Volts crossover: Six big energy questions

This is a co-hosted episode rather than an interview, so there is no single framing to test against: each host proposes his own questions and answers the other’s, and they disagree openly. Roberts notes that the five-to-ten-year bracket was the hard part of the exercise, because most of the questions he wanted to ask will not resolve for twenty or thirty years. Kann’s questions show the bracket working. Rather than asking whether data center fever breaks, which he treats as unanswerable, he asks a second-order version: how much large load goes fully off grid. And rather than asking whether industrial electrification succeeds, he asks whether it will look dead within the decade. Both are downstream of one fact, which is that you now cannot reliably get power on the grid where and when you want it, and its price is rising.

The off-grid question turns on a distinction worth holding onto. Bridge power, where a project self-generates for one to five years until its grid connection arrives, is already everywhere and is not what Kann is asking about. His question is how much load goes fully off grid with no intent to connect at all, which has essentially no precedent outside mining. His chain of reasoning: if demand for data center capacity keeps outrunning what the grid can serve, and the money behind it is willing to take risks hyperscalers historically would not, then some share of it takes the one risk that going off grid introduces, which is reliability, in exchange for being unshackled from siting constraints. Roberts’ gut says no, on the grounds that grids are extremely useful and that a gigawatt facility going off grid is building a city’s worth of electrical infrastructure to avoid a queue. He would rather see operators get serious about distributed capacity, which he argues is faster and cheaper than onsite generation if the financial and institutional arrangements can be lined up. Kann’s reply is not a rebuttal but an escalation: grant all of that, squeeze more out of existing lines, aggregate distributed capacity, and short of building an entirely new transmission system on top of the existing one you still hit a ceiling on time to power. Neither will forecast whether demand reaches the high projections, and their refusals differ in temperature rather than in kind. The caution Roberts insists on at the end is the one most likely to get lost: even if data center demand is radically overstated, electrifying transport, heat and industry means much more electricity and a much stronger grid regardless, and he does not want the two questions conflated.

Kann’s electrification question is where the price problem becomes explicit. His long-standing first-order recipe is to clean up electricity, electrify everything that can be electrified, and fill in the remainder, which implies electrifying a large share of heavy industry. That case has always rested on a trade: higher capital cost up front, lower operating cost afterwards, with the saving governed by the spark spread. Roberts adds less waste and easier permitting. But a new aluminum smelter drawing hundreds of megawatts now struggles to find a site, because data centers have taken the sites, and it faces rising power prices while being acutely price-sensitive. Each half of the value proposition erodes at once. Roberts generalizes this into what he calls the political question of the decade, which is whether anyone can bring electricity prices down while continuing to electrify aggressively, and he is blunt that failure here could sink everything else. Kann’s counterweight is alignment rather than optimism about any particular policy: affordability has become the operative word for utilities, data center companies and politicians simultaneously. Roberts adds that the technologies they favor happen to be the cheap ones, so moneyed interests push the same direction for reasons having nothing to do with the environment.

The other three questions are not about power supply at all, and they are where the two disagree most. On self-driving cars, Roberts’ worry is urbanism rather than emissions: easy, pleasant travel makes it easier to live an hour out, which works against density. Kann argues the climate arithmetic may be fine anyway, since autonomous fleets are electric and high utilization pays back the electric vehicle’s capital premium faster, with more vehicle miles weighing against that; Roberts declines to take a side and says nobody knows. On the softwarization of homes and cars, Roberts expects the platform pattern to repeat, with lock-in followed by exploitation, and wants interoperability and privacy rules written before rather than after. Kann, who owns most of the devices under discussion, reports that they are simply better, sees no evidence of the pattern yet, and says he would happily trade advertising for free robotaxi rides. On recycling, Roberts’ thesis is that recovered materials should be treated as a strategic mineral source rather than an environmental nice-to-have, so that buying a Chinese solar panel is also buying a quantity of minerals that become available when it dies. Kann agrees directionally, calls himself a recycling maximalist, and then supplies the binding constraint: while demand is still climbing the S-curve, the volume available to recycle is the volume of demand from a decade earlier, so recycling matters but does not solve a sovereign supply problem. Roberts accepts the mechanism and narrows his own claim to cost rather than volume. The geoengineering question splits them on whether obscurity is protective: both want deliberate, coordinated work, but Roberts fears that high-profile tests would mainly teach the world how cheap and how hard to detect solar radiation management is, and he doubts any enforcement regime could work when nuclear programs, which are far larger and harder to hide, still slip through.