Hyperscale Power is the latest startup to challenge 140-year-old transformer tech

Startups focused on solid-state transformers have raised $280M recently, and Hyperscale Power closed a €5M seed to build a prototype. The technology—including Hyperscale co-founder Daniel Rothmund's 99.1% efficient design—aims to cut components and shrink power-conversion footprints by operating at tens of kHz. This matters for AI hyperscalers: current Nvidia racks exceed 100 kW and roadmaps target 1 MW racks, where conventional transformers/rectifiers can be >2x the size of server racks, implying material capex and space savings if adoption accelerates.

Analysis

The migration from heavy passive magnetics to high-frequency, semiconductor-dominated power conversion will reallocate margin pools down the supply chain: wafer fabs for SiC/GaN, power IC suppliers, high-reliability capacitors, and thermal management vendors capture a disproportionate share of incremental revenue and gross margin versus traditional metalworking and bulk magnetic manufacturers. Expect top-line growth concentrated among a narrow set of suppliers that can scale wafer capacity and power-module assembly in the next 12–36 months; incremental demand will show up first in their utilization rates and lead times rather than in end-market orders.

A practical gating factor is non‑technology friction — certification timelines, OEM qualification cycles, and grid interconnection rules — which means broad deployment is likely to be staggered across regions and hyperscalers. These regulatory and reliability frictions create idiosyncratic windows for winners to lock design wins (multi-year, low-churn revenue) but also produce binary downside events if a field failure or an EMI/certification setback hits a marquee customer and triggers widespread re-testing.

From a macro view, this shift increases coupling between data-center capex and constrained semiconductor capacity: power-electronics suppliers become a choke point for data-center scale-outs, creating a leverage point for companies that own long-lead fabs or captive SiC/GaN capacity. That coupling also accelerates the optionality for on-site storage and DC-coupled renewables — the faster power conversion becomes compact and controllable, the quicker developers can rationalize integrated battery+compute deployments, changing where and how hyperscalers place future supply nodes.

Key Decisions for Investors

• Long Wolfspeed (WOLF) — 12–24 month horizon: buy shares or 12–18 month call calendar to play SiC capacity tightness. Risk/reward: asymmetric upside (target +80–150% if adoption accelerates) vs drawdown of ~30–40% if SiC pricing or demand disappoints; position size 1–2% of risk budget and stagger entries on pullbacks.
• Long select power-IC names (e.g., ON Semiconductor (ON), STMicroelectronics (STM)) — 6–18 months: buy 6–12 month call spreads to limit premium paid while retaining upside to higher utilization. Risk/reward: 1:3 skew if modules shift to wide-bandgap semiconductors; hedge with small put below 15–20% loss threshold.
• Relative pair — long power-semiconductor supplier (STM/ON) vs short legacy transformer/installation-exposed capex name (ABB) — 12–36 months: size as a market-neutral pair (e.g., 60% notional long / 40% short) to capture technology-driven share shifts. Risk/reward: moderate net beta, target 25–40% relative outperformance; primary risks are sector-wide capex slowdown or simultaneous policy support for incumbents.
• Private/venture allocation — up to 2% AUM reallocation to late-stage SST/ power-electronics funds or structured secondaries: expect 4–7 year illiquidity, binary outcomes. Use as a diversification bucket for high-convexity upside while capping exposure to avoid concentration risk.