The operating story has genuinely arrived: a first $1B quarter, 37% product gross margins, $226M of operating cash flow, and a doubled full-year guide. But a $67B market cap is not underwritten by $3.1B of trailing revenue — it is underwritten by a ~$20B company-defined backlog against $492M of audited remaining performance obligations, a 5 GW capacity ambition, and a supply chain a short seller says runs through China. That gap is the entire debate.
Bloom Energy has stopped being a story stock about fuel cells and become an actual industrial company with an AI-power franchise. Q2 revenue crossed $1 billion for the first time, product revenue tripled, non-GAAP product gross margin hit 37.2%, and the business generated $226M of operating cash and roughly $175M of free cash flow. Management raised the full-year guide to $3.9–4.2B — a doubling of 2025 — with $800–900M of non-GAAP operating income. On execution, there is very little to criticise.
The problem is not the business. It is the price and what the price requires. At ~$230 the equity is worth about $67B against $3.1B of trailing revenue — roughly 22x EV/sales and ~85x the midpoint of this year's own non-GAAP EPS guide. Nothing in the reported numbers supports that. What supports it is a set of forward claims: a ~$20B "contracted backlog," a Brookfield financing framework expanded to $25 billion, an Oracle master agreement for "up to 2.8 GW," and a path from ~1 GW of manufacturing capacity to 2 GW this year and 5 GW eventually.
Each of those numbers is softer than it reads. The audited remaining performance obligation behind the $20B backlog was $492M at June 30 — a ~40x gap, and Bloom's own 10-K says backlog counts revenue from equipment purchased or used by "a financier or an end customer" and "reflects anticipated ITC and other tax incentives." The $25B Brookfield figure is a financing shelf, not orders, and Brookfield vehicles are entities Bloom partly owns; related-party revenue was ~50% of Q1 revenue. Oracle's 2.8 GW is only 1.2 GW contracted — and the expansion was announced four days after Bloom issued Oracle a ~$400M warrant, which now sits on the balance sheet as a $306.5M "customer consideration asset." Management's own disclosure alongside the MiTAC release put AI-infrastructure contracted capacity at roughly 250 MW across about two dozen customers.
Our view: Neutral. We center base fair value near $215 — essentially at the market — with an honest range of $175–260, because this is a genuinely bimodal security. If the backlog converts and Fremont gets to 5 GW, the stock is worth far more than today. If conversion slips, gas-turbine queues normalise, or the scandium supply-chain allegations prove material, the fundamental support sits closer to $90. We would not be short a company compounding this fast. We would also not pay 85x for a forward number whose audited base is $492M.
| Metric | Q2-26 | Q1-26 / Q2-25 |
|---|---|---|
| Total revenue | $1,065.4M | $751.1M / $401.2M (+165.5% YoY) |
| Product revenue | $935.4M | $653.3M / $296.6M (+215.4%) |
| Service revenue | $69.0M | $61.9M / $54.4M |
| Installation revenue | $51.0M | $25.9M / $37.4M |
| Electricity revenue | $10.0M | $9.9M / $12.8M |
| Non-GAAP gross margin | 34.3% | 31.5% / 28.2% (+604 bps) |
| — Product GM (non-GAAP) | 37.2% | 35.3% / 34.3% |
| — Installation GM (non-GAAP) | (1.4)% | (29.4)% / (0.1)% |
| GAAP operating income | $182.2M | $72.2M / $(3.5)M |
| Non-GAAP operating income | $239.6M | $129.7M / $28.6M |
| Adjusted EBITDA | $253.4M | $143.0M / $41.2M |
| GAAP diluted EPS | $0.62 | $0.23 / $(0.18) |
| Non-GAAP diluted EPS | $0.78 | $0.44 / $0.10 · BEAT (~$0.40 est) |
| Operating cash flow | +$226.4M | +$73.6M / $(213.1)M |
| Capex | $51.6M | $26.2M / $7.2M |
| FY26 guide (raised) | $3.9–4.2B rev | ~34% non-GAAP GM · $800–900M op inc · $2.55–2.85 EPS |
This is a clean beat on every line that matters, and the cash-flow swing — from $(213)M a year ago to $+226M — is the single most important datapoint in the release. Bloom spent a decade burning cash; it now self-funds. Cash and equivalents were $2.67B against $2.48B of recourse debt, so the balance sheet is roughly net-cash. The one blemish: installation still runs at a negative gross margin, and G&A grew 56% YoY.
Bloom's re-rating is built on a sequence of very large announced agreements. Read them for what they legally are, not what the headline number implies.
| Counterparty | Headline | What it actually is |
|---|---|---|
| Brookfield | $25B (Jun 30, 2026; up 5x from $5B Oct 2025) | Financing framework, not orders. Vehicles Bloom co-owns; equity-method JVs. Related-party revenue was $373.3M of $751.1M in Q1-26, then $2.8M in Q2-26. |
| Oracle / OCI | Up to 2.8 GW (Apr 13, 2026) | Only ~1.2 GW contracted. Announced 4 days after a ~$400M warrant to Oracle; $306.5M now carried as a customer consideration asset. |
| AEP | 1 GW / $2.65B, 20-yr offtake (Jan 8, 2026) | Real and binding; ~40% of product backlog per Jefferies. But AEP's own decks moved Cheyenne in-service from "no later than 2028" to "no later than 2030." |
| Nebius | Up to $2.6B / 10 yrs, 300 MW (Aug 12, 2026) | Newest and largest neocloud win. Vineland NJ site reportedly received city stop-work orders Aug 6 and Aug 10 (short-seller sourced; unconfirmed by either party). |
| Equinix | >100 MW across 19 IBX sites | The most credible reference account — 75 MW operating, 30 MW under construction, from a 1 MW pilot in 2015. |
| MiTAC | Islanded microgrid, Fremont CA (Aug 6, 2026) | Small, but disclosed alongside the useful reality check: ~250 MW of contracted AI-infrastructure capacity across ~two dozen customers. |
| Crusoe / Tallgrass | 900 MW behind-the-meter (lost) | Paused June 9, 2026 "at the request of our customer"; Bloomberg reported Crusoe was pushed aside after Google raised cost and timeline concerns. |
Note the verb. Validated and approved is not ordered. No direct Microsoft, Google, Meta or Amazon contract has been announced. That is the correct thing to watch: the first named hyperscaler purchase order would do more for this thesis than another framework agreement.
This is the crux of the entire valuation, so it deserves its own section.
| Marketed "contracted backlog" (company-defined, unaudited) | ~$20B |
| — of which product backlog | ~$6B |
| — of which service backlog (5–20 yrs, terminable annually) | ~$14B |
| Remaining performance obligations (ASC 606, audited, 6/30/26) | ~$492M |
| Implied gap — total / product only | ~40x / ~14x |
| Deferred revenue + customer deposits (12/31/25 → 6/30/26) | $143.8M → $445.0M |
| Contract assets (12/31/25 → 6/30/26) | $241M → $428M |
| Disclosed AI-infrastructure contracted capacity | ~250 MW |
Bloom's 10-K defines backlog as revenue attributable to commitments for the purchase or use of Energy Servers "by a financier or an end customer," and states that it "reflects anticipated ITC and other tax incentives." Three things are therefore inside a number marketed as customer demand: offtake arrangements where no one committed to buy equipment; commitments from financing vehicles Bloom co-owns; and expected federal tax credits. For contrast, GE Vernova defines its $150B backlog as RPO.
To be fair to the bulls, the balance sheet does corroborate real forward demand — deferred revenue and customer deposits tripled in six months, and contract assets nearly doubled. That is cash and contractual progress, not a slide. But it corroborates something on the order of hundreds of millions, not twenty billion. The honest position is that the true forward book sits somewhere between $492M and $20B, and nobody outside the company can currently locate it. A quarterly backlog-to-RPO reconciliation would resolve this in one page; management has not provided one.
| Reference price (Aug 13–14, web-sourced) | ~$230 |
| 52-week range | $40.56 – $351.28 |
| 1-year price change | ≈ +422% |
| Market cap / enterprise value | ≈ $67B / ≈ $67B |
| Shares outstanding (6/30/26) / diluted (Q2) | 293.4M / 323.3M |
| Share count YoY | ≈ +25% |
| Cash & equivalents / recourse debt | $2.67B / $2.48B |
| FY26 guide — revenue | $3.9–4.2B (≈ +100%) |
| FY26 guide — non-GAAP EPS | $2.55–2.85 |
| FY26 fwd P/E at ~$230 (midpoint $2.70) | ≈ 85x |
| FY27 consensus non-GAAP EPS (range) | ≈ $4.50 ($2.64–$7.35) |
| FY27 fwd P/E | ≈ 51x |
| EV / FY26 revenue | ≈ 16.5x |
| Short interest | ≈ 6.9% of shares out; ~1.2 days to cover |
| Beta (5-yr) | ≈ 3.8 |
| Street consensus (29 analysts) | Buy · avg PT ≈ $275 |
| Street PT dispersion | $176 (Wells Fargo) – $346 (JPMorgan) |
| Manufacturing capacity | ~1 GW today → 2 GW target end-2026 (~$100M capex); 5 GW long-term ambition |
You cannot value Bloom off trailing numbers — at 22x EV/sales nothing works. The only defensible approach is to underwrite a forward earnings level, apply a multiple appropriate to the confidence you have in it, and discount back. The distribution is unusually wide because the input that matters most (backlog conversion) is unauditable from outside.
| Scenario | Basis | Implied Value |
|---|---|---|
| Bear 25% | Backlog converts far slower than marketed; gas-turbine queues normalise post-2029 and Bloom's time-to-power premium compresses; one or more anchor projects (Nebius, Cheyenne, Jupiter) slips again or fails on permits. Scandium sourcing and/or the securities litigation forces a disclosure reset. FY28 EPS ~$2.90 on decelerating growth; multiple de-rates to ~25x. | $85–100 |
| Base 45% | Fremont reaches 2 GW on schedule and ~3.5 GW by 2029. Revenue compounds ~35% through 2028; non-GAAP operating margin widens toward 25%. FY27 non-GAAP EPS near consensus ~$4.50, FY28 ~$5.75. A ~45x multiple on FY27 discounted back. Growth is real but 45% of it is already in the price. | $195–235 |
| Bull 30% | The backlog is directionally honest. Capacity reaches 5 GW by 2029, revenue clears $13B at ~28% non-GAAP operating margin, FY29 EPS ~$9. Bloom becomes the default onsite-power standard for AI and re-rates as an infrastructure compounder rather than a cyclical. Add S&P 500 inclusion. | $330–370 |
Probability-weighting those (25/45/30 on midpoints of $92.5 / $215 / $350) gives roughly $225 — within a few percent of the market. That is the honest conclusion: Bloom is priced approximately correctly for an extremely wide distribution. We set base fair value at ~$215 with a $175–260 range. The Street's own dispersion — $176 to $346, nearly 2x — is telling you the same thing, and it is a direct read on how much of the thesis rests on a number no auditor has signed.
What would move us to Constructive: a backlog-to-RPO reconciliation, a named hyperscaler purchase order, or evidence Fremont is actually running at a 2 GW rate. What would move us to Cautious: a second anchor-project slip, product gross margin rolling below 35%, or an adverse development on the scandium disclosure.
| Q3 2026 print (late Oct) | Does the $3.9–4.2B guide hold? Watch product GM and installation losses |
| Backlog-to-RPO reconciliation | The single disclosure that would settle the debate |
| Named hyperscaler purchase order | Converts "validated and approved" into revenue |
| Fremont 2 GW milestone (target end-2026) | Proof the ramp is physical, not slideware |
| Securities class action — lead plaintiff deadline Sept 28, 2026 | Discovery could force scandium sourcing disclosure |
| Nebius Vineland permits / Project Jupiter air permit | Permitting is now the binding constraint, not manufacturing |
| AEP Cheyenne in-service updates | ~40% of product backlog reportedly traces here |
| §48E domestic-content thresholds (45%→50%→55%) | Collides directly with the China-sourcing question |
| Gas-turbine lead times (GEV, Mitsubishi, Siemens) | Bloom's moat is a queue; watch it shorten |
| S&P 500 rebalance reviews | Passed over June 2026; still a live flow catalyst |
Bloom Energy has earned its seat at the AI-power table: a $1.07B quarter, 37% product margins, and $226M of operating cash flow are not story-stock numbers. But at ~$230 you are paying 85x this year's own guide for a forward book whose audited component is $492M against a $20B headline. We are Neutral with a ~$215 base and a deliberately wide $175–260 range. The bull case is worth $350 and the bear case is worth $90; both are live, and the difference between them is a disclosure the company has not yet made. Own it small if you own it, demand the backlog reconciliation, and confirm the live tick on your Schwab/TOS feed — the reference price here is web-sourced.