The clean energy transition has a hardware problem

The premise is arithmetic. To have a sustainable energy economy, which will be largely electric, Baglino says the world needs to triple or maybe even quintuple electricity generation and consumption, and his point is about who has to grow to deliver it. Renewable developers are demonstrably good at deploying solar and batteries, and the load side can produce heat pumps and millions of electric vehicles a year. What is not obvious to him is that the electricity sector itself, the utilities and the supply base serving them, can grow by 500%. He says he found the specific version of that problem by accident, meeting founders at venture events working on carbon sequestration, clean hydrogen, next-generation solar, long-duration batteries, fusion and electric steel. Nearly all of them said the core process was figured out and there were no power electronics to buy. Two distinct failures sat behind that. The gear was sold out, and a startup ordering one or two units went to the back of a line where even utilities were struggling. And what they needed was a whole conversion solution rather than a catalog part; going to a scale supplier of solar inverters and asking for a custom build for a clean steel operation could, he says, be really painful a lot of the time.

The obvious objection is that other parts of the stack did ramp, and Johnson raises it with solar panels. Baglino’s answer is that the transformer was never given a reason to. Solar was heavily subsidized in Europe, China and intermittently the United States, and what got subsidized was a genuinely high-technology process: producing polycrystalline silicon and then etching and doping it, which carries specialty capital expenditure. A legacy medium-voltage transformer has no material science in it and is not a high-technology production process, so it was not juiced in the same way and politicians never intervened in that market. He offers that and two further reasons as possibilities rather than findings: repeated federal efforts to raise transformer efficiency standards, which he thinks left investors facing regulatory uncertainty, and the chance that some incumbents saw power electronics eventually climbing to medium voltage and disrupting the product. The scale of what then landed on that under-invested base is the crux. In the mid-2000s there was no utility-scale solar, no utility-scale batteries and under a gigawatt a year of data center development, and medium-voltage transformers served utility distribution at about 1% annual growth. Over the last 10 to 15 years almost a terawatt of transformer demand has appeared globally, and last year alone almost 600 gigawatts of utility-scale solar was deployed, which he sets against a total US peak power of a terawatt. He never scopes the 600 gigawatts and it sits inside that global frame, so the two sides of the comparison are not stated on the same basis. He is careful about what kind of problem this is: not a technology problem but a capacity one. It is hard to ramp something up that quickly.

What Heron sells is power conversion at the megawatt scale and above, and he calls the case for it a combination of factors: efficiency is one, reliability and serviceability another. The argument turns on the second. The incumbent product is monolithic. A medium-voltage transformer weighs multiple tons, needs a crane, and costs weeks of downtime if something goes wrong, which is why data centers overbuild: if four or five will do the job you install six or seven, so a failure does not cost you the load. A power electronics equivalent is modular and fail-operational, so one failed block leaves the rest running, and swapping it could be a 10-to-15-minute job rather than a two-to-three-week one. It is not literally hot-swappable at medium voltage, he says, but it is as fast as if it were. That lets a site drop the overbuild or accept a few minutes at slightly reduced power.

The argument’s real load sits in what it would take to make that real, about which he is unusually plain. The objective is to be the first solid-state transformer solution reliable at mass production scale, and he stresses both words. Successive engineering revisions raise complexity toward the full-power five-megawatt product and maturity at the same time; his framing is that you design for the known knowns, provision for the known unknowns, and only find the unknown unknowns by testing, which they have already hit and expect to hit again. Manufacturing steps up from hand-built prototypes to semi-automatic tooling at the Scotts Valley headquarters to full automation at a factory site still to be announced, and customers are chosen the same way, a few partners who trade project pipeline for product maturity rather than a broad sales push. Why this component justifies all that is a leverage argument: on its current trajectory the medium-voltage transformer alone is on track to consume almost half of all grain-oriented electrical steel in the world, so deleting it frees that supply base for pole-mounted transformers and other electrification needs. On policy he does two things. He asks whether electrical infrastructure should be treated as critical infrastructure, drawing the analogy to the debate over Huawei telecom equipment, and he says plainly that some incentives are needed to drive the behavior, because it has run the opposite way for decades. On the critical-infrastructure question his own answer splits: a passive steel-and-oil transformer carries little hazard, but the more intelligence and software go into the grid, the more where it is made and who owns it matters.