held hostage by a very physical resource. For more on the infrastructure layer of AI and who ends up paying for it, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.