Crossroads Capital: Short Bloom Energy [In-Depth Report]

Crossroads Capital is short shares of Bloom Energy Corp (NYSE:BE).

We are short Bloom Energy (BE) and related derivatives. Below we outline the reasons why we think the stock is not properly pricing in certain risks.

See also our interview with Crossroads Capital in the Q3 2025 Special Edition issue of Hidden Value Stocks.

[Disclaimer: In the last 24 hours, Hunterbrook put out a Bloom Energy short thesis strikingly close to ours. Any resemblance is coincidental. The view is far from consensus, but it’s good to know others see it too, and it prompted us to publish a bit earlier than intended.]

Bloom published a response on scandium yesterday to Hunterbrook. It’s written largely in qualitative terms: scandium is “tiny” in each cell, “abundant” in the earth, sourced from “multiple countries,” with “several hundred tons” available from titanium waste. Those are reasonable things to say; our aim here is simply to put a number next to each one, based on our own analysis, so the discussion can happen in the same units. Here are some parts of Bloom’s response we find lacking:

• Scandium per cell is “a sprinkle of salt.”: At ~3.2 g of scandium oxide per cell and ~2,330 cells in a 100 kW server, that’s ~7.5 kg per server, roughly 75 kg per megawatt, all estimated from Bloom’s patents. Triangulating Bloom’s shipments in 2025 in GW (~0.5), we estimate a total draw of 30–36 tonnes a year, on our math, roughly half of the ~60 tonnes the world consumes, consistent with Bloom’s own statement that it is the largest single consumer.
• Scandium is “abundant… more than lead.”: True, but the whole world still produces only ~80 tonnes a year from every source combined (USGS).
• “Several hundred tons [of scandium] can be produced” from titanium waste: That figure matches what’s contained in the waste, not what’s recovered, and roughly 55–60% of the world’s titanium-dioxide production sits in China; actual recovery from Western titanium is ~3 tonnes a year (Rio Tinto, USGS). Bloom’s own scandium extraction patent (US 9,102,999) recovers scandium from titanium waste acid at 15–20 mg/L, and both titanium dioxide routes (sulfate and chloride) leave scandium in their acid waste at similar levels (10–140 ppm). However, the scandium in those chloride streams is mostly sold off as low-value iron chloride coagulant or landfilled rather than recovered. That recovery-plus-refining-plus-qualification chain is built almost nowhere outside China, which draws ~80% of its scandium from titanium-pigment waste. Western titanium recovers ~3 tonnes a year (Rio Tinto, USGS).
• Supply comes from “multiple countries… no single country”: Bloom’s own 10-K describes its suppliers as “primarily located in Asia”; we estimate roughly 90% of U.S. scandium-oxide imports come from Japan (our read of U.S. trade data; USGS lists imports as “mostly from Japan, China, and the Philippines”).
• The supply chain “can support up to 25 GW per year.”: At Bloom’s own loading, we calculate that as ~1,500 tons of scandium oxide a year at 60 kg of scandium per MW, roughly 19x current world production.

Bloom answered Hunterbrook with words; we analyzed Bloom with numbers, laid out in full below. We’d welcome a similarly quantitative response from management to the ten questions we pose in the introduction below.

I) Introduction

The AI build-out has been booming for nearly everything that touches a data center, and for much of it the enthusiasm is deserved: compute must be powered, cooled, and connected, and the firms selling those picks and shovels have watched their earnings prospects and valuations soar. But as adoption accelerates, undercapitalized sub-sectors—memory, substrates, physical power—are having trouble absorbing what amounts to a demand shock, and thus are attracting supernormal investor expectations and outsized share price moves. Solutions to alleviate these critical constraints that otherwise would never have seen the light of day are now not only being evaluated but are being purchased at volumes that exceed their own capacity to deliver.

The investors swept up in this are best described as “Bottleneck Investors”: chasing constraints in the AI supply chain in the belief that there is multi-year visibility or an evolving moat in the new regime, with little understanding of the technology they are underwriting or the real limitations of what they are buying. While some bottlenecks may have an elongated upcycle and thus have further to go, a demand shock doesn’t typically lend itself to smooth sailing after it is alleviated.

There is one company whose bottleneck-induced upcycle is running straight into a supply wall, one that caps the revenue growth and earnings generation the market is pricing in, and when that becomes visible, the repricing should be severe.

Bloom Energy (BE) has been swept up as one of the bottleneck winners in the power sector. It makes on-site power that does not produce emissions thanks to its solid-oxide fuel cell (SOFC) technology, previously viewed as a science experiment. Now, given AI-driven demand and grid permitting bureaucracy, fuel cells look like an appealing solution despite sitting at the top end of the power-cost curve, with names like Oracle and Nebius booking gigawatts of capacity. The stock has run with additional fuel from retail following the “brand name” investors of this cycle who have what we call “Situational Awareness.”

However, Bloom’s entire growth story rests on a silvery-white metal, scandium oxide, that almost no one has heard of, that only two firms in the world are going to mine on purpose, and that the company would plainly rather not discuss. It discloses no volumes, no costs, and no supply contracts—and the reason that silence matters is where scandium actually comes from: roughly 80% of world supply is Chinese, and the non-Chinese sources are small and largely spoken for. That dependence is about to matter as a paused Chinese export control could retrigger as soon as November, forcing Bloom or its supply partners to apply for an extraterritorial license just to keep the material flowing. Demand is spiking now, while the supply response that could ease it isn’t a meaningful factor until 2029. We are calling this “Physical Awareness.”

We believe the company’s near-term growth is supported by a temporary surplus of scandium that is rapidly diminishing, along with legacy low-cost supply contracts signed years ago that are set to reset at higher prices, all while the necessary supply response to provide even half of Bloom’s aspirational capacity is a 2029-2030 event.

The company does not give true long-term guidance but states a 5 GW nameplate capacity is achievable in a few years’ time and the Street assumes this is as good as revenue, expecting ~4.5 GW in booked fuel cell shipments by 2030. On our numbers, there isn’t a way for Bloom to ship 2.5 GW in 2030, much less 5 GW.

This isn’t an academic supply-and-demand exercise. Once you hold this view on scandium, it throws management’s own statements into a different and more troubling light. The CEO flatly told investors that “there is no China supply chain for us”—but given our supply work, we do not see how this statement could be true. He also told analysts that “to our board or to our vendors, we have not provided any long-term guidance,” because “the horizon at best is six months… nobody has visibility past that.” Yet this is the same company whose equity is valued on multi-gigawatt “capacity” it won’t commit to shipping despite multi-year contracts with Oracle. In our opinion, a management team that won’t commit to a guidance number beyond a six-month horizon—not for investors, not for vendors, not even for its own board—hasn’t earned the benefit of the doubt on filling that 5 GW of capacity, or on much else.

Beyond these physical constraints and management’s apparently (given our research) untrue statements and lack of clear longer-term guidance, the edge the Street is capitalizing as a durable, multi-year advantage—time-to-power—is really a shot clock as cheaper behind-the-meter power arrives at scale to fill the gas-turbine backlog that created Bloom’s temporary opening in the first place.

Once investors trade Situational Awareness for Physical Awareness, we believe the overvaluation we see in the shares is likely to be corrected.

Before we get deep into the thesis, if we could get Bloom to answer the following ten questions, we could properly price the risk we see in the shares:

1. What is BE’s SOFC scandium-equivalent intensity per kW?
2. How many annual scandium-equivalent kilograms purchased?
3. What is the contracted scandium price per kg?
4. What are the remaining scandium supply contract terms?
5. How much scandium do you have in inventory?
6. What share of the scandium in your shipped fuel cells was mined or recovered in China, no matter where it was later refined or who you bought it from?
7. If China reinforces extraterritorial re-export restrictions on scandium on November 10, 2026, will you or your supply partners have to apply for a secondary export license from the Chinese government?
8. Western titanium waste yielded only ~3 tonnes of recovered scandium in 2025 (Rio Tinto). What Western Ti waste extraction projects are being built, how much recovered scandium will they add, and when do they come online?
9. Walk us through how 300 tons per annum of non-Chinese scandium will be sourced to produce 5 GW of fuel cells when total Western supply is, at best, 120 tons per annum in 2030 and the defense industry is likely to take a third of that at prices 6x what Bloom appears to pay today?
10. Can you provide guidance on gigawatt (GW) shipments and long-term revenue and earnings, rather than on manufacturing capacity (which investors often conflate with actual shipments and revenue)?

A key point to keep in mind is that Bloom buys scandium oxide (Sc₂O₃) and then creates scandia-stabilized zirconia (ScSZ) powder, which is the end-use material. Scandium oxide (Sc₂O₃) is only 10-12% of that powder by weight, the rest is zirconia and ceria (or a proprietary mix). Every volume, price, and supply figure in this thesis is stated on a scandium-oxide-equivalent (Sc₂O₃) basis, the contained scandium oxide, not the total powder mass, because that is the exact form in which scandium is mined, refined, traded, and priced. Any response by the company that quotes a total ScSZ powder tonnage (which would be roughly 10x larger) or a per-kg-of-powder price is not comparable to these figures and does not address the constraint: the limiting number is the contained scandium oxide, however it is packaged.

What was once a niche material known only to specialized mining analysts, little more than a footnote in the SOFC bill of materials, has become the true bottleneck of Bloom’s growth story, with the Street modeling deployment at 2-3x what is physically achievable even in a best-case scenario. On top of that, Bloom has obscured its scandium sourcing so effectively that the Chinese exposure has gone all but unnoticed. Across the entire Tegus library, Bloom and scandium appear together just twice, and the one analyst who raised it on a call let it drop without pressing.