The Vendors, the Maturity Curve, and Honest Limits

Having made the case for intelligent conversion, an expert has to be equally honest about where it actually stands, because the gap between a compelling concept and a bankable product is where a lot of infrastructure money gets lost. This lesson maps the vendor landscape and, more importantly, the maturity curve, so you can tell marketing from reality.

Start with who is building this. ABB and its power-grid business, now Hitachi Energy, have been among the most persistent developers of solid-state and advanced conversion technology, with real deployments in traction and grid applications going back years. GE Vernova, the energy-focused company spun out of General Electric, carries deep grid and power-electronics engineering and is active across the conversion and grid-integration space. Eaton is a major force in electrical distribution, switchgear, and increasingly in the power architecture of data centers specifically, where it integrates conversion, distribution, and controls. These are serious industrial firms with the balance sheets and field experience to carry a technology from prototype to product, which matters enormously, because a transformer node is not something you want from a startup that may not exist in a decade.

Now the honest part: the solid-state transformer is not yet a drop-in replacement for the large power transformer, and it will not be for some time. Three limits explain why. The first is cost. A solid-state transformer contains a large number of expensive semiconductor devices, gate drivers, controls, and cooling, against which the conventional transformer's copper and iron look almost free. For bulk voltage conversion where you do not need control, the solid-state device simply cannot compete on price today, and pretending otherwise is how projects go over budget.

The second limit is efficiency, and this one is counterintuitive. The conventional transformer, doing nothing but electromagnetic conversion, achieves efficiency above ninety-nine percent. Every power-electronic stage in a solid-state transformer introduces switching and conduction losses, and stacking several stages means the total efficiency, while good, has historically struggled to match the passive iron device at full load. When you are moving hundreds of megawatts, a fraction of a percent of additional loss is a large amount of waste heat and a real operating cost. The technology is improving as silicon carbide matures, but the idea that solid-state conversion is automatically more efficient is a misconception; its advantage is control, not raw conversion efficiency.

The third limit is reliability and maturity. The conventional transformer's greatest virtue is that it almost never fails and has a track record measured in decades across millions of units. A solid-state transformer is a complex assembly of many components, each of which can fail, and the field history is thin by comparison. For a data center whose entire value proposition is uptime, deploying an immature conversion technology at a critical node is a risk that has to be weighed carefully. The rational pattern is to adopt solid-state conversion first where its unique capabilities are genuinely needed and where redundancy contains the failure risk, rather than ripping out proven iron at the main intake.

So where does advanced conversion actually earn its place in an AI facility today? The strongest case is at the interface between DC assets and the campus: where large battery storage, and increasingly on-site solar, connect to the power architecture, active bidirectional conversion with a native DC port avoids conversion stages and gives precise control. Another strong case is inside the emerging DC power distribution architectures for high-density racks, where the industry is moving toward higher-voltage DC distribution to feed power-hungry AI accelerators more efficiently, and where controllable conversion is inherently part of the design. A third is anywhere power quality and fast control genuinely matter, such as tightly regulating voltage for sensitive equipment or providing grid-support functions. In these niches, the added cost buys something the iron transformer cannot offer at all, so the comparison is not cost-versus-cost but capability-versus-nothing.

There is a broader strategic point about the whole power-conversion chain in AI infrastructure. The industry is rethinking how power moves from the grid to the chip, because the traditional chain, with its many conversion steps from medium-voltage AC down through several stages to low-voltage DC at the processor, wastes energy at each step and consumes space. The move toward higher-voltage DC distribution within the facility, championed by chip and system vendors as rack power densities climb past a hundred kilowatts and beyond, is pulling advanced power electronics deeper into the data center. Solid-state conversion is one expression of a larger trend: as power becomes the binding constraint, the efficiency and controllability of every conversion stage between the substation and the silicon becomes worth engineering hard, where a decade ago it was an afterthought.

Faraday's read: be a discriminating buyer of the solid-state transformer story. It is a genuine and important technology, backed by serious vendors, solving real problems of DC integration and active control that the iron transformer cannot address. But it is not yet cheaper, not yet unambiguously more efficient at bulk conversion, and not yet as proven, and anyone who tells you it is ready to replace the main transformer wholesale is selling. The disciplined position is to deploy advanced conversion precisely where its unique capabilities are required, to keep proven iron where dumb bulk conversion is all you need, and to watch the silicon carbide maturity curve closely, because the point at which the economics cross over is coming, and the operators who have learned the technology in its niches will be ready to scale it when it does. Power is the first-order constraint, and the conversion chain is one of the places that constraint is quietly being loosened, one silicon carbide device at a time.