Marketplace®
Every story is an economic story

How much is AI contributing to climate change? California is trying to find out

California is making big companies, including AI firms, report their carbon emissions. But a recent Ninth Circuit court ruling could make it harder to track AI’s full carbon footprint.

A drone photo shows cooling units on the roof of a data center in Vernon, California, on Nov. 4.
A drone photo shows cooling units on the roof of a data center in Vernon, California, on Nov. 4.
Allen J. Schaben/Los Angeles Times via Getty Images

If you want to know how much CO2 gets emitted on a drive to the supermarket, there's an app for that.

But how much climate gas gets spilled when you ask ChatGPT how to cook with what you just bought? There's not an app for that — no way to know how much CO2 goes into a single AI search, or to figure out the carbon footprint of the AI industry.

"We don't know for sure, and that's because we don't have the data," said Eric Masanet, a professor at the Bren School of Environmental Science and Management at the University of California, Santa Barbara. He’s on a quest to piece together AI's carbon numbers indirectly.

"A lot of the tech companies that operate the data centers don't report emissions, and when they do, sometimes the reporting can vary from company to company," he said.

Even if researchers can estimate some of this, they can't always be sure what's going on inside these data centers — is it AI number-crunching or maybe just streaming some Netflix?

Sure, some companies like Google and Microsoft publish their overall carbon numbers, thanks to disclosure requirements from the European Union. But these reports are a bit like a nonitemized restaurant receipt, where you can't tell the food from the booze.

"Currently, there's no line item for AI in most or any [Environmental, Social, and Governance] report coming from the tech industry," Masanet said.

But last week, some of the biggest companies had to spill the beans on the greenhouse gas they're emitting, plus any CO2 produced by the power they use. It's a California deadline that applies to any company, public or privately held, that does a billion dollars or more of business in the state.

That's about 2,500 firms, and it includes the big AI companies. Until now, there's been no carbon report from, say, OpenAI or Anthropic — the biggest of AI fish — or some big data center firms.

"The hope is that once we start actually disclosing this data, and it's publicly available, companies are going to be pressured into cleaning up," said Mary Nichols, former chair of the state's powerful air quality regulator, the California Air Resources Board, and now at UCLA law school.

"One of the things that, now, people are beginning to realize: We don't know enough about data centers," Nichols said. "What do they actually do? What do they actually use? How much of our grid capacity? And what can be done to make them acceptable in communities or not?"

There's been a flurry of outside consultants offering to help firms gather the data. Barnaby Lynch is head of regulatory reporting at Watershed, a sustainability data platform.

"I have my flight data from all my employees traveling around. I also have all my energy bills," Lynch said. "And we can say, ‘OK, well, where are your trucks driving, and what type of truck are they, and where are you consuming energy?’ And so we can then get the most granular version of what your carbon footprint looks like."

Just hours before last week's deadline, a federal appeals court upheld this set of reports. But in a plot twist, the court decided that more detailed carbon reports due next year are illegal.

Those would have forced companies to assess greenhouse gas emissions up and down their supply chains, touching not just the companies themselves, but third parties — including suppliers. The California Chamber of Commerce called this a victory for business, but the state could still appeal to the U.S. Supreme Court.

The industry prefers a different kind of accounting, according to Masanet.

"Let's be honest, there could be some ways that AI could lead to societal-level efficiencies, or savings, or emissions reductions. That's what the tech industry is promising us," he said. "The challenge is we don't know if they're going to be real, and if they'll be large enough to offset our use of AI data centers and infrastructure."

Proof will require more hard data from the industry.

Related Topics

Latest Episodes

View All Shows
  • Marketplace Morning Report
    6 hours ago
    7:02
  • Marketplace Tech
    12 hours ago
    7:01
  • Marketplace
    3 days ago
    25:28
  • This Is Uncomfortable
    4 days ago
    42:59
  • Million Bazillion
    13 days ago
    29:48
  • Make Me Smart
    5 months ago
    24:33