Bloom Energy

Bloom Energy (NYSE: BE) makes solid-oxide fuel cells — boxes that turn natural gas into electricity on-site, without a turbine, without a grid connection, and, the pitch goes, without the fight. That last part is the business. The Power Wall argues that the binding constraint on AI infrastructure has moved from money to megawatts, and from megawatts to permission — because a grid takes years to reach and a gas plant takes a hearing to stop. Bloom sells the way around both. Where IREN and the miner conversions sit on the scarce side of the wall by owning power that already exists, Bloom sits one layer down: it sells the generator you can put behind the meter when the wall will not move.

That makes it a second-order name here — not a company whose fortunes the AI build-out makes or breaks, but one the build-out reaches for when its first choice fails inspection. And in 2026, the first choice kept failing inspection.

The clarifying event has a date and a place: a multi-gigawatt AI campus in New Mexico called Project Jupiter, planned by Oracle. The plan called for gas turbines and backup diesel. The turbines did not clear. In July, The Information reported that permit hurdles had pushed Oracle off gas turbines and onto costlier fuel cells, adding billions to the cost of its data centers — the Power Wall as a line item. On its first-quarter 2026 earnings call, Bloom’s CEO put the same event from the winning side: Oracle had pivoted Jupiter to an up-to-2.45-gigawatt power block that would be, in his words, 100% Bloom, replacing the turbines and the diesel outright — one of the largest islanded microgrids in the world. He gave two reasons, neither of them price: residents’ concerns about air quality, water use, noise and electricity rates, and a site free of the grid, with more reliability and speed.

The Jupiter block was the headline, but the structure around it is the part worth reading, because it is the same structure that runs through this whole cycle. Two weeks earlier, Oracle had contracted 1.2 gigawatts of Bloom fuel cells and said it intends to procure up to 2.8 — days after receiving a warrant to buy $400 million of Bloom stock, which it exercised on May 1. The customer took equity in the supplier it depends on. That is the AI-capex loop — the pattern where Nvidia invests in CoreWeave and holds a right to invest in IREN, where the buyer and the funder are the same party — reappearing one layer beneath the GPUs, down in generation. When you see the customer take a stake in its power company, the power has become the scarce thing.

The quarter under it was then the strongest in Bloom’s history as a public company. On the same call it reported revenue of $751.1 million, up 130.4% year over year — its first quarter ever above 100% growth — with product revenue at an all-time high and non-GAAP operating income of $129.7 million against $13.2 million a year earlier. It raised full-year revenue guidance to $3.4–3.8 billion, roughly 80% growth at the midpoint (since raised twice, to $4.4–4.6 billion on October 22), and said its current manufacturing footprint can build 5 gigawatts of product a year. A company that spent two decades as a patient, money-losing bet on clean on-site power is suddenly the fast option, because the slow options stopped clearing.

Two cautions the record supports. First, Bloom runs on natural gas; its case is speed and siteability, not zero emissions, and the “clean” framing is relative to a diesel-and-turbine alternative, not to the grid. Second, every fuel-cell sale carries a 10-to-15-year service contract at a 100% attach rate — real annuity, but also a two-decade operating obligation booked against hardware sold today. The growth is real. So is the tail it commits to.

Bloom carries six edges in TEXXR’s knowledge graph as of October 27: three from a 2013 article on its Delaware factory, Oracle’s expansion in April, and two in October on American Electric Power’s Wyoming fuel-cell project. CoreWeave carries 294. That gap is not an oversight — it is the signal. Bloom is a twenty-year-old company that the technology press almost never covered, because nothing it did was a technology story. It became one the moment its product stopped being an energy purchase and started being the answer to a permitting problem. A handful of edges, arriving as the wall did, is a cleaner reading of why this name matters now than a hundred edges accumulated over a decade of being ignored.

Positioning: how much of BE is sold short, on the record brokers file twice a month. It lags by design — read it as where the crowd stood, not where it stands.

Observable signals in the coverage and filing record, not trade ideas:

• Whether the permit stays the moat. Bloom’s premium is the price of siteability — power a community will approve. In Hamilton Helmer’s terms that is close to a cornered resource, but a fragile one: it is cornered by policy, not physics. The interconnection-queue and permitting reforms that would free the grid for everyone else are exactly what would shrink the premium for going around it. Same falsifier as every name on this wall.
• Customer concentration. Oracle is the anchor, the equity holder, and the headline. How much of the backlog rides on one customer — versus the other hyperscalers, neoclouds and co-location providers management says make up well more than half of it — is the disclosed number to track.
• The gas question, and the narrative around it. Bloom’s siting case beats turbines on local objections. It does not beat them on carbon. If the opposition story that stopped the turbines widens to on-site gas generally, the escape route narrows. Watch whether “clean enough to permit” holds as the frame, or slips.
• Inference at the edge. Management’s own argument is that inference — bigger than training in total power, and sited closer to dense population — is where clean on-site generation matters most. Whether that demand shows up in the backlog, or stays a talking point, is in the filings.

Bloom Energy lives or dies by the same constraint as these names — small equal-weighted groups, one constraint each. Not investment products: see all categories.