$BE BLOOM ENERGY SCANDIUM SUPPLY RISK
EXECUTIVE CONCLUSION
The scandium issue represents a genuine strategic supply constraint for Bloom Energy, but the available evidence does not establish an imminent 2026 production stoppage. The risk is nonlinear and directly conditional on Bloom achieving the production ramp embedded in the equity narrative. At approximately 0.7-0.8 GW of annual system and replacement-unit throughput, scandium requirements appear manageable through some combination of existing contracts, Chinese export licenses, Japanese-Philippine supply, inventories and limited Canadian output. At 2.3-2.5 GW of total annual throughput, including field-replacement units, Bloom could require approximately 35-100 metric tons of scandium oxide, with a defensible central range of 46-75 metric tons. At 5.5 GW of total throughput, requirements rise to approximately 55-220 metric tons, with a central range of 110-165 metric tons. The upper end is broadly consistent with Hunterbrook’s 220-metric-ton estimate, but that estimate is not uniquely determined by public technical data.
The most defensible estimate of Bloom’s current fresh scandium oxide requirement is 15-40 metric tons per GW. A 20-30 metric-ton central range is more appropriate for investment underwriting than Hunterbrook’s point estimate of approximately 37 metric tons. Hunterbrook’s estimate is physically plausible under a 100-micrometer electrolyte, approximately 25 W per cell, 10 mol percent scandia, 68 percent manufacturing yield and no recovery of scandium from rejected material. The estimate becomes materially lower if Bloom operates thinner electrolytes, higher cell power, higher yields or meaningful internal scrap recovery. Public patents demonstrate technical pathways toward each of these improvements, but no public disclosure establishes the design, yield or recycling performance of Bloom’s current commercial production lines.
The global market is too small and opaque to support conventional commodity-market analysis. USGS estimates 2025 scandium oxide consumption at approximately 60 metric tons, production at approximately 80 metric tons and nameplate capacity above 90 metric tons. China was the leading producer. Publicly visible non-China operating supply appears to be approximately 10-15 metric tons annually before contractual commitments, government stockpiling and customer qualification are considered. The apparent 20-metric-ton difference between global production and consumption does not demonstrate that Bloom has access to 20 metric tons of freely available 4N material. That difference may reflect inventory accumulation, lower-purity material, intermediates, unqualified output, inconsistent statistical definitions or output committed to other customers.
Bloom’s statement that its supply chain can support 25 GW of annual production is not reconcilable with publicly visible operating supply under reasonable estimates of its current scandium intensity. At 7 metric tons per GW, 25 GW requires 175 metric tons annually. At 20-30 metric tons per GW, the requirement is 500-750 metric tons. At Hunterbrook’s approximately 37-metric-ton estimate, the requirement is approximately 925 metric tons. Bloom states that several hundred metric tons can potentially be recovered from global industrial process streams, but potential recoverability is not equivalent to installed separation capacity, high-purity output, qualified material, export authorization, contracted volume or annual deliverability. Supporting 25 GW from a 200-300-metric-ton supply pool would require intensity of approximately 8-12 metric tons per GW, implying a significant technical change from the legacy electrolyte-supported configuration or an unusually high level of closed-loop recovery. Bloom’s 8-K gives the claim greater legal and informational significance, but the absence of tonnage, suppliers, purity specifications, contract tenor and timing prevents external verification.
The most important investment conclusion is that scandium is more likely to cap Bloom’s medium-term volume and terminal earnings upside than to cause an immediate production collapse. Current scandium prices are not sufficiently high to create a material cost problem at present output. The greater risk is physical allocation. Bloom can economically pay multiples of the quoted Chinese price because scandium represents a small share of system revenue. However, price cannot resolve a shortage of qualified material if exports are not licensed, material is already committed, or new Western production has not been commissioned and qualified.
The short reports are directionally stronger than their precise deficit timing. Hunterbrook establishes that a 5 GW ramp would create a requirement comparable with or greater than the entire present market under legacy technical assumptions. It also presents credible evidence of Chinese participation in Bloom’s direct and indirect supply chain. Crossroads identifies the same strategic contradiction between an AI-driven capacity narrative and a physically constrained critical-mineral input, although the public Crossroads page does not disclose the underlying quantitative model and therefore provides limited independent support for a specific tonnage conclusion. Both firms disclose positions that benefit from a decline in Bloom’s securities. Bloom is economically motivated to minimize the perceived risk, while Chinese suppliers and Western development-stage projects are economically motivated to maximize claims regarding their respective capacities. USGS and primary corporate filings therefore deserve the greatest evidentiary weight.
OVERALL RISK ASSESSMENT
The probability of a scandium-driven interruption to Bloom’s 2026 production should be classified as low-to-moderate rather than high. Current production is likely supported by preexisting contracts, working inventory, licensed Chinese exports and established Japanese-Philippine material. Bloom made an explicit July 2026 representation that it has sufficient scandium oxide for current demand and backlog. No public evidence demonstrates that a current supplier has terminated shipments, that a required Chinese license has been denied or that Bloom’s electrolyte production is operating below plan because of scandium availability.
The risk of procurement tightness, elevated contract prices or mandatory prepayments during 2027-2028 should be classified as moderate-to-high if total manufacturing throughput moves toward 2-2.5 GW. At 20 metric tons per GW, 2.3-2.5 GW requires approximately 46-50 metric tons. At 30 metric tons per GW, the requirement is approximately 69-75 metric tons. At 37 metric tons per GW, the requirement is approximately 85-93 metric tons. The upper part of this range approaches or exceeds current global production before competing demand, export controls, purity requirements and qualification losses.
The risk that scandium becomes a binding volume ceiling during 2028-2030 should be classified as high if Bloom attempts to manufacture approximately 5 GW of new systems annually while also supporting replacement demand. At 5.5 GW of total throughput, even a materially improved 15-metric-ton intensity requires 82.5 metric tons annually. A 20-metric-ton intensity requires 110 metric tons, 25 metric tons requires 137.5 metric tons, 30 metric tons requires 165 metric tons and 37 metric tons requires approximately 203.5 metric tons. A 220-metric-ton requirement is therefore a credible stress case, but not the sole reasonable case.
The risk is self-limiting in a way that is important for equity analysis. Weak end demand, project delays or a failure to convert headline backlog into shipments would reduce scandium requirements and make the supply problem less acute. Strong end demand and successful execution would increase scandium requirements until physical supply becomes the constraint. Scandium therefore functions as a conditional cap on the bull case. It does not independently prove that Bloom’s reported backlog will convert, but it makes the combination of rapid demand conversion and unrestricted manufacturing growth more difficult to underwrite.
WHY SCANDIUM MATTERS TO BLOOM’S TECHNOLOGY
Bloom uses a scandia-stabilized zirconia electrolyte in its solid oxide fuel cells. Zirconia provides the underlying ceramic structure, while scandium oxide stabilizes the crystal structure and improves oxygen-ion conductivity. Higher ionic conductivity permits lower operating temperatures, fewer cells for a given output, better efficiency and potentially lower degradation. Bloom states that scandium improves performance, operating life and fuel efficiency and that alternative materials have been developed, but scandium remains the best material for the current platform.
Scandium’s importance is not adequately described by its low percentage of system mass. The electrolyte is thin, but every cell requires it, and a utility-scale system contains a very large number of cells. A material that represents a small share of each cell can therefore become a large absolute requirement when production is expressed in GW. The relevant measure is not scandium as a percentage of total system weight. The relevant measure is fresh kilograms of qualified Sc2O3 required per 100 kW of system output after manufacturing losses and recycling.
Bloom patents describe scandia concentrations generally in the range of 8-11 mol percent, with 9-11 mol percent and approximately 10 mol percent appearing in representative formulations. Mol percent should not be confused with weight percent. Based on the molecular weights of Sc2O3 and ZrO2, 8 mol percent Sc2O3 corresponds to approximately 8.87 weight percent, 9 mol percent corresponds to approximately 9.97 weight percent, 10 mol percent corresponds to approximately 11.06 weight percent and 11 mol percent corresponds to approximately 12.15 weight percent. The oxide content is therefore materially higher on a weight basis than a casual reading of the mol percentage would suggest.
The central engineering issue is the trade-off between scandia concentration, ionic conductivity, phase stability and degradation. Bloom’s patents discuss ageing in conventional scandia-stabilized zirconia and the use of ceria, yttria, ytterbia and other co-dopants to stabilize the cubic phase. Lowering scandia content cannot be assumed to generate a proportional reduction in total scandium demand without consequences. Lower ionic conductivity can require a thinner electrolyte, higher operating temperature, larger active area, more cells or lower output per cell. Each response can offset the apparent material saving or introduce thermal, mechanical and service-life risks.
BLOOM’S SCANDIUM INTENSITY
A bottom-up reconstruction provides the strongest framework because Bloom does not disclose scandium consumption. Under the Hunterbrook-style legacy case, a cell has approximately 100 cm² of active electrolyte area, an electrolyte thickness of 100 micrometers, ceramic density of approximately 5.7 g/cm³ and 10 mol percent Sc2O3. These assumptions produce approximately 0.63 g of Sc2O3 contained in each cell. At approximately 25 W per cell, 1 GW requires approximately 40 million cells, resulting in approximately 25.2 metric tons of Sc2O3 physically sealed into finished cells.
Manufacturing yield then becomes decisive. With 68 percent first-pass yield and no recovery of scandium from rejected material, fresh purchases rise from approximately 25.2 metric tons sealed into product to approximately 37.1 metric tons per GW. This is the central foundation of Hunterbrook’s approximately 37-metric-ton estimate. The arithmetic is internally consistent. The uncertainty lies in whether 100 micrometers, 25 W, 68 percent yield and zero recovery accurately describe Bloom’s current commercial process.
The zero-recovery assumption creates meaningful upward bias if Bloom can reprocess green ceramic tape, trim, unused slurry or other presintered scrap. Under a 68 percent yield, recovery of 50 percent of rejected scandium reduces fresh requirements from 37.1 to approximately 31.2 metric tons per GW. Recovery of 80 percent reduces the requirement to approximately 27.6 metric tons. Recovery of 90 percent reduces it to approximately 26.4 metric tons. Complete recovery reduces purchases to the approximately 25.2 metric tons physically contained in finished product. Sintered, metallized, brazed or contaminated rejects are more difficult to recycle than unfired material, so complete recovery would be unrealistic, but zero recovery is also an aggressive assumption. USGS reporting of no scandium recycling refers to the broader market and does not resolve the amount of internal manufacturing rework conducted inside Bloom or its ceramic suppliers.
Electrolyte thickness, cell output and scandia concentration can lower intensity further. A moderate improvement case using a 75-micrometer electrolyte, 30 W per cell, 9 mol percent scandia and 80 percent yield produces approximately 17.8 metric tons of fresh Sc2O3 demand per GW before scrap recovery. An advanced case using 55 micrometers, 35 W, 8 mol percent scandia and 85 percent yield produces approximately 9.3 metric tons per GW. An aggressive case using 50 micrometers, 40 W, 8 mol percent scandia and 90 percent yield produces approximately 7 metric tons per GW. Bloom patents contemplate electrolyte-supported cells in approximately the 50-100-micrometer range and also describe substantially thinner anode-supported structures, but patent coverage does not demonstrate deployment at commercial scale.
Anode-supported cells can reduce electrolyte thickness to a small fraction of the electrolyte-supported design, but that transition is not a simple materials substitution. The anode becomes the primary mechanical support, modifying shrinkage control, thermal expansion, redox behavior, gas diffusion, flatness, fracture risk, sintering sequence and stack assembly. Qualification would need to cover efficiency, degradation, thermal cycling, start-stop performance, sealing and long-duration field reliability. A company selling systems under long service commitments would have limited tolerance for an accelerated architecture change that reduces scandium consumption but damages stack life.
The appropriate underwriting range for current or near-current Bloom production is therefore 15-40 metric tons per GW. The lower end incorporates material progress in thickness, cell power, yield and recycling. The upper end approximates Hunterbrook’s legacy case. A 20-30 metric-ton range provides a balanced central estimate. A sub-10-metric-ton intensity should be treated as a future technology case until commercial evidence is disclosed.
A top-down market cross-check supports a material but not precisely measurable consumption level. Hunterbrook estimates that solid oxide fuel cells represent approximately 2-thirds of global scandium demand and that Bloom represents approximately 75 percent of solid oxide fuel-cell demand. Applying these shares to approximately 60 metric tons of 2025 global consumption produces approximately 30 metric tons attributable to Bloom. Against estimated 2025 throughput of approximately 0.7-0.8 GW, including replacement units, the implied intensity is approximately 37.5-42.9 metric tons per GW. This result supports the upper end of the bottom-up range, but every component is estimated rather than disclosed. The cross-check is useful corroboration, not proof.
CURRENT AND FUTURE BLOOM DEMAND
Estimated 2025 Bloom throughput of approximately 0.7-0.8 GW, including approximately 0.3 GW of field-replacement-unit activity, implies annual scandium demand of approximately 10.5-32 metric tons under the 15-40-metric-ton range. The 20-30-metric-ton central range produces approximately 14-24 metric tons. Hunterbrook’s top-down market-share method produces approximately 30 metric tons. A reasonable current-demand estimate is therefore approximately 15-30 metric tons annually, with material uncertainty surrounding replacement accounting, cell architecture and fresh-material recovery.
Replacement demand is strategically important because it is less discretionary than new-system growth. Installed systems require continued service performance, and replacement modules protect uptime, contractual service revenue and customer relationships. If replacement throughput increases from approximately 0.3 GW toward 0.5 GW, replacement requirements alone could consume approximately 7.5-20 metric tons annually, with a central estimate of 10-15 metric tons. In a shortage, material would rationally be prioritized toward service obligations and the most valuable customers, potentially reducing new-system shipments before installed-system support is interrupted.
A 2 GW new-system production target combined with 0.3-0.5 GW of replacements produces total throughput of 2.3-2.5 GW. At 15 metric tons per GW, demand is approximately 34.5-37.5 metric tons. At 20 metric tons, demand is 46-50 metric tons. At 25 metric tons, demand is 57.5-62.5 metric tons. At 30 metric tons, demand is 69-75 metric tons. At 37 metric tons, demand is approximately 85.1-92.5 metric tons. At 40 metric tons, demand is 92-100 metric tons.
The 2 GW threshold is therefore where scandium moves from a manageable procurement item to a market-structure issue. At a 20-metric-ton intensity, Bloom would require most of 2025 global consumption. At a 30-metric-ton intensity, Bloom would require more than estimated 2025 global consumption. At 37-40 metric tons, Bloom would require approximately all current annual production, before aerospace alloys, semiconductors, defense stockpiles, other fuel-cell manufacturers and inventory demand.
A 5 GW new-system target combined with approximately 0.5 GW of replacements produces total throughput of 5.5 GW. At 10 metric tons per GW, demand is 55 metric tons. At 15 metric tons, demand is 82.5 metric tons. At 20 metric tons, demand is 110 metric tons. At 25 metric tons, demand is 137.5 metric tons. At 30 metric tons, demand is 165 metric tons. At 37 metric tons, demand is approximately 203.5 metric tons. At 40 metric tons, demand is 220 metric tons.
Hunterbrook’s 220-metric-ton estimate is therefore best interpreted as a high-intensity stress case based on legacy manufacturing assumptions. It should not be dismissed as physically impossible, but it should not be presented as the sole answer. Even the lower 82.5-110-metric-ton cases require Bloom to absorb an amount equal to or greater than current global output or consumption. The short thesis remains economically relevant even after substantial adjustment to Hunterbrook’s intensity assumptions.
THE GLOBAL SUPPLY BASE
Scandium is geologically abundant but economically scarce. It is rarely concentrated at grades that justify dedicated mining and is generally recovered as a by-product from titanium, nickel, cobalt, uranium, zirconium and other industrial process streams. The amount contained in global ores and tailings is therefore not the relevant near-term supply measure. The relevant measure is material that can be economically separated, refined to the required purity, qualified by Bloom, exported to Bloom’s manufacturing chain and delivered under enforceable contracts.
Supply should be analyzed through 6 sequential filters. Geological resource is the broadest pool. Recoverable material in an operating host process is smaller. Installed separation and purification capacity is smaller again. Output meeting Bloom’s purity and particle specifications is a further subset. Material with export authorization and a qualified logistics route is narrower. Uncommitted volume that Bloom can actually purchase is the smallest and economically relevant pool. Capacity at any upstream layer should not be treated as available supply at the downstream layer.
USGS estimates global 2025 scandium oxide production at approximately 80 metric tons, consumption at approximately 60 metric tons and capacity above 90 metric tons. China was the leading producer. The United States commercially mined or recovered no scandium in 2025, imported approximately 4 metric tons and remained 100 percent import-dependent. Reported 2021-2024 U.S. oxide imports were attributed 89 percent to Japan and 11 percent to China, but USGS explicitly excludes scandium contained in value-added intermediates and finished products. A Japanese or Thai customs origin can therefore coexist with Chinese upstream material.
The Hunan supplier data illustrate both China’s importance and the unreliability of market statistics. The supplier claims a 40-metric-ton high-purity facility, approximately 150 metric tons of Chinese capacity and approximately 90 metric tons of Chinese 2024 output. It estimates only approximately 13 metric tons of overseas production, consisting principally of 7.5 metric tons from a Philippine-Japanese route, approximately 3 metric tons from Canada and approximately 2.5 metric tons from Russia and other sources. The supplier’s claimed 90 metric tons of Chinese output exceeds USGS’s approximately 80-metric-ton estimate for the entire world. The discrepancy likely reflects inclusion of hydroxide, oxalate and other intermediates converted to oxide equivalent, nameplate capacity, lower-purity material or different reporting periods. It demonstrates that supply estimates with apparent precision should be discounted.
China can simultaneously experience domestic oversupply and create a Western shortage. Chinese suppliers have reported rapid capacity expansion, weak domestic demand and price competition. However, material available within China is not automatically exportable to a U.S. fuel-cell customer. Export licensing, end-user disclosure, product classification, defense sensitivities, bilateral relations and government discretion segment the market. A low Chinese domestic price is therefore compatible with a much higher price or limited availability for secure, traceable Western supply.
EXISTING NON-CHINA SUPPLY
The most established non-China route is the Philippine-Japanese supply chain operated by a major Japanese mining and refining company. A nickel and cobalt high-pressure-acid-leach facility in the Philippines produces a scandium intermediate, which is refined into scandium oxide in Japan. The project was designed for approximately 7.5 metric tons of annual Sc2O3-equivalent capacity and was supported by a long-term agreement with a major U.S. fuel-cell manufacturer. The manufacturer was not named, but Bloom is the economically logical candidate. This route is commercially credible and likely strategically important to Bloom, but its scale is modest relative to a 2-5 GW production case.
The Canadian titanium-process route is also commercially established but small. USGS cites planned expansion to approximately 9 metric tons annually. More than 6,000 kg is intended for delivery to the U.S. Defense Logistics Agency over 5 years, and the Canadian government has also entered an offtake arrangement. These commitments reduce the volume that should be assumed to be freely available to Bloom. Even complete access to 9 metric tons would support only approximately 0.2-0.6 GW of production at 15-40 metric tons per GW.
Russian and other former Soviet supply may contribute several metric tons, but availability to a U.S. strategic-power manufacturer is exposed to sanctions, logistics, financial restrictions and geopolitical risk. It should not receive full credit in a secure-supply model. The same principle applies to nominal output that cannot meet Bloom’s qualification standards or cannot be traced to an acceptable upstream origin.
Publicly visible operating non-China supply therefore appears closer to 10-15 metric tons than to the several hundred metric tons referenced by Bloom. Bloom may have proprietary arrangements and recovery technologies not visible publicly, but those arrangements would need to bridge a very large gap between currently visible output and the 25 GW claim.
