Forecasting the Load

Weather forecasting in this business used to be a supply-side instrument. You forecast clouds because they hit solar output, you forecast wind because it hits turbines. Today a company called Climavision is selling forecasts to the other side of the meter: data centers, so they can “respond to changing weather conditions” and “play a more active role in grid management.” Same tool, flipped. The load itself is now the unpredictable thing, the one that needs a forecast and might surprise you on a Tuesday afternoon.

It fits a pattern I’ve been watching build for weeks — Virginia data centers dropping gigawatts of demand in seconds, MISO trying to legally define “computational load” so it holds still, PJM warning about a 2030 reliability crisis if nobody solves this. What’s new is the market answer: if data centers are the volatile variable now, sell them the same instrument you’d sell a wind farm. Twenty years spent building tools to predict nature’s mood swings, now pointed at the mood swings of GPUs.

Meanwhile the House passed a bill to stop utilities from shifting data center costs onto regular ratepayers — a real idea that, per the analysts quoted, is unlikely to clear the Senate before midterms and would mostly just formalize what states are already doing through large-load tariffs. Congress arriving to ratify a trend after the trend already happened, again.

And underneath it, Meta pushed its solar portfolio with Apex to 1.2 GW across five states, and Burlington — the discount retailer, not the city — is quietly bolting solar onto its warehouses because it’s cheaper. The federal mood on clean energy has curdled all month. The buildout doesn’t seem to have gotten the memo.