Our Sector Is Getting AI and Energy Wrong
Twenty years ago, an oil company had a problem: people were beginning to point fingers at it. So BP hired one of the biggest ad agencies in the world, and together they made a household phrase out of an obscure bit of jargon — the "carbon footprint." They built the first footprint calculator. They ran the ads inviting each of us to measure our own little contribution to the crisis. It worked beautifully. For a crucial decade, a movement that should have been kickstarting the green energy revolution was staring at our grocery bags instead.
I've been watching our sector talk about AI and the environment this year, and I keep seeing something that genuinely worries me. We're running that campaign on ourselves. The posts tallying the water cost of a chatbot question. The guilt over the electricity used when drafting with AI. The quiet and loud abstentions on principle. Nobody had to pay an agency this time. We're taking our eyes off of the real levers of change, for free.
Let me say plainly where I stand, so nothing that follows gets misread. I believe climate change is one of the defining emergencies of our lifetimes. I think we're already seeing devastation and we're on a path toward catastrophe. The people in our sector worried about what AI is doing to the planet are directionally correct.
And I run an AI company that serves this sector — so discount me accordingly. But notice what I'm asking for, because it isn't "relax." BP wanted our attention on our own footprints so that we wouldn't see efforts they were really afraid of. I'm asking us to take back our attention, and aim it in the right direction.
Carbon footprints were never the point. The point was to manipulate our feelings so that after a year of dutifully sorting your recycling, the first time you see a photo of the Great Pacific Garbage Patch, you let out a deep sigh and say to yourself, What's the point? Nothing I do matters.
Convince people that a planetary crisis is theirs to carry personally, and they will carry it straight to resignation. A movement of individuals privately auditing themselves toward hopelessness is no threat to anyone.
And that's what I feel we're rerunning now. Where does the AI-guilt conversation actually land? It brings us to I shouldn't use these tools, and then, quietly, to I can't affect any of this anyway. We end up feeling helpless about AI and helpless about the climate at the same time — which is exactly where the original BP play was designed to leave us. We've seen this before.
The climate movement learned this lesson painfully and in public. Andy Masley — a former physics teacher whose essay on AI and the environment is the one I'd hand anyone in our sector — puts it best:
"One of the most important shifts in talking about climate has been the collective realization that individual actions like recycling pale in comparison to the urgent need to transition the energy sector to clean energy. The current AI debate feels like we've forgotten that lesson. After years of progress in addressing systemic issues over personal lifestyle changes, it's as if everyone suddenly started obsessing over whether the digital clocks in our bedrooms use too much energy and began condemning them as a major problem."
His whole essay is worth twenty minutes of your time — he's an independent writer on a philanthropic grant, not an industry employee, and his figures have since been corroborated by measured disclosures and by the climate-data researcher Hannah Ritchie. I'll steal just three of his findings and send you to him for the rest.
For example, Masley finds the full cost of one AI prompt — training, chips, and all — is about 0.28 grams of CO2. That's equivalent to driving a sedan four feet at constant speed. Running a space heater for under a second. Printing a fifth of one page of a paperback book. To match the energy you used on one transatlantic flight, you'd need to ask a chatbot 400 questions a day for 80 years.
To match one transatlantic flight: ask a chatbot 400 questions a day, every day, for 80 years.
Full lifecycle per prompt — electricity, model training, and chip manufacturing — per Andy Masley, corroborated by Hannah Ritchie and Google’s measured figures.
My favorite of his comparisons isn't about AI at all. A digital clock uses a million times as much power as an analog watch. A million times! Nobody cares and nobody should if your concern is how to conserve energy.
"ChatGPT (the digital clock) uses ten times the energy of a Google search (the analog clock)." Both true and both meaningless without the baseline of how much of our total energy consumption Google searches and AI chats comprise. Even at scale, with billions of people using these systems, the sum is still marginal in terms of gross energy consumption when compared to other common activities in our lives. It's just not the priority, not even close.
So is it all fine, then? No. And this is where I need the concerned half of our sector to stay with me. The buildout is real and it is enormous. Data centers already draw about 1.5 percent of the world's electricity, and the International Energy Agency projects that roughly doubling by 2030, with AI the fastest-growing slice. Google's emissions are up 51 percent since 2019. Microsoft's rose 25 percent in a single year. The journalists at MIT Technology Review who did the most careful independent math put the trajectory in one line: our AI footprint today is "the smallest it will ever be."
And the local harms are not hypothetical. In one Georgia county, a data center quietly drew 29 million gallons through an unmetered line, during a drought, while residents sat under watering bans. It got a back-bill and no fine. Deals like that one are increasingly negotiated under NDAs — in Virginia, the data-center capital of the world, four out of five localities with facilities have signed one. Two-thirds of the data centers built since 2022 sit in water-stressed basins, because getting to power fast beats water prudence every single time. And the biggest neighbor-borne harm has no address at all: data centers drove almost two-thirds of one grid operator's capacity-price spike last year — 9.3 billion dollars, spread across the electric bills of thirteen states.
But look at the shape of every one of those harms. No chatbot ever drained a well. A siting deal did. A water permit did. A utility contract did. There was a public process. Elected officials invited the data center, for the tax base — the revenue that balances the budget and pays for the services their voters want. The public got its hearing, voiced its concerns, and discovered that a hearing is not leverage. The developer and the officials were at the table. The community is in the audience.
So here's what I'm actually advocating for: nonprofit expertise and support that helps communities balance the economic and public-health concerns of hosting a data center. That looks like policy support — objective criteria for where these facilities should go, environmental remediation and community-health commitments written into the deals, companies paying for the grid infrastructure they use. It looks like organizing for more public voice than a three-minute comment window. And it looks like regular old political organizing — electing local candidates who actually reflect the community's priorities. If a town weighs all of it and really doesn't want a data center, and is willing to forego the economic benefits, that should be its choice.
The scale of the mismatch is the point. Remember the Amazon HQ2 auction — hundreds of municipalities tripping over themselves for a company that had already decided? That's the buildout today, county by county, wrapped in NDAs, with subsidies in the biggest deals averaging about two million dollars per permanent job. Communities are negotiating alone against the most resourced companies on Earth. So pool the lawyers, the hydrogeologists, the rate experts who can sit on the community's side of the table. Share the record of what other counties actually got — the one thing NDAs are designed to prevent. There are toolkits, and there are moratorium campaigns. Almost nobody is building the negotiating infrastructure in between. It is the most nonprofit-shaped gap I have ever seen.
Most times when I get asked about AI energy use, I ask if anyone in the audience knows how much electricity AI uses? So far I have not had anyone raise their hand to answer. The reason I press so hard on the math is because credibility is our sector's entire currency, and guilt theater spends it. When our loudest claims don't hold up to scrutiny, the public and policymakers we most need to move learn to discount everything else we say.
None of this suspends personal responsibility. This is not a license to be wasteful. We need to understand the flows of energy that power our lives — including AI. We are mostly highly principled and well-meaning people. What we need is knowledge of this era-defining technology, and dialogue, not about whether or not to use it, but how and when.
What about truly principled abstention? I used to work at an environmental nonprofit. We were good people doing real work — and we commuted to a climate-controlled office tower, drove all over the state shooting photos for social media, and went through paper by the case. Is that hypocrisy?
No. That's what working inside an industrial civilization looks like. But it means abstention is extraordinarily difficult to achieve. 40 percent of our waking lives already runs through data centers. To truly abstain from AI and work that flows through data centers, you'd need to be off the grid, closed loop, and analog. I will never discourage folks from trying to do this, but let's be clear about what abstention actually means, when it makes a statement, and what it actually takes.
What do I think should happen? I refer folks to a conversation between Ezra Klein and Bill McKibben where they highlight efforts to green the grid, electrify transportation, improve battery tech, build more cleanly, and fix food systems. All this stuff is mundane, well-understood — and working. Last year, for the first time, all of the net new electricity demand on Earth — the AI boom included — was met by clean generation, three quarters of it by solar alone.
The climate movement knows what to do. "Limit your AI use" as a means of making an impact on climate is a distraction from the list of real priorities. A well-understood climate agenda needs our continued focused effort.
So let me summarize my perspective. The concern is righteous, but misplaced and risks our credibility. Personal AI use is a rounding error on the energy of ordinary life, and personal guilt about it is self-defeating. We have upstream levers to pull — siting deals, water permits, utility contracts — and the communities negotiating at those tables deserve our sector's expertise and support. And the climate fight itself lives many levels up from our chatbots — in the grid, the buildout, the global energy transition — where the movement needs our support.
And use AI to fight! Chat with AI all day if it helps you organize your community. Use it to build the presentation for the town-hall meeting. Use it to draft your comments on the siting ordinance so they are air tight. Use it to organize the volunteers, to summarize the meeting notes, to cut the social videos that grow the movement. If you feel the pang of guilt, run Masley's math — the prompts behind your entire campaign cost less energy than one committee member's drive to the hearing. Most people in this sector can do orders of magnitude more good with AI than they can do harm with it.
The oil industry needed one of the biggest ad campaigns in history to make us audit ourselves instead of them. For some reason we're doing it to ourselves now with AI. If we avoid climate catastrophe, it won't be because we counted our prompts. It'll be because we rebuilt the global energy system.
