One AI Forecast Moved $5.7B in Power Markets. Read That Again.

I was three tabs deep into Australian inflation tables at 2am when a number made me stop scrolling. Audio, visual, and computing equipment. Up 6.1% in a single quarter.

That's the biggest move in that category since the Australian Bureau of Statistics started the series in 1980. Forty five years of electronics getting cheaper and faster, and one quarter flipped the whole trend.

Anon, let me explain why that print matters way past Sydney.

The $5.7 Billion Forecast

US power markets run capacity auctions. Generators get paid to promise they'll be available when demand spikes. The clearing price is set off a forecast of future load.

One forecast. That's the input. And when that forecast got revised upward on AI data center demand, the auction cleared higher. The extra cost came to $5.7 billion.

Here's the thing most people miss. Capacity auctions are a feedback loop, not a snapshot. A higher load forecast pushes the clearing price up immediately. Ratepayers start covering that number on their next bill. The actual generation capacity shows up three to five years later, if it shows up at all.

So you pay now. You get the megawatts later. And if the AI forecast was too aggressive? The money's already gone.

This is bigger than people realize. We're not talking about a mispriced stock here. We're talking about a single modeling assumption, made by a small group of analysts, that repriced a national input cost by billions. Nobody voted on it. Nobody hedged it.

And that Australian print? That's the same thing showing up on the consumer side. Memory, storage, and compute are getting scarce and expensive at the exact moment everyone needs more of it. The electronics deflation we all built our budgets around wasn't a law of physics. It was a supply chain subsidy. That subsidy is getting clawed back.

Everyone Pays First

Real talk: AI has an inflation front end and a productivity back end, and they don't arrive at the same time.

The front end hits fast. Higher power contracts. Higher hardware costs. Higher prices for anything with a chip in it. The back end arrives slowly, as automation actually reduces cost per unit of work.

That gap is where the pain lives. Asymmetric, ugly, and completely predictable in hindsight.

And it's not just utilities. It's every business that budgeted for falling compute prices. It's every household that assumed a bigger TV would cost less next year than this year. That assumption just broke for the first time in four decades.

Should we be worried that a forecast model has more influence over your electricity bill than your state regulator does? I think the honest answer is yes.

What I'm Watching

Load forecasts are becoming the most important number nobody publishes. Watch the next round of capacity auction results. If clearing prices keep stepping up, the $5.7 billion wasn't a one off. It was the opening bid.

Watch the CPI sub-indices too. Australia went first. If computing equipment inflation shows up in US and European prints, the deflation era is officially over and every model built on it needs rebuilding.

And keep an eye on Bitcoin miners. They're sitting on contracted power, interconnect queues, and land. In a world where a forecast can move billions, real megawatts with real contracts become the scarcest asset in tech.

The chain doesn't lie. Neither do auction results. They're just harder to read.