My daughter Anaya was nine when she asked me, over breakfast, whether the planet was going to be okay. She'd done a project at school about polar bears and she'd clearly been chewing on it for days. I gave her the answer parents give — that lots of clever people are working on it, that we'll be fine, eat your toast.
Then I went to work, where I spend my days around people building AI systems, and I thought: am I actually sure that's true? Are the things we're building helping her future, or slowly making the air she'll breathe a little worse?
So I went and looked. Not at the marketing. At the actual ledger. And the thing I want to tell Anaya, when she's old enough for the longer version, is that the answer is "it depends entirely on what the grown-ups choose to build" — which is both more honest and more frightening than the toast answer.
The part the brochures skip
Let's get the uncomfortable bit out first, because the people selling "green AI" tend to mumble through it. Training a large model burns through a serious amount of electricity. Running it — every time millions of people ask it something, every day — burns through more. And most of the world's data centres are still plugged into grids that run on gas and coal.
That's a real cost, and it lands on the same atmosphere Anaya's polar bears are stuck in. Anyone who tells you AI is "carbon-negative" today is, I'm afraid, selling you something. I've sat in those pitches. The slide is always very confident and the assumptions underneath it are always doing a lot of work.
Where it genuinely earns its keep
And yet. When I dug into where AI is actually pointed at the climate problem, some of it made me genuinely hopeful — for her, not for the quarterly deck.
The one that surprised me most was the electricity grid itself. Once you add solar panels on roofs, wind that comes and goes, and electric cars feeding power back in, running a grid becomes fiendishly complicated. Predicting demand and supply well enough to lean harder on renewables, without the lights flickering, is unglamorous work that almost nobody outside the industry thinks about. It might also be the single most useful thing on this list.
Then there's the heavy industry nobody photographs — the factories and plants where shaving even a small slice off energy use saves more carbon than a whole town of us remembering to switch the lights off. That's where the near-term wins really live, not in the apps on our phones.
There's the search for better materials — battery chemistries, more efficient solar cells, cheaper ways to make clean hydrogen. The number of possible combinations is so vast that testing them one by one would take lifetimes; letting a model narrow the field down speeds up work that Anaya's generation will need finished, not started.
And there's the watching. Satellites paired with detection models can spot illegal logging almost as it happens, or trace a plume of methane back to the exact facility leaking it. The clever part isn't the model. It's that someone with the power to do something finally has the evidence.
How a kid's future gets traded away in the small print
Here's what I've learned to listen for, because this is where good intentions go to die.
People love to talk about training a model because it's a big visible number. Almost nobody talks about running it a billion times a day, because that cost is spread thin and easy to ignore — right up until it isn't. If you're not measuring it, every reassuring claim about it is a guess.
Then there's the trick where a fossil fuel company runs a sweet little "AI for carbon capture" pilot for the press while expanding drilling everywhere else. Judge the whole company, not the photogenic pilot. Anaya's school project would have seen through that, frankly.
And the slipperiest one: "more efficient per query" and "less energy overall" are not the same sentence, no matter how often they get said in the same breath. A factory that gets more efficient often just makes more stuff. Efficiency is not the same as actually using less.
What I'll tell her when she's older
The bits we can control are mostly choices, and they're boring, and boring is fine. Power the things with the cleanest electricity you can find, and keep pushing the providers who won't say where theirs comes from. Don't reach for the biggest, hungriest model when a smaller one does the job — the emissions roughly follow the cost. Measure your carbon like you measure your spending, as a real number on a real line, not a vibe. And before you build anything, be honest about whether it actually helps the planet, does nothing, or makes it worse. Plenty of AI sits in that middle, neutral bucket and gets sold as salvation anyway.
So, Anaya — when you're ready for the long version. The planet's fate isn't being decided by the technology. It's being decided by whether the people building the technology are willing to look at the unflattering number and choose against the comfortable story. Some of them are. I work to make sure more of them do. That's the most honest answer I've got, and you deserved an honest one all along.
