The PR Machine Powering Big Tech’s AI Energy Story

Google doesn't want you to think AI uses up much energy

Google announced last week that the energy consumed by AI isn’t worth worrying about. The announcement featured new Google data about how much energy asking Google’s AI product uses per search. It came via a 10 page PDF paper (not peer reviewed), a few blog posts, and videos with prominent executives.

This is a big topic of debate right now. People living beside data centers are getting sick. Across America, electricity bills are rising, with one report finding electricity bills in Ohio going up $15 a month because of data centers. In Ireland, data centers use 21% of the country’s energy supply. In San Antonio, Texas, local residents are being asked to ration water amidst a statewide drought, yet the nearby data centers have no such restrictions.

The big AI companies are spending unfathomable amounts of money on building even more data centers to keep up with the energy needed for AI. Per a recent WSJ story:

Capex spending for AI contributed more to growth in the U.S. economy in the past two quarters than all of consumer spending, says Neil Dutta, head of economic research at Renaissance Macro Research, citing data from the Bureau of Economic Analysis.

Despite the massive amount of data they already store, from big AI’s perspective, they need to be way, way bigger. AI proponents have predicted that one day the whole world could be covered with data centers. And yet, despite hundreds of billions of dollars going into these things, the build out and the energy they’re using is not yet a regular front page issue.

The backlash against data centers has barely got off the ground. A group of people last year in Indiana stopped one, just a few weeks ago in Tucson, Arizona facing pressure from locals, the city council rejected an application to build one. These whack a mole efforts are few and far between if you consider the number of data centers built here in the U.S. and elsewhere.

How Big Tech Spin Works

The blog post goes out early in the morning. Within hours you see which reporters accepted the news under embargo, and which of them wrote it up. MIT Tech Review, Axios, CBS, Ars Technica, Politico, Semafor, The Verge, Time Magazine, Fast Company, and others. Google will typically give each reporter a short amount of time to turn the story around, making it difficult for them to speak to experts who have the knowledge to critically assess the claims.

The Google PR team were really happy with the MIT Tech Review write up. While it picked holes in the cherry picked data, the Google Cloud VP of Comms highlighted a quote from it on Linkedin:

“Excited about the level of transparency we are providing here: "It’s the most transparent estimate yet from a Big Tech company with a popular AI product, and the report includes detailed information about how the company calculated its final estimate." Thanks to the entire team that worked on this!”

They were happy with that one because the same publication did an incredible deep dive earlier this year on data centers and their usage (Power Hungry: AI and our energy future). The reporting was extensive and it included stories with critical headlines such as this one like: “We did the math on AI’s energy footprint. Here’s the story you haven’t heard.”

In fairness to the PR team, they did a good job to give reporters a very repeatable, but impossible to verify claim:

“Google says it has calculated the energy required for its Gemini AI service: Sending a single text prompt consumes as much energy as watching television for nine seconds.”

Effectively every story who wrote about the paper repeated the claim. NYT columnist Kevin Roose welcomed Google’s blog and dismissed concerns about the environmental cost of AI as “more vibe-based than anything”.

The only publication that didn’t give a positive or neutral headline was The Verge, they went with this: “Google says a typical AI text prompt only uses 5 drops of water — experts say that’s misleading”. Reporter Justine Calma took the time to speak with a few experts who said Google was “hiding critical information” and using “apple to oranges” comparisons. When pressed with specific questions,

“The company declined to respond on the record to a list of other questions from The Verge.”

So much for transparency. This release and the positive response from tech media would have been months of strategy and careful execution by the AI and PR teams. Other Googler Linkedin posts shared how they want to be perceived on this:

“these numbers are a lot smaller than most people think.”

and

“we believe in a transparent and comprehensive methodology that accounts for the full picture of serving AI at Google’s scale.”

and

“Our teams are sharing this methodology to encourage industry-wide consistency and collaboration”

They’re trying to get out in front of it by presenting Google AI as better than their competitors. They want to be seen as transparent, when others are not. They didn’t answer specific questions from The Verge, yet maintain that they are being transparent.

Ketan Joshi, a climate and energy writer and analyst debunks every claim in this blog post. Give it a read if you’ve time, if not he summarized the blog on this Bluesky thread. He explains how Google presented incomplete and handpicked data to come up a message to advance their agenda. With full control over the data, Google is providing us with an incomplete view which allows the context and framing to become the message.

This isn’t a new issue for Google.

Back in 2015, when I was on Google’s public policy team, I was tasked with developing a strategy to reshape public perception of the company’s growing ambitions around data centers. I remember asking a colleague to explain why this was getting the attention of the global policy and comms boss at the time. He told me we needed to do a campaign to greenwash the data centers as it was going to become a problem in the future.

In 2020, as I was reminded while reading Karen Hao’s excellent book Empire of AI this weekend, one of Jeff Dean’s chief objections to a paper co-authored by then–Google researcher Timnit Gebru was that it misrepresented Google’s energy usage. Gebru was ultimately fired for refusing to take her name off the critical paper, and Dean reportedly spent months insisting to anyone who would listen that the paper had gotten the energy numbers wrong. Which makes it all the more striking to now see Dean front and center in Google’s latest video announcement on energy use—downplaying public concerns, insisting the company was being transparent, and touting Google’s supposed innovations to reduce the climate impact of AI. “Just trust us”. We’ve heard that before.

Fast forward to last week, nearly every tech and business publication in the country carried the claims. No shade to Joshi’s blog but I imagine only a few hundred climate experts will read it. Millions of people saw Google’s headline, spread via the tech media headlines, and in the end many more will repeat the sound bytes the spin doctors at Google came up with (“It only consumes as much energy as watching television for nine seconds!). We’re in the middle of an information war about how much energy AI companies consume, and what the true impact of it all is. Tech corporations have all the data, on the outside all we can do is try to assess the impacts and mount creative efforts to pushback.

Big tech PR people and their bosses are right to be doing their best work on this. As we’ve seen with President Trump’s efforts to dismantle federal agencies that publish information he dislikes, narrative can be just as powerful—if not more so—than reality. Big tech understands this well. They can’t afford to let consumers blame them for rising electricity bills, because the entire business model of the AI boom depends on unchecked enthusiasm. If people start hesitating before asking an AI to draft some text or generate a video, the rosy revenue forecasts could quickly unravel.

What I’m reading:

An opinion piece from former Palantir worker, Juan Sebastian Pinto, in the Guardian raising the alarm about the danger of Palantir’s most recent contracts.

Andreessen Horowitz has funded dinners at an all-female hacker house in San Francisco. Tech billionaires invest heavily in shaping Silicon Valley culture because they know their fortunes depend on it.

“How Google Broke Google” - More Perfect Union broke down how Google’s products have gotten worse over the years.

Via his X account, Paul Graham continues to be one of the only Silicon Valley executives to speak out about what the Israeli government is doing to the Palestinian people. Why is that?

Status published an interview yesterday with Katie Drummond, editor in chief at Wired. She said: “Ultimately, this era in history is a WIRED story: The United States’ authoritarian turn, aided and abetted by the tech industry’s complicity. And that’s how we’re covering it.”