AI's Energy Appetite Is Bigger Than Anyone Expected

According to KKR's Raj Agrawal, the market is making a massive mistake: it's drastically underestimating how much power AI will actually need in the coming years. And this isn't small change. If machine learning data centers start consuming electricity like crazy servers gone rogue, costs explode, development timelines stretch, and the whole digital disruption party crashes against a much more earthly reality: electricity bills.

The problem isn't new, but the timing is rough (or perfect, depending which side you're on). AI is growing at a bonkers pace—new models every week, ruthless competition between tech giants, everyone trying to build the next ChatGPT. Each of these experiments guzzles massive amounts of power. But the official numbers floating around? Probably way too optimistic.

What does this mean in practice? Companies could hit a wall way sooner than they think. Not the "killer AI rebels" wall (that's fiction), but the boring, concrete one: there isn't enough energy infrastructure. Data centers are already battling fiercely for access to renewable energy, and if AI accelerates further, it could become a serious bottleneck.

Agrawal's basically saying: "Folks, you calculated everything except the most important fixed cost." It's like buying a sports car without checking if your garage has a spot. Cool car, but where does it go?

Here's the actual plot twist? This might create openings for anyone who's already invested in renewable energy or efficient infrastructure. Tech companies that did their homework could find themselves with a massive competitive advantage. The rest? Less cheerful.