Bloom Energy crossed a billion dollars in quarterly revenue for the first time, and every headline led with it. Revenue of $1,065.4 million against $401.2 million a year ago, product revenue up 215.4%, GAAP operating income of $182.2 million versus a loss, non-GAAP EPS of $0.78 against a consensus near $0.41. Guidance went from $3.4–3.8 billion to $3.9–4.2 billion. The stock, which had shed roughly 18% in the five sessions before the print and closed near $167, traded up about 11% after hours to $186. All of that is accurate and none of it is the interesting number.
The interesting number sits in the working capital section of the cash flow statement. Accrued warranty rose $39.4 million in the quarter, against product revenue of $935.4 million — an accrual equal to 4.22% of product sales. In the same quarter last year the figure was $1.7 million on $296.6 million, or 0.58%. In Q1 it was 2.81%. The balance-sheet line went from $20.0 million at year-end to $77.8 million in six months. Because that provision runs through cost of product revenue, it is already sitting inside the 36.5% GAAP product gross margin Bloom just reported. Hold the accrual rate flat at last year’s level and product margin prints somewhere near 40% instead. The change is net of claims paid, so the arithmetic is directional rather than exact, but the direction is unambiguous: Bloom is reserving against its installed base at a rate that has multiplied sevenfold in twelve months while reporting the best margins in company history.
There are two readings and the release does not distinguish between them. The generous one is that solid oxide stacks degrade on a known curve, the installed base is compounding faster than revenue, the newest high-density configurations have limited field history, and management is provisioning conservatively into a period when it cannot afford a reliability event with hyperscaler customers who have just standardized on the product. Related-party accrued warranty — the Brookfield joint venture assets — went from $0.8 million to $8.6 million, consistent with fleet growth rather than fleet trouble. The unkind reading is that something in the field is costing more than modeled. The service segment, historically a margin sink, printed 18.7% GAAP gross margin against 9.3% a year ago, which argues against a live problem. On balance the conservative reading is the better one, and it means the reported product margin understates the underlying unit economics. That is a bullish conclusion drawn from a line item that looks bearish at first glance, which is precisely why it is worth the attention it did not get.
Two other things in the release deserve more scrutiny than they will receive. First, a new asset appeared on the balance sheet from nothing: $91.0 million current and $215.5 million long-term of “customer consideration asset,” described in the footnotes as upfront share-based consideration payable to a customer’s customer. That is $306.5 million of Bloom equity handed to an end user to secure a relationship, and it explains most of the $576.6 million increase in additional paid-in capital during the half — the remainder is roughly $143 million of convertible notes converting into stock and about $100 million of ordinary stock compensation. Consideration payable to a customer is normally a reduction of transaction price. Bloom capitalized it, which means it is a deferred revenue or margin haircut rather than an avoided one, and the current portion alone is $91.0 million coming through over the next four quarters. It does not appear as a non-GAAP add-back, because it has barely started hitting the income statement. Second, related-party revenue was $373.3 million in Q1 — half the quarter — and $2.8 million in Q2. The billion-dollar quarter was almost entirely third-party. Whatever else can be said about the print, it is not circular.
The cash flow is real and that matters more than usual for a company that spent twenty-five years not generating any. Operating cash flow of $226.4 million breaks down as $198.9 million of net income plus roughly $277 million of net income and non-cash charges, less a $50 million working capital build. Deferred revenue and customer deposits grew $211.7 million in the quarter and now stand at $445.0 million against $143.8 million at year-end — a threefold increase that functions as the backlog disclosure Bloom does not provide. Contract assets rose to $428.3 million. Receivables and unbilled revenue together run about seventy days of sales, expanding but funded. The company ended with $2.69 billion of cash against $2.68 billion of total debt and financing obligations: a balance sheet that nets to almost exactly zero, with $2.47 billion of that debt being 2029 converts now deep enough in the money that the 36 million share gap between basic and diluted count already treats them as equity.
And the tax line is a $200.3 million pre-tax profit carrying a $1.5 million provision — an effective rate of 0.7%. The accumulated deficit is $3.72 billion. Bloom will not pay meaningful federal cash tax for years. Every dollar of operating income converts to shareholder earnings at close to par, which is a structural advantage over any of the industrial incumbents it competes against and one that no analyst model can extend indefinitely but every model must respect for the next several years.
The moat is not the fuel cell. Solid oxide electrical efficiency of roughly 60% beats a simple-cycle turbine handily and loses to combined cycle, which is a wash in the applications that matter. The moat is air permitting. Bloom’s servers convert gas electrochemically rather than by combustion, which puts them under minor-source thresholds for NOx, SOx and particulates. In the exact places where data centers want to be — Northern Virginia, Phoenix, the Dallas corridor, the Chicago suburbs — nonattainment status turns a gas turbine into a twelve-to-twenty-four-month permitting exercise with a real probability of denial. The blocked pipeline for Oracle’s New Mexico project earlier this month is the template. Bloom sells the one dispatchable onsite generation technology that clears the permit. Layer on qualification lock-in — the CEO’s claim that every major US hyperscaler and more than a dozen neoclouds and colocation operators have validated the platform is the power-sector version of a design win, and displacing an approved reference architecture requires re-qualification nobody has time for — plus turbine order books at GE Vernova, Siemens Energy and Mitsubishi that are booked into the late decade, and the position is genuinely defensible for the next two to three years. It is partly borrowed defensibility. If turbine capacity meaningfully loosens in 2028 and 2029, the permitting argument has to carry the whole moat by itself.
The capital efficiency is the underappreciated part of the competitive story. Bloom is annualizing north of $4.2 billion of revenue against $443 million of net property, plant and equipment — better than nine turns — and did it on $77.8 million of first-half capex. The Energy Server is assembled, not fabricated. Capacity expansion here costs a fraction of what a turbine OEM spends for equivalent output, which means the growth does not need to be financed with dilution or debt. It also means the constraint on Bloom is supply chain and labor, not plant.
Which brings the guidance into focus. Full-year revenue of $3.9–4.2 billion against first-half revenue of $1,816.4 million implies second-half quarters of $1,042 million to $1,192 million. Bloom just did $1,065 million, having grown 41.8% sequentially. The non-GAAP EPS guide of $2.55–2.85 against $1.22 booked implies $0.67 to $0.82 per quarter versus the $0.78 just delivered. The raise, stripped of its framing, embeds zero sequential growth at the low end and roughly 12% at the high end. Management is either sandbagging or bumping a capacity ceiling, and given a CEO describing demand as accelerating every quarter, it is very hard to argue it is demand.
The stock has run 386% in twelve months and something over 100% year to date, from a 52-week low of $32.52 to a high of $351.28, and still sits roughly 47% below that high even after the after-hours move. Market capitalization at $186 is about $54.5 billion, $60 billion fully diluted — 13.5 times the midpoint of 2026 revenue guidance and 69 times the midpoint of non-GAAP EPS guidance, with GAAP earnings a quarter lower because stock compensation runs at 4.9% of revenue. Street targets are wide: JPMorgan raised to $346 from $267 on July 21, the aggregate average across roughly nineteen analysts sits near $276–290, and the discounted cash flow crowd has fair value at $122.81 after itself doubling from $66.98. Insiders sold 32 times in six months with zero purchases, one officer alone moving 205,334 shares for about $41 million.
Base case is $190–240 over twelve months: guidance met, 2027 revenue in the $5.5–6.5 billion range, non-GAAP EPS of $4.00–4.50, a 50-to-55 multiple as the growth rate halves from triple digits. Bull case is $300–360, requiring the Brookfield framework — expanded from $5 billion to $25 billion — and the $1.7 billion Oaktree and IDF commitment to convert into shipped gigawatts, with 2027 EPS above $5. Bear case is $95–130, and it is a cohort derating rather than a company failure: Bloom now trades with Vertiv, GE Vernova, Talen and the rest of the AI power complex, not with Plug Power at $2.23 or FuelCell, and any hyperscaler capex digestion signal or credible turbine capacity relief takes 30 points off the multiple regardless of what Bloom ships.
The line to watch on the next print is accrued warranty. If it compounds another $40 million against sequentially flat revenue, the field is telling on itself and the margin story needs rebuilding. If the accrual rate reverts toward 1%, Q2’s product margin was understated by roughly 350 basis points and Bloom is a materially more profitable business than the one that just reported the best quarter in its history.