Five big questions about the future of energy

Kann opens by asserting that the power market is undeniably undersupplied and asking when the imbalance could flip the other way. Lubershane will not accept the framing at that strength. He separates two different things: the imbalance being alleviated over the next five years, which he can imagine, and the market actually flipping into oversupply, which he calls practically impossible. He also puts the uncertainty asymmetrically. The huge error bars live on the demand side, because nobody knows how much energy data centers will consume; the supply side has no order-of-magnitude uncertainty at all, since the grid can add some tens of gigawatts through the rest of the decade and cannot add hundreds. That asymmetry is what makes demand the only realistic source of a surprise. Three demand vectors come up: whether the money going into AI is chasing real economic demand, which Kann raises through a Sequoia investor’s running series on the capital spending bogey; whether algorithmic efficiency improves fast enough that the same amount of AI consumes ten or a hundred times less energy, which is what the DeepSeek moment briefly felt like; and whether inference moves off the big centralized facilities onto devices, leaving training as the only centralized load. The one supply-side move big enough to matter is the mirror image of all this: building data centers off the grid in the American Southwest with their own solar, batteries and a little backup gas, then running fiber to them, because new fiber to a sunny region is far easier to imagine than new interregional transmission. Asked for a yes or no on 2030, both men bet no.

On winners and losers, the interesting part is how carefully Lubershane refuses the obvious causal story. His winner is anyone selling basic power system equipment, transformers, switchgear, conductor, turbines and engines, which he concedes is boring; Kann’s is utilities, for the same reason. His loser is electricity consumers generally, and he immediately blocks the inference most people would draw from that. It is not the fault of any individual data center development, and a single well-structured deal with a creditworthy operator paying its own system upgrade costs and then some can genuinely lower rates for everybody else in that territory. What raises prices is the collective pressure of all this growth on every element of the system, arriving at the same time as a core grid that needs upgrading and hardening regardless. Kann adds a second loser, large industrial electricity loads trying to site new plants, and argues siting may hurt them worse than price does: if you need a hundred megawatts, one of the hundred-odd data center real estate developers has almost certainly already tried to buy your site, and your willingness to pay is lower because your process is less profitable than AI. Lubershane’s escape hatch is the same off-grid idea, and he thinks manufacturing suits it better than data centers do because factories are not latency-sensitive. Kann pushes back mildly that data centers are increasingly being sited anywhere too.

Nuclear gets the sharpest answer of the five. Kann’s question is whether a nuclear renaissance would be a Cambrian explosion of new reactor designs or mostly one or two that have already cleared the gauntlet. Lubershane’s answer is that we can already watch it happen: the renaissance is under way globally, if not yet in North America or most of western Europe, and only a few designs are getting traction, with China driving much of it and, unusually, still buying significant technology from a western vendor to do so. The industrial logic rules out proliferation. Getting nuclear down the cost curve requires economies of scale through the whole supply chain and a learning curve that runs from policymakers and regulators all the way down to the crews pouring concrete, and none of that accumulates if the work is split across designs. He says four or five competing designs relevant in any one region would be bad for the industry, and that the ceiling is perhaps one advanced-reactor champion and one small-reactor champion per region. Kann’s supporting analogy is the gas turbine market, where three manufacturers hold more than 70% of the market, and he draws the uncomfortable conclusion that the startup reactor companies face a bloodbath with a few survivors. Lubershane frames the next five to ten years as the window in which any of them must cement a position.

The last two questions rhyme, and together they are where the episode’s real disposition shows. On geothermal, Kann asks how much it would mean if the leading enhanced-geothermal developer succeeds at its first commercial project. Lubershane says he has complicated feelings and gives both halves. Yes, it is a flag planted and a starting gun, and fast followers can draw on the same oil and gas fracking and horizontal drilling supply chain and the same service providers. But he thinks geothermal is inherently slower to roll out than solar was, because drilling and exploration risk do not go away and the supply chain is more complex, where utility-scale solar around 2008 or 2009 only needed open land near an interconnection point. So the marquee moment will be followed by a few disappointing years, with the real take-off in the 2030s rather than the late 2020s. He adds a wrinkle: the oil and gas workforce is both geothermal’s greatest enabler and its main competitor for talent, so a rich oil and gas market makes drillers harder to recruit into a riskier, slower-paying geothermal field. Grid-enhancing technologies get the same verdict from a different direction. Advanced conductors and dynamic line ratings both look like no-brainers on paper, and Lubershane gives three reasons they have moved slowly anyway: until roughly three years ago there was no demand pressure to force the issue; utilities are conservative about high-voltage assets for reasons he thinks are good, given the safety consequences; and the pitch understates the complexity, since raising one line’s throughput also means upgrading the substations at both ends and studying knock-on effects across an integrated system. Kann’s reading is that this sector never gets a galvanizing moment but that momentum, once started, runs for decades. Lubershane agrees and sets the expectation precisely: an inflection here means one to two, then two to three, then maybe three to five, reaching ten times current deployment over ten to twenty years.