Something changed in this market in 2026, and it did not change because of anything in the ceramics industry. AI data centres started buying solid oxide fuel cells, in gigawatts, because they could not wait for a grid connection. Between October 2025 and January 2026 roughly USD 7.65 billion of binding fuel-cell agreements for data centres were signed. On 30 June 2026 Bloom Energy and Brookfield expanded a partnership from USD 5 billion to USD 25 billion. In January, American Electric Power exercised an option for up to 1 GW at USD 2.65 billion, backing the first phase of a 1.8 GW campus in Wyoming.
If you sell electrolyte powder, the reasonable next question is what that does to demand for the material inside those cells. That question is being answered publicly in tonnes, by several parties, and the answers do not agree with each other. This guide is about why they disagree, and what you would need to know to settle it — not a fourth number to add to the pile.
Why This Reaches a Materials Buyer at All
A solid oxide fuel cell needs an electrolyte that conducts oxygen ions and almost no electrons. Zirconia stabilized with yttria (YSZ) is the long-standing default. Zirconia stabilized with scandia (ScSZ) conducts better at intermediate temperature, which is the trade we have written about in detail in the YSZ versus ScSZ comparison.
For the company at the centre of the data-centre story, the composition is not a matter of inference. Bloom Energy holds a family of granted US patents on doped scandia-stabilized zirconia electrolyte compositions — zirconia stabilized with scandia plus at least one of magnesia, zinc oxide, indium oxide or gallium oxide, with optional ceria, and in one family scandia and indium oxide together at 10–13 mol%. Patents describe what a company has claimed, not necessarily what is in the cell shipping today, and we are not going to pretend otherwise. But the direction is disclosed rather than guessed: this is a scandium-bearing electrolyte technology.
That is the whole reason a data-centre power story lands on a powder buyer's desk. Scandium is not a bulk commodity with a deep market. Global output is on the order of 40 tonnes a year of scandium oxide, from byproduct recovery rather than dedicated mines, for the reasons set out in our piece on why scandium is so scarce. A demand shock measured in gigawatts meets a supply base measured in tens of tonnes.
The Numbers in Circulation, and How Far Apart They Are
Here is what is publicly claimed, with the source of each claim rather than an average of them.
Read the right-hand column before the left. Two of these are a company describing its own supply position. Two are market projections. One is an estimate that the underlying statistical sources decline to state firmly. They are not the same kind of number, and averaging them would produce something with no meaning at all.
They also answer different questions. "Our supply chain can support 25 GW" is a statement about procurement capability, and it is not the same claim as "25 GW of cells can be built from 40 tonnes of scandium." A supply chain that can support a volume may be doing so through byproduct streams not currently counted in the 40-tonne figure — which is precisely what the disclosure describes, naming titanium, nickel, cobalt, bauxite and uranium processing as the sources.
What the Arithmetic Would Actually Require
We are not going to convert gigawatts into tonnes here, and it is worth being explicit about why, because the conversion looks like a one-line calculation and is not.
To get from a gigawatt of installed capacity to a mass of scandium oxide, you need, at minimum:
- Power density, in watts per square centimetre of active cell area, at the operating point the system actually runs at — not at peak. This sets how much cell area a gigawatt requires.
- Electrolyte thickness. Anode-supported thin-film designs run in the single-digit to low-tens of micrometres; electrolyte-supported layers run an order of magnitude thicker. This alone moves the answer by 10× or more.
- Scandia loading, in mol% converted to a mass fraction, and then the scandium fraction of the scandia. A conversion between the oxide and the element is required here, and getting it backwards is a common error — the same oxide-to-element problem that shows up on every COA.
- Density of the sintered electrolyte, to turn a volume into a mass.
- Manufacturing yield, including material lost in tape casting, screen printing and sintering. Powder purchased is not powder shipped in a cell.
- Stack replacement rate over the contract life. A twenty-year power agreement with periodic stack replacement consumes more material than a one-time build, and the replacement interval is a commercial term, not a public one.
Every one of those is either proprietary or varies by an order of magnitude across designs. Change electrolyte thickness alone from 60 µm to 6 µm and the mass falls tenfold with no change to any other input. That is why the published figures span from 200 to 1,000 tonnes a year for what is nominally the same technology: the range is not disagreement about the market, it is disagreement about the assumptions, and none of the sources shows theirs.
If someone quotes you a scandium tonnage for the data-centre buildout, the useful question is not whether the number is high or low. It is which of the six inputs above they used, and where they got them.
What a Buyer Can Reasonably Conclude
Three things survive the uncertainty.
The direction is not in doubt, even if the magnitude is. Whether SOFC scandium demand is 200 or 1,000 tonnes a year, both are large against roughly 40 tonnes of current output. A byproduct-recovery supply base does not respond to price the way a mined commodity does, because the operator's primary product is titanium or nickel, and scandium recovery is a secondary decision with its own capital cost and lead time.
Supply diversification claims deserve the same scrutiny as any other spec claim. "No single supplier or nation controls its supply" is a reasonable thing for a manufacturer to say about a byproduct-sourced material, and it is materially different from the position on yttrium, where China's April 2025 licensing regime bites directly — a divergence we traced in the rare earth controls piece. Scandium sits under the same MOFCOM announcement, which is covered separately.
If you buy ScSZ, the constraint you will feel first is not price. It is qualification. A change of scandia source or a change in dopant package changes sintering behaviour and long-term phase stability, and requalifying an electrolyte is measured in months. That is the same pattern as the yttrium disruption: the binding constraint turned out to be requalification time rather than the price line.
For anyone specifying this material, the practical step is unchanged by the news. Write the specification so a substitution cannot arrive silently: state the stabilizer and its mol%, name the dopants, state the measurement method, and require production-lot certificates before the lot ships rather than after. The supplier landscape covers who can actually quote against that.
Frequently Asked Questions
Does the AI data centre buildout mean scandium prices will rise?
Direction and magnitude are different questions. Demand pressure on a byproduct-recovered material with roughly 40 tonnes of annual output is real. But scandium already repriced sharply after the April 2025 export controls, so a further move depends on how much of the demand expectation is already in the price — and on whether recovery capacity from titanium and nickel streams expands, which is a capital decision by producers whose main product is something else. We do not have a defensible price forecast and are not offering one.
Can I calculate scandium demand from the announced gigawatts myself?
Only if you have the six inputs listed above, and most of them are proprietary. The calculation is straightforward arithmetic once you have power density, electrolyte thickness, scandia loading, sintered density, yield and stack replacement rate. Without them, any figure you produce is the assumptions talking, not the gigawatts.
Is YSZ affected by this too, or only ScSZ?
Both, but differently. They are alternative electrolytes for the same job, so demand for one interacts with the other, and the choice between them is an engineering trade covered in our comparison. Their supply risks are not the same shape: yttrium is under Chinese export licensing at commodity volume, while scandium is a byproduct-supply problem. A programme exposed to both is exposed to two unrelated failure modes.
Does every solid oxide fuel cell use scandium?
No. YSZ remains the long-standing default electrolyte, and many systems use it. Scandia-stabilized zirconia buys conductivity at intermediate temperature and pays for it in price and phase stability. Which one a given manufacturer uses is a design decision, and where it is not publicly disclosed we do not assume it.
Is this article saying the published tonnage figures are wrong?
No. It is saying they are not comparable. Each may be internally sound under its own assumptions; the problem is that the assumptions are not published alongside them, so the figures cannot be reconciled or checked. Treat any of them as a scenario with hidden inputs rather than as a measurement.
References (Public Sources)
- Bloom Energy scandium oxide supply chain disclosure (7 July 2026), as reported 14 July 2026 — 25 GW annual production support, byproduct sourcing from titanium, nickel, cobalt, bauxite and uranium processing, "several hundred tonnes" recoverable from titanium processing.
- US patent family, Bloom Energy Corporation — doped scandia-stabilized zirconia electrolyte compositions (scandia with magnesia, zinc oxide, indium oxide or gallium oxide; optional ceria; scandia plus indium oxide at 10–13 mol% in one family).
- Reported data-centre fuel cell agreements: approximately USD 7.65 billion October 2025 – January 2026; Bloom Energy–Brookfield partnership expanded from USD 5 billion to USD 25 billion, 30 June 2026; American Electric Power option for up to 1 GW at USD 2.65 billion, January 2026, supporting a 1.8 GW campus developed by Crusoe and Tallgrass.
- Market commentary citing SOFC scandium oxide demand assumptions of 200–300 t/yr, and a historical projection as high as 1,000 t/yr.
- US Geological Survey, Mineral Commodity Summaries — scandium (crustal abundance, recovery routes; no firm global production total published).
Figures in this article are as reported at the time of writing and this sector is moving weekly. Deal values and capacity claims are the parties' own statements, not independently verified by us.
None of the circulating tonnage figures can be reconciled with the others, so the firmest number available to a buyer is the one a mill will put on a quotation. Nami Tech Solutions (NTS) approaches those mills directly, in their own language and against one written specification — stabilizer and mol%, named dopants, measurement method — so the quotations come back comparable instead of assembled from assumptions.
