We value BE at $137 a share. The report's $217.77 price (Oct 22, 2026) was 59% above fair value: Avoid.
Restated Oct 26, 2026 (method case-spread-2026-10-24); the report's original value was $137, Avoid.
- Five-year return
- 0.2%
- a year, at the report's price
- Return
- -6.4%
- vs SPY
- -6.5%
- vs benchmark
- -6.5%
Close of Oct 22, 2026 to close of Oct 26, 2026
What would change the view
- Volume growth in 2028-2031 about 15 points a year faster than our central path (the optimistic world, 7 GW in 2031): the price is justified
- A 2027 revenue guide midpoint above $6.6B, our full 2027 central, given Bloom's habit of low first guides: our central is too low
- Product GAAP gross margin above 41% for a full year: the price premium is lasting rather than fading after 2028
- Product GAAP gross margin below 33% for a full year, or price per kilowatt falling more than 5% a year: the pessimistic world
- Turbine lead times back to about two years before 2029, or a large cloud capex cut: the pessimistic world
Catalysts
- Q4 2026 results (tentative date), the 2027 guide and the year-end product backlog. We expect Q4 revenue of $1.70B (range $1.35-1.96B) and a 2027 revenue guide midpoint near $5.6B; each $1B of 2027 revenue above our $6.6B is worth about $28 a share
- The large cloud companies' 2027 capital spending guides with their Q4 results (late January to early February 2027; we use 2027-02-05). A cut is our pessimistic world
- Oracle's Project Jupiter in New Mexico: whether the 2028 start holds after the September 2026 force majeure notice (press). A delay past 2028 moves 2027 toward 1.75 GW, -$23 a share
Next results preview
Results expected Feb 4, 2027. Street: FMP earnings revenueEstimated, captured Oct 22, 2026, 7:40 PM PDT.
| Metric | Ours | Our range | Street |
|---|---|---|---|
| Revenue ($B) | 1.7 | 1.35 to 1.96 | 1.23 |
| Diluted EPS ($, non-GAAP) | 1.37 | 1.04 to 1.61 | none |
| Product gross margin (GAAP) | 0.38 | 0.36 to 0.4 | none |
| Operating cash flow ($M) | 250 | -100 to 600 | none |
| 2027 revenue guide midpoint ($B) | 5.6 | 4.8 to 6.6 | none |
| Product backlog at year-end ($B) | 11 | 8 to 16 | none |
This text argues the original call. The current figures are at the top.
Avoid: Bloom is now a profitable AI power supplier, but at $217.77 it trades 59% above our $137 fair value. It sells fuel cells that make electricity from natural gas on the customer's own site. AI data centres buy them because the grid cannot connect new load for years and gas turbines are sold out, while Bloom delivered Oracle's Utah site in under two months (management, Q3 2026 call). Revenue more than doubled in Q3 2026 (+138% to $1,236M), product margins rose and Bloom raised its 2026 revenue guide for the third time, to $4.4-4.6B from $3.1-3.3B in February.
Our central case has volume rising from about 1.3 GW in 2026 to 4.2 GW in 2031 and revenue from $4.5B to $13.5B, with operating margins after stock pay near 27%. Bloom holds its price through 2029, while GE Vernova, the largest turbine maker, is sold out (its 2026-07-22 call), and cuts it 3% a year after. Even that is worth $137, because the growth slows to 9% by 2031 and the business then earns returns closer to an industrial equipment maker's. Most of the value is the exit: $110 of the $137 is the 2031 exit price brought back to today, $18 is the cash earned on the way and $9 is today's net cash and tax losses. The price needs about 6.4 GW in 2031 instead of 4.2, close to our optimistic world.
Terms. The central case is our best estimate; the pessimistic and optimistic cases are coherent alternative worlds we weight 25% each. GW is gigawatts of fuel-cell capacity shipped and recognised as revenue in a year; Bloom does not publish it, so we infer it from revenue at about $2,950 per kilowatt (from AEP's reported order, press). Time to power is how fast a customer gets electricity after ordering. Operating margin after stock pay is GAAP-like operating profit over revenue. Fade is our long fade: returns on new investment glide from today's level to a mature level over 15 years. Outside view is the price the 2031 operating profit would fetch at the median enterprise-value-to-operating-profit multiple of five power-equipment makers.
1. What Bloom sells and what limits it
Bloom makes solid-oxide fuel cells in Fremont, California ("Energy Servers"). They run on natural gas (or hydrogen), make power without combustion and emit little besides carbon dioxide, which makes local air permits easier than for turbines or engines (management). Revenue in Q3 2026 was product 87%, installation 7%, service 5% and legacy electricity 1% (10-Q). Service is a long maintenance contract on each installed server.
Who pays. Large buyers often do not pay directly: Brookfield's funds own the power projects and buy the servers (a $5B framework in August 2025, raised to $25B in June 2026, management, Q2 2026 call); those Brookfield joint ventures, a related party, were 26% of Q3 2026 revenue, and one other customer was 43% (10-Q note 1). Nebius's New Jersey site is financed by Oaktree, MUFG and Morgan Stanley (management, Q2 2026 call). So Bloom's customers' ability to raise project money is part of Bloom's demand.
Who the end customers are (management unless noted): Oracle (an agreement for up to 2.8 GW, 1.2 GW contracted, April 2026, press; Project Jupiter in New Mexico, up to 2.45 GW, Q1 2026 call); AEP, a utility, which agreed in November 2024 to buy up to 1 GW (10-K), exercised a further 900 MW option worth about $2.65B (Reuters via TradingView, press) and is using 700 MW to supply Google in Wyoming with carbon capture (Q3 2026 call); Nebius, 300 MW in New Jersey; Aligned Data Centers, 2 GW in Pennsylvania (September 2026, press). All the large US cloud companies and more than a dozen smaller AI cloud companies have approved Bloom's product (Q2 2026 call).
What limits it. First, factory capacity: Bloom planned 2 GW a year by the end of 2026 (10-K), its buildings can hold more than 5 GW, and it has leased more space in Fremont (10-Q). Second, gas supply and permits at each site: Oracle sent a force majeure notice on Jupiter in September 2026 after a gas pipeline slipped (Bloomberg via TradingView, press), then named Bloom primary power in an El Paso Electric agreement (press). New York and other places have proposed moratoriums on data centres (10-Q risk factors). Third, competition once turbines and the grid catch up: Bloom sells speed, and speed is worth less when the queue is short.
The AI thesis in plain words. AI data centres need gigawatts now; the grid adds them over five to ten years and turbine makers are booked for years, so on-site power that arrives in months is worth a premium. Bloom is the only fuel-cell maker shipping at that scale, so it captures most of that premium for as long as the shortage lasts. Our central case assumes the shortage holds price through 2029 and fades after. The thesis is wrong if any of these happens: the large cloud companies cut capital spending; turbine lead times fall to about two years before 2030, or the grid speeds up interconnection; Bloom's price per kilowatt falls more than 5% a year; or its large projects slip for gas or permits, as Jupiter has threatened.
2. The Q3 2026 results against the bar
We read the release, slides, 10-Q and the whole call, prepared remarks and Q&A (assumptions.md lists what was read and how far). The five things that matter:
- Growth beat Bloom's own outlook again. Revenue $1,235.8M (+138%), product $1,073.6M (+179%) (10-Q). The 2026 guide rose to $4.4-4.6B, its third raise this year; each raise lifted what was left of the year by 16% to 45% (preview.txt). In 2025 Bloom held its $1.65-1.85B guide through July and delivered $2,024M (releases).
- Margins rose with scale. Product GAAP gross margin 37.9% against 35% a year earlier; non-GAAP operating margin 23.1% against 8.9%, as operating cost grew 26% on revenue up 138% (10-Q, CFO on the call). GAAP operating income $225.4M (18.2%), GAAP diluted EPS $0.78, non-GAAP $0.94 (release). Service gross margin 17.8% GAAP. Management says pricing "remains strong" (CFO).
- Cash went out in the quarter. Operating cash flow was -$27.1M and capex $153.5M, so free cash flow was about -$181M; nine-month operating cash flow is +$272.9M against $231M of capex (10-Q). Contract assets ($858.6M) and inventory ($927.7M) carry the growth; the CFO says contract assets bill on milestones and turn to cash. Cash is $2.49B, against $2.5B of 0% convertible notes due November 2030.
- Backlog: no number, but commitments say a lot. Management says backlog "is continuing to expand faster than our revenue" and will give the number with the 2027 guide in February. Bloom has committed to $5.8B of purchases under long-term supply agreements, $2.6B of it in 2027 (10-Q note 12), about 1.4 times what 2026's product costs imply for materials (our assumption of a 75% materials share). Firm unsatisfied obligations for product and installation are $661M, up from $394M at December (10-Q note 3). At end-2025 product backlog was about $6B, up about 2.5 times, and total backlog about $20B (Q4 2025 release).
- Concentration and the AI pipeline. Two customers were 43% and 26% of the quarter, and 97% of revenue was in the US (10-Q). New this quarter (management): AEP's 700 MW for Google with carbon capture; first units shipping to Texas; the first block of Nebius's 300 MW ready; a US Army project; utilities buying in front of the meter. The CEO says the 5 GW milestone is coming into view and Bloom is preparing beyond it, adding capacity "in an analog dial", not in steps.
3. Our forecast and value
Alpha, our quantitative model, ranked Bloom 22nd of 702 on 2026-10-22 (score 97.0, tier Strong: growth A, expectations A, momentum B, quality B, valuation E). It ranks on growth, expectations and momentum; this report values the business, and the two disagree here.
The model (model.py, output.txt) builds product revenue from volume times price, and product cost from volume times unit cost, so the margin is a result. Inputs and sources are in assumptions.md (the blind record; section 9 lists the one input changed since); the main ones:
| 2026 | 2027 | 2028 | 2029 | 2030 | 2031 | |
|---|---|---|---|---|---|---|
| Volume (GW) | 1.33 | 2.00 | 2.70 | 3.29 | 3.79 | 4.19 |
| Price ($/kW) | 2,950 | 2,950 | 2,950 | 2,950 | 2,862 | 2,776 |
| Revenue ($M) | 4,500 | 6,604 | 8,925 | 10,970 | 12,398 | 13,499 |
| Product gross margin | 37.0% | 38.6% | 39.5% | 41.3% | 41.3% | 41.3% |
| Operating margin after stock pay | 17.4% | 22.3% | 24.8% | 27.2% | 27.1% | 26.7% |
| Free cash flow, modelled full year ($M) | 309 | 328 | 1,016 | 1,690 | 2,140 | 2,536 |
2026 free cash flow is the model's full year; the value counts only Q4 ($130M, our assumption), since cash at 2026-09-30 is counted at face. Reported nine months plus that Q4 would be $172M.
Where we differ from management and history. 2026 is management's guide midpoint; 2027's +47% revenue is our assumption, backed by the 2027 purchase commitments. The 2026 Q4 in the model is the guide's $1.45B, while our preview below expects $1.70B: a beat in Q4 2026 barely moves the value, because 2026 enters only through Q4 cash. The product margin of 41.3% from 2029 is 3.4 points above Q3 2026's 37.9%, because unit cost falls 1.5-3% a year while price is held to 2029. The operating margin of about 27% after stock pay is above anything Bloom has reported (18.2% GAAP in Q3 2026); it comes from operating cost growing about half as fast as revenue, as it did this year. We read it as scale, not a peak, but it is above any year Bloom has reported; the dated support is GE Vernova's sold-out position through 2030 and the CFO's 2026-10-22 remarks that pricing is strong and cost programs are still reaching cost of goods. Ending 2031 at Q3 2026's levels instead (product margin 37.9%, non-GAAP operating margin 23.1%; ours are 41.3% and 29.6%) gives $98, $38 less; the product margin alone is $14 (c6.py, c6.txt).
Stock pay ($234M in 2026) is charged as a cost, and the share count is today's 310.9M diluted shares; the 2030 notes add 12.8M shares only in the optimistic case, whose exit price clears the $195 conversion price, and are repaid in cash in the other two. Tax is 23% on operating profit; the $1.45B of federal losses left after 2026 is worth $352M today with the research credits, counted once.
The exit. At the end of 2031 the central's long fade is worth 15.9 times 2032 operating profit after tax, about 13.2 times 2031 pre-tax operating profit, and the outside view 20.1 times operating profit (the median of Caterpillar, Cummins, Generac, Vertiv and Eaton at their year-ends; GE Vernova has too few years). Their mean, after the $2.5B notes, is $185 a share. We used the method's default throughout, with no changes. The two halves differ by 55% ($225 against $145 at the exit): the fade lets growth glide from 8% to 3% and returns from today's high level to 25%; the peers' year-end multiples include the recent re-rating of power equipment on AI demand (our inference). A terminal growth near 6.8% instead of 3% would close the gap (model auditor, on the blind model). Each half is a fair reading; the method averages them.
Cases. Pessimistic ($31): the data-centre build slows and turbines catch up from 2029; volume 1.75 GW in 2027, falling in 2028 and growing 6% after; price falls 3% to 8% a year from 2028; 2031 revenue $5.8B at an 11.7% operating margin. This is the bust we cannot date. Optimistic ($303): Bloom keeps its time-to-power lead and ships 7 GW in 2031; revenue $22.9B at a 33.9% margin.
Sensitivities (central, output.txt): volume growth 2028-31 ±10 points a year -$48 / +$58; 2027 volume 1.75 / 2.30 GW -$23 / +$25; price 2029-31 -3 / +2 points a year -$36 / +$23; unit cost ±2 points a year -$27 / +$23; operating cost ±5 points a year -$13 / +$9; growth after 2031 4% / 12% -$9 / +$9; Bloom's own beta (Blume 2.86) instead of the peers' 1.27 -$48, to $89.
4. Market expectations and how the price reacted
Bloom closed at $217.77 on 2026-10-22 before the release and traded at $211.27 after hours, about 3% lower, despite the revenue beat and the raised guide (Robinhood). The Street's average target is $296, $267 when brought back one year at our cost of equity (FMP, 2026-10-22).
To see what the price expects, we ran the Street's consensus for each year 2027-2030 through our own model (street_path.py, street_path.txt; FMP consensus via research-check, 2026-10-22). The Street stops at 2030, so 2031 either keeps our growth or carries the Street's 2030 growth on:
| Step | Value a share |
|---|---|
| Our central | $137 |
| The Street's revenue each year ($6.9B in 2027 to $15.6B in 2030), our unit costs | $181 (+$44) |
| Plus the Street's EPS each year, read untaxed as Bloom's own non-GAAP EPS is | $154 (-$27) |
| Same, read taxed at our 23% | $204 (+$24) |
| Untaxed or taxed, with 2031 carrying the Street's 2030 growth on | $164 or $219 |
Bloom's non-GAAP EPS carries almost no tax while its losses last (Q3 2026 adjusted net profit $302.8M against non-GAAP operating income $285.7M, release), so the untaxed reading is the closer match for the next two years; we did not see the Street's models (inference). Either way the volume step is the largest, and on the untaxed reading the Street's margins are below ours. The price, and more so the targets, then need growth well past 2030 or an exit multiple far above the peers' 20.1x: the model auditor found the discounted target needs about 32x operating profit on the Street's path (blind model).
5. Where we differ from the market
We agree with the market that Bloom is a real winner of the AI power shortage. We differ on how big and how long. Our central triples volume by 2031 and holds price to 2029; the price needs about 6.4 GW in 2031, or the Street's revenue plus a higher exit multiple than power equipment has carried. The difference from the Street's revenue alone is $44 a share; the rest is duration.
What settles it: the 2027 guide and year-end backlog on about 2027-02-04 (a guide midpoint above $6.6B points to the Street's revenue), Bloom's capacity plan past 5 GW, and the product margin through 2027 (holding above 41% with rising volume says the premium is lasting). If volume heads to 6 GW by 2030, our optimistic world ($303) is the better description and we would revise.
6. Why the other side could be right
Bulls. Bloom has beaten its own outlook every time for two years, by 16% to 54% of the remaining period; our central's 2027 (+47%) may be low. The CEO is preparing beyond 5 GW, and customers such as Oracle, AEP and Aligned have each signed for gigawatts; if half of that converts, our optimistic case (7 GW in 2031) is in reach. Data-centre power may stay short for a decade, and carbon capture and 800-volt direct current could open markets we give nothing for. The optimistic path is worth $303.
Bears. One customer was 43% of the quarter and Brookfield's funds 26%: the business is a handful of AI projects financed by others. Q3 burned cash and contract assets are large; the short seller Hunterbrook challenged Bloom's scandium supply and accounting in July 2026 (Bloom rejected it, 8-K 2026-07-09), and a securities class action followed. Oracle has already signalled it may defer payments on Jupiter. Turbine makers are adding capacity, and if the queue shortens Bloom's price premium goes with it. Our pessimistic case, $31, is cash plus a business whose profits shrink after 2028.
7. Next results: Q4 2026, tentatively 2027-02-04 (not confirmed by the company)
From preview.py (preview.txt), written before the market pass. Bloom guides only the full year, so Q4 is the guide less the first nine months.
| Forecast | Central | 80% range | Reason |
|---|---|---|---|
| Revenue ($B) | 1.70 | 1.35-1.96 | guide-implied $1.45B lifted by half Bloom's median surprise of 35% |
| Diluted EPS ($, non-GAAP) | 1.37 | 1.04-1.61 | Q3's $0.94 plus 30% of the added revenue |
| Product gross margin (GAAP) | 38.0% | 36.0-40.0% | Q3 37.9%, nine months 36.5% |
| Operating cash flow ($M) | 250 | -100 to 600 | contract assets turning to cash, as the CFO said |
| 2027 revenue guide midpoint ($B) | 5.6 | 4.8-6.6 | our 2027 $6.6B less Bloom's habit of a low first guide |
| Product backlog at year-end ($B) | 11 | 8-16 | about $6B a year earlier, growing faster than revenue |
The 2027 guide matters most. Each extra 0.1 GW of 2027 volume (about $300M of revenue) is worth about $9 a share in the model, so a 2027 outcome $1B of revenue above our $6.6B is worth about $28 a share (our inference from the 2027 volume sensitivity).
8. Dates to watch, and what would change the view
- 2027-02-04 (tentative): Q4 2026 results, the 2027 guide and the year-end backlog. A 2027 guide midpoint above about $6.6B, or backlog above $16B, moves us toward the optimistic case.
- Oracle's Project Jupiter, through 2027. A formal delay past 2028 or a cancelled order moves us toward the pessimistic case (2027 at 1.75 GW, -$23).
- The large cloud companies' 2027 capital spending guides (late January to early February 2027). A cut is the pessimistic world.
- Turbine lead times. GE Vernova's and Siemens Energy's order books and slot dates; a return to two-year deliveries before 2030 means price falls sooner than our central.
- Bloom's price per kilowatt and product margin. Product GAAP margin below 33% for a full year says price is falling faster than cost; above 41% says the premium is lasting.
9. What changed from the blind value, and the review
The blind model (commit 06116d2) gave $124. Two changes since, both made after the price was known, take it to $137:
| Change | Evidence, dated and independent of the price | Before | After | Effect |
|---|---|---|---|---|
| Upkeep capex at least covers depreciation | The odds simulation drew shrinking worlds where the model's depreciation, which never retires assets, made net investment negative and the method refused to value them (alpha research odds, 2026-10-22) |
central $124.10, pessimistic $31.35 | $124.10, $31.07 | central $0; pessimistic -$0.28 |
| Price per kilowatt held through 2029, cuts from 2030 instead of 2029 | GE Vernova, 2026-07-22 Q2 call (company transcript, primary, read for this point): "mostly sold out through '30" and more than half of its 2031 slots sold by the end of 2026. Found by the opposite-side reviewer; we read the transcript ourselves | price path 0, 0, -3%, -3%, -3% | 0, 0, 0, -3%, -3% | central +$12.56, to $136.66; exit $166 to $185 |
The second change moves the value toward the price. We made it because our blind reason for the 2029 cut was turbine supply catching up, and the largest turbine maker's own dated statement puts that a year later. The pessimistic world keeps its undated bust from 2028.
Reviews. research-eval was run by us (the analyst) on commit 3e8b88304; the two outside
reviewers were Codex (codex exec, reasoning effort xhigh, read-only), a model auditor and an
opposite-side advocate, on the same commit, which carried the $124 model. Their dollar figures
below are on that model. Both ran their scripts in memory: the read-only sandbox refused to
save them, so their figures are not rerunnable from saved files; the advocate's source is in
its review, and we reproduced the GE Vernova fact and the Street fit ourselves.
| Finding | From | Effect | Answer |
|---|---|---|---|
| Turbine supply is sold out through 2030, so the 2029 price cut is early | advocate | +$12.56 | Fixed (table above) |
| Our Street fit used only 2027 and 2030 endpoints and stretched both improvements into 2031 (flag 17) | auditor, advocate | $228 overstated; exact years give $204 (auditor) to $222 (advocate) | Fixed: street_path.py now fits every year and shows both tax readings (section 4) |
| Street EPS read as taxed; the non-GAAP definition and interest are unclear | auditor | -$9.70 on the Street path | Fixed: the untaxed reading is now shown and is the closer match to Bloom's own non-GAAP EPS |
| "12.2x pre-tax" for the fade omitted the next year's growth | advocate | none | Fixed: about 13.2x |
| A Buy world: volume 12 points a year faster and flat prices after 2028 gives $243 | advocate | n/a | Disclosed: it is close to our optimistic world (6.1 GW and a 44.8% product margin in 2031); it needs both a larger factory than Bloom has committed to and no price give-back, and at Bloom's own beta it is $154 |
| The single flip: volume growth 16 points a year faster (13 after our fix) | advocate | n/a | Agreed; in the Price needs line |
| Margins end above anything Bloom has reported (research-eval C6) | research-eval | -$38 if 2031 ends at Q3 2026's margins; -$14 for product margin alone | Disclosed (section 3), not changed: the scale argument and GE Vernova's sold-out position are dated support |
| Working capital stops growing in the fade | auditor | -$2.09 | Disclosed: a method-wide property, raised in process.md |
| The 2030 notes are repaid at the valuation exit, not their November 2030 maturity | auditor | -$0.57 | Disclosed |
| Research credits ($47.9M) are not recognised as deferred tax assets; losses valued at once, not when used | auditor | -$0.15 and -$0.07 | Disclosed |
| Options counted without exercise proceeds; basic shares from the quarter-end, not the cover; finance leases omitted | auditor | +$0.25, -$0.05, -$0.02 | Disclosed; together with the rows above about -$2.7, under 2% |
| Unvested awards in the share count may overlap with future stock pay | auditor | +$0.78 if so | Rejected for now: not established; future stock pay is a share of future operating cost, not today's awards |
| Payouts at mid-year rather than year-end | auditor | -$0.81 at year-end | Disclosed: the method's convention |
| The Oracle warrant asset's amortisation has no explicit schedule | auditor | not sized | Disclosed: it sits in the margins |
| Q4 2026 is $1.45B in the model and $1.70B in the preview; the 2026 free cash flow row differs from reported nine months plus Q4 | advocate | none on value | Fixed: labelled in section 3 |
| Flags 3, 4, 17 (research-check) | both | 3: no double charge found by either reviewer; 4: the gap is the fade's growth path against the peers' multiple (section 3); 17: section 4 |
Not checked: neither reviewer reran the odds simulation; no technical-analysis read was done.
10. Sources
Primary, in data/, fetched from EDGAR 2026-10-22/23: Q3 2026 earnings release and
supplemental slides (8-K 2026-10-22), Q3 2026 10-Q, Q2 2026 release, slides and 10-Q, Q1 2026
release, Q4 2025 release, 10-K 2025, Q3 2025 10-Q, releases for Q4 2024 and Q1-Q3 2025, and
8-Ks of 2025-10-30, 2025-11-04, 2025-12-19, 2026-04-13, 2026-04-27 and 2026-07-09.
Secondhand: call transcripts for Q4 2025 to Q3 2026 (Quartr via TradingView). Press:
TradingView news on AEP, Oracle, Aligned, Hitachi and Hunterbrook (to 2026-10-22). Data: FMP
statements, peer betas and year-end multiples (2026-10-22). Method: case-exit-2026-10-22
(src/alpha/services/analyst/research_report/method.py), unchanged.
Details
- Report date
- Oct 22, 2026
- Price source
- Robinhood, official close 2026-10-22 (SIP listing-exchange close); after the Q3 release it traded at $211.27 after hours (Robinhood, 2026-10-23 02:20 UTC), Oct 22, 2026, 1:00 PM PDT
- Bear / bull
- $27.38 / $365
- Exit price
- $185 (on Dec 31, 2031)
- Confidence
- 15%
- Price against our scenarios
- The price is justified in 20-50% of our scenarios4,000 simulated outcomes, each drawing the uncertain drivers from ranges we chose: our judgement, not a measured chance
- Worlds
- 3 stories about the businessEach outcome first picks one of these stories by its weight, then draws that story's own driver ranges; the answer above is their mix.
- World: base (50%)
- the base drivers. The price is justified in under 20% of its scenarios; middle outcome $128, 10th to 90th $64.99 to $257Weight: Our judgment, matching the central case's 50% weight.
- World: bear_lean (25%)
- AI data-centre build slows and gas turbines and engines catch up from 2029. The price is justified in under 20% of its scenarios; middle outcome $21.20, 10th to 90th $7.74 to $40.69Weight: Our judgment, matching the pessimistic case's 25% weight.
- World: bull_lean (25%)
- Bloom keeps its time-to-power lead and scales past 5 GW by 2030. The price is justified in over 50% of its scenarios; middle outcome $279, 10th to 90th $155 to $523Weight: Our judgment, matching the optimistic case's 25% weight.
- What moves the answer
- Volume growth 2028-2031, points a year above or below the central path (35%, 22%, 15%, 10.5%) 55%; Price per kilowatt 2029-2031, points a year beyond the central path (flat to 2029, then -3% a year) 50%; 2027 Energy Server volume recognised, gigawatts 49%; NOPAT growth in the year after 2031 43%; Mature return on new fixed and intangible capital 24%How much of the spread in value each driver explains on its own; drivers that move together share credit, so the shares need not add to 100%.
- Leaning cases
- 2 other views we ranThe same simulation with some ranges moved to another view, to show how much the answer depends on our ranges.
- Case: bear_lean
- AI data-centre build slows and gas turbines and engines catch up from 2029. The price is justified in under 20% of this case's scenarios; middle outcome $21.20, 10th to 90th $7.74 to $40.69
- Case: bull_lean
- Bloom keeps its time-to-power lead and scales past 5 GW by 2030. The price is justified in over 50% of this case's scenarios; middle outcome $279, 10th to 90th $155 to $523
- Cost of equity (original)
- 10.6%
- Original figures
- fair value $137 (bear $31, bull $303), exit $185; Avoid
- Street
- average $296, median $303, low $176, high $351; 16 targets in the last quarter; FMP price-target-consensus, Oct 22, 2026, 7:40 PM PDT
- Restated
- Oct 26, 2026, method case-spread-2026-10-24; forecast.json valued by method.py at market inputs 2026-10-01
- Cost of equity calibration
- +0.0 points
- Calibration source
- none: the 1-point cut set 2026-10-16 to pull the batch toward the Street was removed on 2026-10-18 (Dong); the cost of equity is risk-free plus beta times the premium
- View rule
- strong buy when the price is at or below 0.80x fair value; buy when at or below 0.90x fair value; avoid when fair value is below 0.85x the price; otherwise hold
- Blind step
- Partly unblinded; the model was committed in 06116d2 before the record
- Method
- As of Oct 22, 2026, built without: Mature return holds own capital
Original call, Oct 22, 2026 (before the Oct 26, 2026 restatement)
We rate Bloom Energy an Avoid: at $217.77 the price is 59% above our fair value of $137 a share (pessimistic $31, optimistic $303). Bloom is a real and now profitable AI power supplier: Q3 2026 revenue rose 138% and it raised its 2026 guide for the third time, to $4.4-4.6B. Our central already has volume tripling to 4.2 GW and revenue reaching $13.5B in 2031 at about a 27% operating margin, holding price until turbines catch up in 2030. The price needs the Street's path: revenue near $15.6B in 2030 and product margins near 45%, or volume growth 13 points a year faster than ours.
BE — Bloom Energy Corporation — research view (market: Avoid)
As of 2026-10-22 · partly unblinded: the brief gave the 2026-10-22 close and a Street EPS
estimate, and contract headlines showed price targets, before the blind commit
(assumptions.md lists seven exposures). Blind commit 06116d2; the blind value was $124,
review moved it to $137 (section 9).
Fair value today $137 per diluted share, discounting at a 10.64% cost of equity.
Weighting the pessimistic, central and optimistic cases 25/50/25 (our
weights, not calibrated) gives $152.
Price $217.77, Robinhood, official close 2026-10-22 (SIP) → fair value is
37% below the price. After the Q3 release it traded at $211.27 after
hours (Robinhood, 2026-10-23 02:20 UTC).
Five-year exit price $185 a share on 2031-12-31: the average of two estimates, one from
letting Bloom's returns fade toward its cost of capital ($145) and one
from what makers of turbines, engines and electrical gear have traded at
($225, at 20.1x operating profit). Fading over 10 or 20 years instead
gives $179 or $191.
Five-year return 0.2% a year from $217.77 to the central exit, cash paid out included,
against a 10.64% cost of equity
Bear / bull today $31 / $303: our pessimistic and optimistic cases, each one consistent
business outcome.
Confidence About 15%: our judgment of the share of simulated values within 20% of
$137 (interpolated from the simulation's percentiles in odds.json;
subjective, not calibrated)
Moat Our judgment: narrow and time-limited. Bloom is close to the only
fuel-cell supplier data centres use at scale ("very high 90s" share of
fuel cells, management), and it can deliver power in months where the
grid and gas turbines take years. The lead is speed and a proven
factory, not a lock-in: it lasts as long as the power shortage does.
Market expects More volume than ours for longer. The Street's revenue each year to 2030
($15.6B in 2030; ours $12.4B) is worth $181 in our model; its EPS adds
little once read on Bloom's untaxed non-GAAP basis. The rest of the price
needs growth past 2031 or an exit multiple above the peers' 20.1x
(street_path.txt)
Price needs Volume growth in 2028-31 13 points a year faster than our central, alone
(6.4 GW in 2031 against our 4.2 GW); no other single driver reaches the
price inside its range (odds.json)
Odds in our scenarios P(price justified): 20-50%, the share of our simulated scenarios in which
the value is at or above the price. Eight in ten simulated values fall
between $18 and $345. Subjective, not calibrated (odds.json)
Key driver Volume growth after 2027: 10 points a year slower or faster in 2028-31 is
-$48 / +$58 a share. Then price per kilowatt in 2029-31: 3 points a year
lower is -$36; and unit cost, 2 points a year either way, is about ±$25.
Street Average 12-month target $296 (median $303, range $176-351); 64 targets
over the last year averaged $210, the last quarter's 16 $264 and the last
month's 6 $281 (FMP via research-check, 2026-10-22). An expectations
reference, not validation