Everyone's celebrating AI breakthroughs while ignoring the infrastructure cost. Training and running these models requires energy on a scale most people don't appreciate. The environmental bill for the AI revolution is coming due, and it's enormous.

Major tech companies that were making real progress on clean energy commitments are now backsliding. Google, Microsoft, Amazon—all the companies that made big climate pledges a few years ago—are watching their emissions climb because AI data centers are power hungry.

Let's put some numbers on this. Training a large language model can emit as much carbon as five cars over their entire lifetimes. That's just training—before anyone actually uses the thing. And we're not training one model. We're in an AI arms race where every company is training multiple models, running constant experiments, and scaling inference to serve millions of users.

The promise was that AI would help us solve climate change. Optimize energy grids, improve efficiency, model complex systems. And maybe it will! But right now, the energy cost of the AI boom is undoing years of progress on renewable energy commitments.

Here's the uncomfortable truth: Silicon Valley wants AI progress more than it wants climate progress. When those goals conflict—and right now they do—AI wins. Companies are signing deals with natural gas plants to ensure reliable power for data centers. They're pushing back renewable energy timelines. They're finding creative ways to account for their emissions that make the numbers look better than they are.

This isn't unsolvable. We can power data centers with renewables. We can make models more efficient. We can be strategic about what actually needs to run on massive GPU clusters versus what can run locally or on smaller models. But those solutions require choosing efficiency over capability, which runs counter to the current race for bigger, more powerful models.

The AI industry likes to say we're in the early days, that we'll optimize later. But later is too late for climate commitments that were already ambitious. Every ton of carbon we emit now is a ton we can't emit later if we want to stay within safe warming limits. The marginal improvements AI might enable for climate modeling don't offset the concrete emissions happening right now.

I'm not anti-AI. I'm a former founder who built an ML product. But we need to be honest about the tradeoffs. If we're going to power the AI revolution by reversing climate progress, we should at least admit that's what we're doing instead of pretending we can have both without sacrifice.