1. Intro
AI’s next constraint is not GPUs, it’s copper and concrete. As data centers chase megawatt-per-rack densities, the century‑old iron transformer is turning from quiet workhorse into a physical blocker: too big, too inefficient, too inflexible for an AI‑driven grid. A small European startup, Hyperscale Power, thinks it can change that by radically shrinking the transformer itself. If it works, this isn’t just another efficiency tweak. It rewires who controls the interface between cloud data centers and the power grid – and that has consequences for utilities, chipmakers, and European industry alike.
2. The news in brief
According to TechCrunch, Hyperscale Power has raised a €5 million seed round led by World Fund and Vsquared Ventures to build a prototype solid‑state transformer (SST) aimed at high‑density data centers and grid applications.
Traditional iron‑core transformers, a 140‑year‑old technology, are still the backbone of electricity distribution. But as TechCrunch reports, a new wave of SST startups has collectively raised around $280 million in recent months, promising smaller footprints, fewer components and better grid stability. Competitors include Amperesand, DG Matrix (backed by ABB) and Heron Power (founded by a former Tesla executive and funded by Andreessen Horowitz); together they have attracted over $330 million, per PitchBook figures cited in the article.
Hyperscale claims it can go further than existing SST designs by operating at significantly higher switching frequencies, allowing even more compact hardware. The company’s CEO previously built a 99.1% efficient SST during his PhD at ETH Zürich, and now wants to commercialise a system tailored to AI‑era power densities, where Nvidia server racks already draw over 100 kW and 1 MW per rack is on the roadmap.
3. Why this matters
At first glance, replacing a transformer sounds like an engineering detail. It isn’t. It’s about who owns the chokepoints of the AI economy.
Today, hyperscale data centers are running into a simple physical reality: power‑conversion equipment is starting to occupy more space than the compute it feeds. When a rack consumes 1 MW, the transformers, rectifiers and switchgear around it balloon in volume and cost. Hyperscale’s pitch is that by pushing SSTs to much higher frequencies, you can collapse that power hardware into something closer to the footprint of the racks themselves.
The immediate winners, if the tech works, are:
- Cloud providers and AI firms: More compute per square metre and potentially lower capex per watt of delivered power. That directly impacts the economics of training and running large models.
- Grid operators: SSTs can offer finer‑grained control, faster fault response and easier integration of renewables and storage, turning substations into programmable power routers instead of passive bricks of steel and oil.
The potential losers:
- Incumbent transformer manufacturers that are slow to adapt. Their margins are built on mature, low‑risk hardware; SSTs are more like power electronics systems, closer to inverters than to classical transformers.
- Data center developers who keep building around legacy assumptions. If SSTs reach commercial readiness within this decade, facilities optimised for bulky 50/60 Hz gear may age quickly.
Hyperscale’s smaller form factor is particularly interesting because it shifts where in the stack power intelligence lives. Instead of one big transformer per building, you can imagine modular, rack‑proximate units that are software‑defined. That opens the door to new business models (power‑as‑a‑service inside data centers) and new ways for cloud providers to arbitrage electricity markets in real time.
The flip side: this is critical infrastructure. Reliability, certification and long‑term cost matter more than clever topology diagrams. Hyperscale is stepping into a market where a single failure can write off a facility or damage a grid operator’s reputation.
4. The bigger picture
Hyperscale Power is not emerging in a vacuum; it’s a symptom of three converging trends.
1. AI’s power hunger.
Over the last few years, every big cloud and AI company has revised its power forecasts upward. We’ve moved from tens of megawatts per campus to hundreds, and now to gigawatt‑scale plans. Nvidia talking about 1 MW racks is not marketing theatre; it is a signal that the balance between IT load and facility infrastructure is breaking. Power electronics that used to be “good enough” suddenly look oversized, inflexible and too dumb.
2. Power electronics have quietly matured.
For years, “electronic transformers” were mostly lab curiosities. What’s different now is the wide commercial availability of silicon carbide (SiC) and gallium nitride (GaN) devices, which can switch at higher frequencies with lower losses. That’s exactly the regime Hyperscale wants to exploit. We’ve already seen similar shifts in EV inverters and fast chargers; the grid and data centers are next in line.
3. The grid is becoming bidirectional.
Renewables, batteries, EV fleets and flexible loads mean power no longer flows neatly from big plants to passive consumers. SSTs fit neatly into this new world because they can, in principle, handle multiple inputs and outputs, manage harmonics and provide synthetic inertia and voltage support through software.
Historically, attempts to radically change grid‑side equipment have taken decades; even high‑voltage DC (HVDC) lines faced long adoption cycles. The difference now is that the pressure is not coming only from utilities but from hyperscalers with far shorter planning horizons and much deeper pockets. If Amazon or Google decides that SST‑based campuses are a strategic edge, they can co‑fund and de‑risk deployments in a way utilities rarely can.
Against that backdrop, Hyperscale’s late arrival versus Amperesand, DG Matrix or Heron is less of a problem than it appears. The SST market is so early that differentiation will be defined more by execution, integration partners and regulatory approval than by who filed term sheets in 2025.
5. The European / regional angle
Hyperscale Power is a telling case study for Europe. The company is Europe‑based, the lead investors (World Fund, Vsquared) are European climate‑tech specialists, and the underlying know‑how comes from ETH Zürich – exactly the kind of deep‑tech pipeline the EU says it wants to scale.
From a policy standpoint, this intersects with several European priorities:
- Energy efficiency and climate targets. The EU’s Green Deal and “Fit for 55” package rely on squeezing more useful work out of every kilowatt‑hour. SSTs that trim conversion losses in data centers and substations speak directly to that agenda.
- Data center scrutiny. Countries like Ireland, the Netherlands and parts of Germany are already pushing back on new hyperscale facilities due to grid congestion and land use. More compact, flexible power infrastructure could make it easier to integrate large compute clusters without overbuilding physical substations.
- Strategic autonomy. Europe missed the GPU wave; it cannot afford to miss the power‑electronics layer that will define how those GPUs are powered. Companies like ABB and Siemens are obvious future partners or acquirers, but early‑stage players such as Hyperscale help keep innovation on the continent.
For European grid operators and regulators – under frameworks like the Clean Energy Package and, indirectly, the Digital Services Act’s reporting obligations on large platforms – there is also a governance question. If data centers deploy highly programmable SSTs at their grid connection points, who ultimately controls grid‑stability functions: the utility or the cloud provider? That debate has barely started.
6. Looking ahead
What happens next will be decided less in venture decks and more in dusty substations and fluorescent‑lit data halls.
In the near term (2–4 years), expect:
- Prototype deployments at edge data centers, research campuses or industrial microgrids, where operators are willing to accept higher technology risk for flexibility.
- Partnerships with incumbents, as startups like Hyperscale realise they need ABB/Siemens/Schneider‑class manufacturing, certification and service networks to be taken seriously by utilities.
- Regulatory test cases, where national regulators and TSOs/DSOs evaluate how SSTs fit into existing grid codes, fault‑ride requirements and protection schemes.
Key technical milestones to watch:
- Demonstrated multi‑year reliability in real‑world conditions, not just lab MTBF estimates.
- A clear cost trajectory per kVA, showing that SSTs can compete not only on footprint and features but also on total cost of ownership.
- Evidence that very high frequency designs, like Hyperscale’s, can manage electromagnetic interference and thermal issues at scale.
There are also downside scenarios. If AI demand plateaus, or if regulators cap data‑center growth in key hubs, the urgency to adopt risky new power hardware could fade. Conversely, a few high‑profile SST failures could set the whole category back by years, reinforcing utilities’ instinct to stick with iron and oil.
Still, the macro forces – AI expansion, renewable integration, grid congestion – are unlikely to reverse. That suggests SSTs in some form are more a question of when and where than if. The open question is whether young firms like Hyperscale can stay independent long enough to shape the architecture, or whether they will be absorbed as feature teams inside industrial giants.
7. The bottom line
Solid‑state transformers have quietly moved from academic novelty to serious contender for the most strategic box in the data‑center and grid value chain. Hyperscale Power’s bet on ultra‑compact designs aligns perfectly with AI’s space‑and‑power crunch, but it still has to clear the brutal hurdles of reliability, certification and cost. For everyone watching AI, climate tech and European industry, the message is simple: start paying as much attention to the “boring” power cabinets as to the shiny GPUs. That’s where the next bottleneck – and opportunity – is forming.
