An Inverter Shortage Is a Storage Shortage

The cost curve broke in the part of the system nobody forecasts

The storage industry forecasts in kilowatt-hours and buys in megawatts of power conversion, and in 2026 those two numbers stopped moving in the same direction. Cells got cheaper. The installed asset did not.

Between 2024 and 2025 the price of a lithium-ion cell fell roughly 20 per cent, to about USD 115/kWh. IRENA measures the whole system rather than the cell — the installed asset, inverter and interconnection included — down 93 per cent since 2010, to about USD 192/kWh. Those two facts produced the sentence that opens every storage deck: the cost curve is still falling, and it is falling because of chemistry.

The 2026 numbers say otherwise. Lazard’s 2026 levelised cost of storage puts utility-scale storage without the investment tax credit at USD 210–292/MWh, against USD 115–254/MWh in 2025. That is +36.04 per cent on the midpoints in twelve months — the first year in a decade in which the direction reversed.

With the credit, 2026 lands at USD 148–209/MWh: a midpoint 3.25 per cent below the 2025 midpoint that carried no credit at all. Which produces the uncomfortable reading of the year. The tax credit is not a subsidy sitting on top of a falling cost curve. On these figures the credit is the curve — and an incentive that depends on a foreign-entity-of-concern test is not a subsidy at all when the test fails. It is equity.

That is the financial story, and it is the one the market trades on. It is not the story that governs whether projects get built.

The physical gate is a semiconductor

An IGBT or silicon-carbide module is a small line in a storage invoice and a hard gate on the schedule. The 2026 capacity gap in power semiconductors is above 18 per cent. Lead times are running 30 to 45 weeks, and at least one manufacturer has pushed customer allocations out a further 15 weeks on top of that.

Those three numbers do something that a cell price cannot: they convert a cost problem into a schedule problem. A pipeline can be fully contracted on cells, fully financed, fully permitted, and still miss its commercial operation date because the converters arrived a quarter late. Storage is sold per megawatt-hour and energised per megawatt, and it is the megawatt where the queue forms.

The cleanest illustration is a tender. POWERGRID’s 500 MW / 1,000 MWh bid in July 2026 needs 500 MW of conversion regardless of what cells cost. The ratio between the two halves of that specification — one megawatt of conversion for every two megawatt-hours of storage — is fixed by grid requirements and by the duration the buyer wants, not by the battery. Nothing on an LFP line shortens a fab qualification, and nothing on a cathode roadmap changes how many modules a 500 MW interconnection requires.

Why large-format cells answer a different question

The industry’s answer to cost pressure has been the larger cell: fewer racks, fewer busbars, fewer fuses, fewer high-voltage connections per megawatt-hour. That is a real saving in balance-of-system cost and in assembly labour, and it is where a substantial part of the recent cost reduction has come from.

It is also, precisely, an answer about components per megawatt-hour. A converter shortage is a constraint on converters per megawatt. A bigger cell reduces the first and leaves the second untouched. This is why an announcement that a container now holds more energy says nothing about whether the project it belongs to can be energised — and why large-format cells have not, on their own, moved a single allocation date.

The statutes have dates; the fabs have lead times

What makes 2026 unusual is that the demand side of storage became a schedule of statutes rather than a forecast of costs. Germany’s transmission system operators opened inertia procurement in January 2026. POWERGRID tendered 500 MW / 1,000 MWh in July 2026. New Hampshire passed HB 1718. Each of those is a dated obligation with a procurement attached, and each one pulls forward a delivery date that a fab decides.

A statute has a date on it. A fab has a lead time. When the two collide, the lead time wins — and that is the sense in which a storage forecast built on statutes is more reliable than one built on chemistry, right up to the point where the equipment does not arrive. The procurement is knowable. The allocation is not.

What would change this argument

Three things, in order of how quickly they would show up in the data.

One: capacity that converts into qualified supply inside 45 weeks. Silicon-carbide and IGBT expansion announcements are not the test. Qualification is. A fab that is announced in 2026 and qualified in 2029 does nothing for a 2027 commercial operation date, and the announcements should be read against the qualification calendar rather than the construction calendar.

Two: an ex-credit LCOS print that resumes falling. If 2027 comes in below USD 210/MWh without the credit, the 2026 move was policy and the underlying equipment curve is intact. If it does not, the constraint was never financial and the credit was masking a physical shortage rather than causing a cost one.

Three: evidence that a larger format reduces converters per megawatt. That would be a genuine structural relief rather than a component-count saving, and it has not been demonstrated.

One caveat that belongs in print. The base year in Pheonix’s stationary storage report is derived from a single published web estimate. The growth rate and the direction are the usable output; the level should not be defended to three significant figures. The cost figures above are Lazard’s and IRENA’s, quoted as published, and the arithmetic on them is simple and checkable.

The companion to this piece, The Cheaper Cell and the Dearer Megawatt-Hour, sets out why the 2026 levelised cost reversal sits on top of an ITC that is really a supply-chain test. Read together, they describe a market whose price is now set by statutes and whose schedule is set by fabs.