The battery that rusts: my most expensive assumption finally got sources

Image above: AI impression, not a construction drawing.

For weeks, my assumptions register has carried the same uncomfortable label next to the price of seasonal storage: low confidence, decisive for the conclusion. Twenty euros per kilowatt-hour of storage capacity, the hydrogen route. The sensitivity analysis showed the main conclusion survives the whole price band from 20 to 150 euros: seasonal storage beats the pure battery route everywhere. But the word "cheap" hangs entirely on that one control.

So I gave it a research day, with one fixed rule: every figure gets a label. Measured, list price, contracted, claim, analyst estimate or model study. A narrow band without sources is worthless; a wide band with sources is gold.

The model is parameterised on hydrogen: electrolyser, pressure vessel, fuel cell. It turns out exactly one system in that class was ever actually for sale for buildings: the German HPS Picea 2, around 100,000 euros for roughly 1,500 kWh of usable storage. That is 67 euros per kilowatt-hour, everything included. Not 20.

It gets sharper: HPS was declared insolvent in April 2025. So the list price was apparently already a loss-making price; the real cost sat higher. Bosch, meanwhile, stopped its small fuel-cell system. The route my model computes on is commercially disappearing at neighbourhood scale.

The explanation is scale. In a salt cavern, hydrogen storage costs 2 to 5 cents per kilowatt-hour. In pressure vessels at neighbourhood scale, with compression, piping, safety zoning and permits on top, hundreds of times as much. My 20 euros was the industrial tank price. A neighbourhood does not buy an industrial tank.

While the hydrogen route shrank, something else surfaced in the research. Iron-air batteries store energy by letting iron rust in a controlled way and then reversing the rust. Slow, bulky, 40 to 50% round-trip efficiency. But the energy carrier is iron, and iron is dirt cheap per kilowatt-hour.

The world's first grid-connected iron-air battery has been running since July 2025. Not in California. In Delft. Ore Energy, a TU Delft spin-off, builds them in containers of 4.2 megawatt-hours. The seasonal store my model computes for the 90% neighbourhood fits in two of those containers, from a Dutch supplier, without a hydrogen permitting process. Claimed prices sit at 30 to 60 euros per kilowatt-hour, at an efficiency that beats the hydrogen route. Budget Thuis signed for a gigawatt-hour; in America, Google ordered thirty from Form Energy.

I parameterised a hydrogen system and meanwhile the market built something that looks better for a neighbourhood. That is no disgrace for the model. It is a correction, though.

The defensible band for neighbourhood-scale seasonal storage comes out at 30 to 100 euros per kilowatt-hour, with a midpoint around 60. My old 20 stays in the sensitivity analysis, but with a new label: optimistic lower bound, not a base case. The explorer on this site fortunately already used 100 euros as its honest default; in hindsight that choice was right.

The next step is a scenario variant: iron-air alongside hydrogen, same physics, different price structure and efficiency. Not as a replacement. The comparison itself is the insight.

• Not a single figure in this story is a quotation. Ore Energy publishes no prices; the 30 to 60 euros comes from analysts and one old factory claim by Form Energy from 2021.
• Pure seasonal cycles, one or two per year, are not what these batteries are designed for; they are built for multi-day storage with dozens of cycles per year. At one cycle per year only the capital cost counts, and that is exactly what is low. But measured it is not.
• The hardest anchor remains a bankrupt company: the Picea list price bounds the total, not the split per kilowatt-hour and kilowatt.
• The first real quotation request would be the first contracted anchor at neighbourhood scale. If you have one, or want to request one: I will read along. The full source dossier, every label included, is in the lab repository.