Heron Power cuts MV‑to‑rack cost by 1/3

Heron Power Electronics has published a detailed case for rewiring how AI data centers are powered, arguing that the legacy path from medium voltage to the rack has become a construction bottleneck as rack densities climb. On its website and in a May 20 post on X, the company said its 800 VDC blueprint can cut MV-to-rack electrical cost by 65%, reduce installation labor by 90% and shorten construction schedules by 25%. The company’s pitch is not a small component swap. Heron says a 4.2 megawatt solid-state transformer with an integrated battery buffer can replace the medium-voltage transformer, UPS, switchgear and PDU that sit in a conventional 480 VAC chain. In Heron’s framing, that also removes the need for a dedicated electrical room, or “gray space,” in the data hall. ### Why is Heron targeting the MV-to-rack segment? (heronpower.com) AI rack power is the pressure point Heron is trying to address. In its blueprint, the company says conventional 480 VAC architecture was built for lower-density racks and now requires four conversion steps, four separately procured equipment categories and four sequential installation phases before power reaches GPUs. Those layers add cost, lead time and conversion losses, according to the paper. (heronpower.com) NVIDIA has been pushing the same broader shift toward 800 VDC. On its official data-center architecture pages, NVIDIA says data centers will gradually move from AC distribution to 800 VDC and that the approach is meant to support 1 megawatt IT racks and beyond, starting in 2027. NVIDIA says the goal is to reduce conversion stages, lower losses and provide a path from current facilities to all-800 VDC designs. (heronpower.com) ### What exactly is Heron saying it replaces? Heron’s product page says Heron Link converts 34.5 kV AC to 800 V DC while providing ripple stabilization and ride-through support. The company says that lets operators eliminate low-voltage transformers, main switchboards, UPS systems, PDUs, remote power panels and rack power supply units from the traditional chain. The hardware block Heron describes is rated at 4.2 MW and 98.5% efficiency, according to the company’s data-center page. (nvidia.com) Heron also says the system is designed for native, protected, per-rack 800 VDC outputs and no derating up to 45 degrees Celsius. ### Where do the cost and labor claims come from? Heron’s own blueprint is the source of the headline numbers. The company says its comparison uses a traditional 480 VAC, 12 MW data hall as the baseline and measures the alternative against that design. (heronpower.com) In the executive summary, Heron lists four claimed gains: 65% lower MV-to-rack electrical cost, 90% lower installation labor, 25% faster construction schedule and 50% lower power-delivery losses. Those figures should be read as company claims from a vendor-authored blueprint, not as independently audited industry benchmarks. Heron has made the white paper available through its resources page and a dedicated download page for “A Blueprint for 800 VDC Datacenters.” ### How does this fit into Heron’s broader business? Heron Power is a Scotts Valley, California company founded by former Tesla executive Drew Baglino. (heronpower.com) In May 2025, the company said it had raised $38 million in Series A funding led by Capricorn Investment Group, and said the money would help complete engineering of Heron Link, its solid-state transformer platform. (heronpower.com) By February 2026, Heron had announced a larger $140 million raise to scale production of solid-state transformers for data centers and the grid, according to TechCrunch. That financing underscored that the company is moving beyond concept-stage messaging and trying to industrialize the hardware behind the architecture it is now promoting. ### What should readers watch next? (capricornllc.com) The next concrete test is deployment. Heron’s website now pairs the blueprint with product pages for Heron Link and a separate technical paper on “Unlocking Efficient Power Conversion for Gigascale AI,” both aimed at hyperscale and AI-factory builders. Whether the cost, labor and schedule claims hold up will depend on customer projects, procurement decisions and field installations rather than the blueprint alone. (heronpower.com) (techcrunch.com)