Enter the electric supercycle

The starting conditions are bad and he does not soften them. Anything ordered today takes three or more years to arrive, gas turbines more like five, and essentially every common building block of the grid, turbines, transformers, aluminum and copper conductor, switchgear, costs two to three times what it did five years ago. The expectation after the pandemic was that these crunches would ease the way other global supply chains did; they did not, because demand kept surging, so prices have stayed well above general inflation. Worse, the political consequences have not arrived yet. Averaged across the United States, retail electricity prices over the past five years have risen roughly in line with inflation, with real heterogeneity by region and rate class, so in real terms customers have not yet felt this. Equipment prices are a leading indicator, and he says with a very high degree of confidence that retail prices will now rise faster than inflation and, because electricity is a foundational input, become a driver of inflation themselves. He calls this frightening and says the affordability debate is only getting started.

All of which sounds like a case against optimism, and the pivot is not that the crunch resolves. It is that the demand growth underneath it is not really an AI story. Lubershane inverts the usual ordering of certainty here, and this is the most important move in the episode. Data centers are by far the largest source of new demand and he thinks they rightly get the attention, but in his opinion they are the least certain form of demand once you look five, ten or fifteen years out. What he is confident in is the slower, older trend: electrification of transport and heating, and the fact that advanced manufacturing is already electrically driven because electricity is the form of energy that delivers precision. Kann supplies the headroom figure, that electricity is only about 20% of final energy demand in western countries, leaving 80% still to take. And Lubershane offers a corrective datapoint: globally in 2025, electric vehicles added almost exactly as much electricity demand growth as data centers did. He does not expect that to repeat in 2026 or 2027, and offers it to show that a US-centric view anchors too hard on data centers. In the near term he agrees AI is a problem for everything else, crowding out sites where you might have put a hundred-megawatt industrial process, even as the money it attracts upgrades infrastructure that will eventually serve the rest.

The actual thesis is about supply chains rather than demand. Four building blocks compound on each other: solar photovoltaics, lithium-ion batteries, electric vehicles and their motors, and power electronics, which he calls the least visible leg of the stool and the connective tissue between the others. Deploying more solar raises the value of storage, which creates a market for grid-scale batteries. Electric vehicles are the keystone species because unit volumes are so large that they build mass manufacturing capacity for everything else. And the loop closes rather than running one way: electric vehicles themselves benefited from early solar-industry investment in wide-bandgap semiconductors, the materials that allow higher voltage and higher frequency switching, and that technology, scaled up by vehicles, is now being brought back to the grid for solar inverters and transformer replacements. Microgrid controls follow the same path, developed over a decade at hospitals and campuses, taken to gigawatt scale by data centers, and plausibly reused at electric vehicle charging depots of ten or twenty megawatts with batteries and perhaps co-located solar. Robots extend it again, since anything autonomous runs on batteries, on power electronics converting battery direct current to drive motors, and on high power density motors, and drones already manufactured in the hundreds of millions could pull battery energy density forward because defense buyers pay for performance in a way vehicle buyers do not. That mutual reinforcement, not the current crunch, is what he means by a positive supercycle.

He is careful about how much weight that carries, and two limits are his own. The first is size. He sized the robotics load himself, called it a back-of-envelope exercise done for fun, and found that a household humanoid run five or six hours a day would consume three to four times a refrigerator’s annual electricity, making it the largest appliance in a home apart from heating and cooling but only about a fifth of what a typical electric vehicle driver uses. He adds that he is not especially bullish on humanoids as a form factor anyway, and thinks the larger load sits in industrial robotics combined with industrial electrification, where a single large chip fabrication plant might draw tens of megawatts up to around a hundred. Nothing on this list, he says, compares to data centers, which are so big they block out the sun. The second limit is the one the episode’s title does not prepare you for. Asked to pick the rate limiter, he answers the grid, and then narrows it to electric transmission, because he has not seen a technology that changes the amount of investment new lines and corridors require. The binding constraint on those is societal willingness to tolerate large infrastructure, and he says plainly that no technology solves that. His only offered route around it is moving large loads off-grid, powered by a hybrid of solar, gas and batteries rather than off-grid gas alone, and he frames that as something he hopes for and as the option he sees no alternative to if the data center boom runs well past five years.