Fuel cells, geopolitics, and the solar growth curve

The answer with the most reasoning in it is the first. Kann expected community opposition to data centers and did not expect it to arrive this fast or to be this bipartisan, with versions of a pause or moratorium now in both New York and Texas and, as he reads it, President Trump as the notable voice saying not to worry. His prior was that opposition would come mostly from the left, and part of it does, most sharply from Bernie Sanders, who responded to Dario Amodei’s public argument for pacing AI development by saying pacing is not enough and it should stop. But the right did not line up the way a crypto-friendly, libertarian-leaning coalition would suggest, which leads Kann to conclude that opposition is not really a partisan issue at all. It is a local one, sitting on top of a more general anxiety about AI. Asked whether all this slows the buildout, he says yes relative to a no-guardrails counterfactual, and immediately bounds it: Texas will still build a great deal, delays of six months to a year are realistic and five years is not, and the constraint on frontier AI, if one comes, will be something like a regulatory body rather than an inability to build data centers.

The same discipline about attribution runs through his answer on the war in Iran. He sees no good evidence of a large direct effect on electric vehicles. The correlation between fuel prices and adoption is weak to begin with, and the price move was smaller than expected: talk of $200 oil at the outset, roughly $100 at the time of recording. The sales data points elsewhere anyway, with the US down 21% because the tax credits went away, Europe up 29% in line with its existing trend, and a doubling in the rest of the world. Whatever the war did in the US was swamped by the incentive change; whatever it did elsewhere at most boosted a trend already underway. The effect he thinks is genuinely underreported is slower and not about prices at all. Between the Russia-Ukraine war and this one, Europe has become serious about sovereignty, visible so far mostly in defense spending but working its way into energy strategy, with a durable tailwind for anything that can be produced domestically or regionally rather than imported.

The forward-looking answer he has actually done arithmetic on is the one worth watching. The question is whether a data center and a utility will announce a tariff that substantially subsidizes everyone else’s electricity prices, and he says the math works, but only under a specific combination. Rates are set across a whole territory, so a large utility cannot direct the benefit at the county hosting the project; the subsidy gets diluted across too many customers to notice. Shrink the utility and it concentrates. His illustration is a roughly three-gigawatt peak system taking on another gigawatt of load, where the revenue and margins expected from a gigascale data center could fund a very large subsidy, possibly approaching the Ohio proposal to take everyone’s prices to zero, though he does not expect it to go that far and names 20% or 50% without committing. The catch is that the same smallness makes the case rare: you need a site, and you need to energize a gigawatt-scale load on a three-gigawatt system quickly enough for the subsidy to matter, which he calls a unicorn set of circumstances while noting that things of that order are already happening.

On solar he is making a forecasting claim rather than a physical one. A recent industry forecast reads this year’s strong build as partly a rush to beat the tax credit expiration and projects flat volumes for the rest of the decade; he is at high confidence it rises instead, and expects record volumes each year for five years. His reasons are that generation is needed more urgently than capacity, that solar is fast and deployable, that domestic manufacturing is expanding, and that its cost pressures are ordinary ones shared with everything else rather than the specific supply chain congestion afflicting gas. Then he undercuts his own contrarianism: forecast notwithstanding, he thinks the market already understands this. The related segment on commercial and industrial solar shows the limit of the optimism. That segment is stuck between two business models, unable to standardize the way residential can and unable to spread transaction costs over a large project the way utility-scale can, and reviving it requires both better relative economics, a premise he audibly doubts, and a radical cut in soft costs he thinks would have to come from automation. His closing hot take applies the same patience to the grid: utilities move on multi-year regulatory planning cycles, those cycles only started turning a couple of years ago, and once the momentum exists the system can build a great deal more than the “grid can’t keep up” narrative allows.