Is the future of low-carbon industry off grid?

The post is built on a white paper by people at Scale Microgrids, Paces AI and Stripe, which asks whether off-grid solar microgrids in the US could be big, fast and cheap enough to be a compelling near-term alternative to new gas plants for AI demand. Its answer is probably yes, and his own gloss is that the costs it reports are close enough to grid power that a hyperscaler with a carbon commitment would not blink at them. He says he was not surprised by that, since the cost assumptions are reasonable and going off-grid avoids both the interconnection charge and the decade it takes to build any transmission line of consequence. What surprised him was the scale. Screening for practical siting rather than raw sunlight, the study still found more viable Southwest capacity for mostly-solar data centers than the whole United States grid peaks at.

He then supplies the caveat that turns the post. Clustering that much computing that far from population centers is almost certainly unrealistic, because training a model is not sensitive to communications delay but using one is, and he suspects that is a large part of why the strategy is not already happening. The paper asks itself the same question. That caveat is exactly what does not apply to a factory. From an energy point of view a data center is a big box drawing a steady load, and so is an electrified manufacturing plant; the difference is that the distance goods travel often matters less than the distance data travels. With automation pushing labor down as a share of production cost, he argues, energy is what is left to decide where industry sits, and industry has always gravitated toward cheap primary energy.

Two further things make the industrial case cheaper rather than merely equivalent. The white paper priced solar and batteries at what they cost today, and he expects further cost reduction to outweigh the escalation risk he sees coming from trade conflict with China, pointing to multi-day iron-air storage and to storing energy as high-temperature heat as the routes to running a system at 90% solar or better. And most industrial facilities do not need a data center’s uptime: for some of them 90 to 95% firm supply may be good enough, so long as the interruptions are not unexpected, which means many can probably get away with less gas capacity and a lower total cost. That is how he gets to an off-grid cost of energy well below today’s, and to a conclusion about the country rather than the technology. America already has the gas and the pipelines to serve as the backstop fuel, and it is the solar and wind resource on top that could make it a clean manufacturing superpower.