Bloom Energy's window is someone else's backlog

The word in the annual report

When Bloom Energy's board granted its chief executive a new performance award in June, the company's filing noted that its market capitalisation had grown "from approximately $5b as of December 31, 2024 to approximately $79b as of June 15, 2026" ^[1]. Four months earlier, the annual report had described what Bloom was selling into, in a two-sentence passage on competition:

major turbine OEMs are currently supply constrained and are typically prioritizing capacity expansion only after fulfilling existing commitments. This supply constraint, combined with the long lead times for traditional grid infrastructure, creates a market window for alternative distributed generation technologies that can deliver capacity more rapidly ^[2].

"Window" is the company's own word, and an unusually candid one. The filing names what sits on either side of it: a turbine industry that cannot build fast enough and a grid that cannot connect fast enough. Neither is a property of Bloom's product; what the product contributes, the same section says, is that it "can be deployed rapidly, giving us a competitive advantage when customers have an urgent need for power" ^[2]. On its own account, the window Bloom sells into is someone else's backlog.

For now the window is wide open. Revenue for the quarter ended June 30, 2026, Bloom's fiscal second quarter, was $1,065.4 million, up 165.5%, and full-year guidance of $3.9 billion to $4.2 billion represents "100% year-over-year growth at the midpoint" ^[3]. The shares closed at $262.87 on September 28, a market capitalisation of about $77.4 billion on this site's record ^[4], which by our arithmetic applies the July 10-Q share count ^[15].

So the useful question is not whether Bloom is winning. It is how much of what it wins will still be Bloom's if turbines and grid connections stop being scarce. The filings show what the window has bought, how little of it is locked in, and what Bloom is paying for the rest.

What the window is made of

Speed. Bloom says that in certain project scopes its onsite power can be available "within approximately 90 days" ^[2]. It cites a Lawrence Berkeley National Laboratory finding that the time from a grid interconnection request to commercial operation "has more than doubled to 55 months in 2024 from less than two years in 2008" ^[2].

The risk factors describe the ordinary case: "Generally, it takes between six to twelve months or more from the entry into a sales contract until the installation of our products," with the "availability of gas infrastructure" among the reasons ^[2]. Both clocks are far shorter than 55 months, though the grid's runs from interconnection request to operation and Bloom's from contract to installation.

Permitting and scale. In most states that require air permits above emission thresholds, permits have "only been required for larger Energy Server system installations" ^[2]. The 10-K said Bloom believed doubling its Fremont factory to 2 gigawatts of annual run rate by the end of 2026 "remains on schedule and within our planned budget" ^[2]; neither 2026 10-Q repeats a gigawatt figure ^[21][15].

What the window is not, on the company's own account, is a first-cost advantage. In February, per an unofficial published transcript of the earnings call (the source of every call quote here), the chief executive said "the customers who evaluate the entire value proposition will choose us", but conceded: "Customers who just look at first cost, surely, if you can find it and if you can install it in a short amount of time, you'll be able to use a combined-cycle gas turbine" ^[6]. The window lives inside the conditional in that sentence.

The evidence it is still open

GE Vernova said its gas power equipment backlog plus slot reservation agreements "grew from 100 to 116 GW" in its second quarter, and that it expects "at least 125 GW by year-end 2026" ^[7]. A quarter earlier it had expected at least 110 GW ^[8], so by our arithmetic the year-end expectation rose 15 GW in three months. Its January annual report said that "due to increasing demand exceeding available capacity for products and services that supply the electrical power industry, we face growing competition from emerging threats" ^[9].

Buyers describe the same gap. American Electric Power offers fuel cells to data centers "while it completes grid infrastructure enhancements to accommodate demand" ^[10], a bridge in all but name. Nebius said in May that its first Bloom project, with 328 MW of installed capacity, was expected to be operational this year, eliminating "the need for reciprocating engines at the site" ^[11].

One case runs the other way. WHYY reported that New Jersey regulators fined the developer of a 300-megawatt Vineland data center more than $1 million for running 62 natural gas-powered generators without a permit; the developer said the engines were a temporary measure until its Bloom fuel cells are operational ^[13]. The buyer needed a bridge to the bridge.

What narrows it

The first force runs on GE Vernova's schedule. It says it is "on track to deliver 20 GW of annual gas turbine output in the third quarter of 2026, with 24 GW in 2028", and is "implementing actions to produce 30 GW in 2030" ^[7]; by our arithmetic, 50% above the 2026 rate. Its annual report warns that "capacity expansions and related commitments may outpace realized demand" and that "slot reservation agreements may not result in orders" ^[9]. The queue that holds the window open could shorten.

The second is the grid. Among the things that "could adversely affect AI data centers' demand for our solutions", Bloom's 10-K lists "improved grid interconnection timelines" ^[2]. Reforms are under way: a Southwest Power Pool pathway for generation serving large loads, approved by the Federal Energy Regulatory Commission in January 2026, promises interconnection studies "completed within 90 days" ^[2]. A 90-day study is not 90 days to power, and the filing says shorter queues would help Bloom's own grid-connected systems too, "increasing the value of onsite generation solutions" ^[2].

The third is every other fast source of power, and the chief executive concedes it. Asked in July about engine and turbine capacity, he said: "given the huge supply-demand gap right now, I think every single technology that can provide power quickly for the next few years is going to have a play" ^[14]. Looking past that, he argued that what matters is "the total cost of power to token revenue" ^[14]. "For the next few years" reads as his estimate of the window's width; no filing yet tests the second claim.

What the window has bought

At its strongest, the record shows a business being built, not only a queue being exploited.

Service, which earns after the sale, stopped losing money at the gross-profit line: service gross margin was 10% in 2025 against (1)% in 2024 ^[2], and 18.7% in the June quarter, or 22.0% on Bloom's non-GAAP basis, which excludes stock-based compensation and some other items ^[3]. Maintenance contracts run "from 5 to 20 years, subject to termination for convenience on an annual basis", and "nearly every customer has renewed historically" ^[2].

GAAP operating income has stopped being marginal: $72.8 million in 2025, about 3.6% of total revenue of $2,024.0 million by our arithmetic ^[2], then $182.2 million in the June quarter alone ^[3]. Net income to common stockholders went from a 2025 loss of $88.4 million ^[2] to $70.7 million in the March quarter and $196.3 million in June ^[3].

June-quarter operating cash flow was $226.4 million ^[3]. Customer deposits and deferred revenue added $211.7 million, but receivables took $101.4 million and contract assets $123.0 million; across all the working-capital lines, by our arithmetic, working capital used about $50 million, so the cash came from net income of $198.9 million and non-cash charges ^[3].

Qualification may be the most durable asset, because an approval does not depend on a turbine shortage. In the July release the chief executive said "all the major US hyperscalers and over a dozen US neoclouds, AI labs, and colocation data center operators have validated and approved our power solutions" ^[3]. Asked on the call how many were using the systems and how many sat in backlog or pipeline, he said all were users, booked orders or holders of "definitive agreements", but: "We are not going to split that up" ^[14].

How much is locked in

Backlog. With its 2025 results Bloom reported a "Total current backlog of ~$20 billion; current product backlog of ~$6 billion" ^[16]. That is a management measure: product backlog "reflects anticipated ITC and other tax incentives" (the 30% section 48E investment tax credit), and service backlog counts contracts terminable every year ^[2]. The GAAP counterpart, unsatisfied performance obligations, was $394.4 million, primarily for product and installation, plus $25.0 million, mainly for service, at December 31, about $419 million on our addition ^[2]; at June 30 the two were $442.4 million and $51.7 million, about $494 million ^[15]. Bloom publishes no bridge between the measures, and its July release gives no backlog figure ^[3]. On the July call the chief executive said backlog was "growing at a faster pace than revenue" ^[14], without a number.

Ceilings. Under a master services agreement, Bloom says, Oracle "intends to procure up to 2.8 gigawatts", of which "an initial 1.2 GW of capacity has been contracted" ^[5]. Brookfield "increases its framework to finance" power projects to $25 billion ^[17]; the 10-K called the original framework "prospective", for projects "that meet agreed investment and contractual criteria" ^[2]. Nebius's own regulatory filing is more guarded than its press release: it will pay monthly fees "of up to $2.6 billion in the aggregate" for capacity "expected to come on in three phases", each with a 10-year supply term, "subject to commencement of the Capacity Agreement upon satisfaction of certain conditions" ^[18].

The firmest is AEP, whose 2024 purchase agreement with Bloom carried an option to reach up to 1 gigawatt in total ^[10]. In January an AEP subsidiary signed "an unconditional purchase agreement to acquire a substantial portion of its option" for a facility near Cheyenne, Wyoming, "for approximately $2.65 billion" ^[19]. The facility's offtake "remains subject to specified conditions"; if they are not met in December 2026, AEP is "entitled to additional consideration" ^[20]. Bloom's own annual and quarterly reports do not name the Wyoming order ^[2][21][15].

The customer is often a financier. Bloom counts concentration by "the contractual counterparty", which "in certain transactions may be a project‑finance affiliate rather than the ultimate end user of the products" ^[2]. In July the chief financial officer said most customers "pay over time rather than own", so Bloom brings in "a financier" ^[14]. On that basis one related party was about 43% of 2025 revenue ^[2]; Bloom's 2025 sales to the Brookfield joint ventures, $809.8 million of product and $52.3 million of installation ^[2], come to about 43% of the $2,024.0 million total by our arithmetic. One related party was about 50% of the March quarter ^[21]. In the June quarter, per a 10-Q/A that corrected transposed periods, one customer that is not a related party was about 73% of revenue; for the half year, two customers were about 44% and 21%, the second a related party ^[22]. Related-party revenue fell from $373.3 million in the March quarter to $2.8 million in June ^[3].

Those figures allow one deduction. By our arithmetic, 73% of the June quarter's $1,065.4 million ^[3] is about 43% of half-year revenue of $1,816.4 million ^[15], so that customer must be the half year's 44% customer ^[22], leaving it about $22 million of March-quarter revenue, at most about $36 million allowing for rounding, out of $751.1 million ^[3]. The record quarter's largest customer was nearly absent three months earlier, and no filing names it.

Management reads this as timing. In February the then acting principal financial officer said Bloom did "not have oversized concentration to any one customer", speaking of backlog ^[6]; in July the chief financial officer said concentration "reflects delivery timing, not the composition of our backlog" ^[14]. No filing breaks out backlog by customer.

Hunterbrook Media, which disclosed that its affiliate Hunterbrook Capital was short the stock, argued in July that "nearly half of Bloom's 2025 revenue came from buyers Bloom helped create" and, as its lead charge, that Bloom relies on Chinese scandium, "the rare earth at the core of each Bloom fuel cell" ^[23]. Bloom "categorically rejects the Report's claims regarding the Company's financial results and accounting" and said "our supply is not dependent on China" ^[24]. The tax credit is exposed too: the 10-K warns that failing rules on "material assistance" from "prohibited foreign entities" may cut it, and that financiers "may require us to indemnify them", though it expects no such limit ^[2]. A class action filed on July 30 pleads omissions about scandium sourcing ^[25]; its allegations are unproven.

What the conversion costs

Stock paid for a customer. In October 2025 Bloom agreed to issue Oracle a warrant for up to 3,531,073 shares at $113.28 ^[2]. Bloom's financial statements book it as consideration payable to a "customer's customer" ^[2][15]: Oracle is treated as the customer of the party Bloom sells to. The 10-Qs put its grant-date fair value at about $261.3 million ^[21] and later $251.6 million ^[15], without explaining the change. After Oracle exercised on May 1, with 248,798 extra shares worth $72.3 million to induce net settlement, the shares were worth $324.4 million in aggregate ^[15]. Only $17.9 million had come off revenue by June 30; the rest, $306.5 million by our arithmetic, sits as a Customer Consideration Asset of $91.0 million current and $215.5 million long-term, to be netted against revenue as Oracle-arrangement systems are delivered ^[15].

One-offs that happen to cancel. June-quarter product costs included "a $37.4 million recovery of previously paid import tariffs", only $5.0 million of it refunded by June 30 ^[15]. Against it, a "specific warranty reserve of $19.7 million for identified product issues" at March 31 ^[21] had grown to $58.3 million by June 30 ^[15], a $38.6 million build by our subtraction; neither 10-Q says which product is affected. GAAP product margin was 34.3% in the March quarter and 36.5% in June, on product revenue of $653.3 million and $935.4 million ^[3]. Excluding the tariff credit and the warranty charges from both, it was about 37.3% and about 36.6% by our arithmetic. The two items roughly offset, but neither is a run rate.

Cash advanced for a customer project. In May Bloom paid $50 million for option rights on a customer project, assigned in July to a Brookfield vehicle that is to pay Bloom $50 million on exercise; if it does not proceed, "Brookfield may put the option rights back to the original holder, Oracle", with Bloom recovering the $50 million ^[15]. Bloom expected to realize the asset "within the next three months" ^[15].

The equity. Shares outstanding went from 230,398,527 in February 2025 ^[26] to 294,527,346 in July 2026 ^[15], 27.8% more by our arithmetic. Most of the new shares retired older convertible notes: 42,407,945 in a 2025 induced conversion ^[2] and 6,676,924 on 2026 conversions ^[15], against 1,905,433 plus 248,798 issued to Oracle ^[15]. The $2.5 billion of 0% notes issued in November 2025 convert at $194.97 a share, or 5.1290 shares per $1,000: about 12.8 million shares by our arithmetic, up to 19,554,000 with make-whole adjustments ^[2]. The "just 48%" opex growth the chief financial officer cited ^[14] is the non-GAAP line, $125.7 million against $84.7 million ^[3], or 48.4% by our arithmetic; it leaves out stock-based compensation, whose total rose 86.9% year over year in the quarter, while GAAP operating expenses rose 56.7% ^[15].

Jupiter: where the window meets the ground

At up to 2.45 GW of Bloom fuel cells, replacing "previously planned gas turbines and diesel generators" ^[12], Oracle's Project Jupiter in New Mexico is the largest Bloom project named in these documents: by our arithmetic, 87.5% of Oracle's up-to-2.8 GW ^[5]. Neither company says whether any of it sits inside the contracted 1.2 GW.

New Mexico's environment department says the Bloom fuel cell microgrid proposed by Yucca Growth Infrastructure "will be a major source for Title V", the Clean Air Act program for large emitters; its notice gave a preliminary intent to issue the permit "on or before August 25, 2026" ^[27]. KRWG, which calls Yucca Growth Jupiter's developer, reported on September 18 that the state Supreme Court had lifted an August stay, but that the department "has yet to appoint a new hearing officer and set a new hearing date" ^[28].

On September 24 TechCrunch, crediting Bloomberg with the first report, said Oracle had sent a force majeure notice to the project's developer; according to Bloomberg's sources, Oracle was not seeking to exit, and the notice would let it delay payments "should the facility miss its 2028 target to come online" ^[29]. In its own reporting TechCrunch added that the gas pipeline for the site "has also been delayed nearly six months, to February 1, 2027, after regulators repeatedly denied permits" ^[29]. Reuters, citing a person familiar with the matter, said the notice cited "potential delays in securing power", that securing power "is Oracle's responsibility under the contract", and that Oracle "cannot terminate the lease under any circumstances" ^[30].

The counterparties played it down. Oracle told CNBC that "Project Jupiter remains on our planned schedule"; Blue Owl, whose unit received the notice, said it "does not change the financial commitments to this multi-year project"; and Bloom said on X, according to CNBC, that it remains "committed" to the contract and the "planned timeline" ^[31].

Bloom's shares fell 3.10% on September 24, rose 8.27% the next session and fell 8.95% on September 28 ^[32]. No primary document explains that last fall, which by our reading of the closes came after the shares had recovered to $288.70, above the $275.19 close before the news ^[32].

On these reports, Jupiter's constraint sits upstream of Bloom's factory ^[29][30]: a fuel cell campus needs gas, gas needs a pipeline, and both need permits. It is the 10-K's "availability of gas infrastructure" clause ^[2] at gigawatt scale.

What would show the window narrowing, or its gains locked in

Each is a disclosure to be filed or published, not a forecast.

• GE Vernova. Its next backlog-plus-reservations figure, against 116 GW at the end of June and "at least 125 GW" expected at year-end ^[7].
• Bloom's September-quarter 10-Q. Customer concentration; the $58.3 million warranty reserve ^[15]; unsatisfied performance obligations against the backlog; the $50 million option.
• Jupiter. The air permit decision, which TechCrunch said is due by November 23, and the February 1, 2027 pipeline date ^[29].
• AEP and Nebius. Whether the Wyoming offtake conditions are met in December ^[20], and whether the Nebius capacity agreement commences ^[18].

The 10-K chose the word window. On the filings through June, what could outlast it is a service line with an 18.7% gross margin ^[3] and approvals the chief executive says span the major US hyperscalers. What is enforceable is about $494 million of unsatisfied performance obligations ^[15]; the rest of the headline book is ceilings, frameworks and conditions.