BE · Forward model · The AI book is still small case
The The AI book is still small case, 20 quarters out
Each segment is projected from its own operating driver, rolled up into consolidated cash flow, and discounted back to a fair value per share. The assumptions below are editable — change them and every number on this page moves with them.
Bloom reports four revenue streams — Product, Installation, Service and Electricity — on the face of every income statement, so all four verticals carry REPORTED revenue and nothing is apportioned. Every quarter from 2023 Q1 to 2026 Q2 is a disclosed split except the three fourth quarters, which are the full-year segment figure less the nine-month year-to-date and are flagged estimated. The stream revenues sum exactly to reported consolidated revenue in thirteen of fourteen quarters; 2023 Q4 differs by $1 thousand, a rounding artifact of that subtraction. What is NOT disclosed, and is therefore ours: the end-market split (AI data centre, utility channel, colocation, C&I) is nowhere in the filings and is deliberately not modelled; megawatts shipped or accepted are disclosed in no period, so every megawatt here is back-solved from an assumed $3.0M/MW; the installed service fleet is assumed; and Bloom allocates no operating expense to a stream, so verticals carry gross margin only and all opex sits in the corporate layer — which is what makes 33.38% gross margin less 16.27% opex reconcile to the reported 17.11% operating margin. One material item the engine cannot express: the Oracle warrant is contra-revenue, not cost. Its $324.4M fair value is recognised as a reduction of revenue as the Oracle systems ship, and $306.5M of that still sits on the balance sheet as a customer consideration asset. The product line here is GROSS of it, so cumulative product revenue through the Oracle delivery window is overstated by up to $306.5M — roughly 8.5% of that tranche. Net cash of -$17.6M treats the $206.5M of financing obligations as debt; excluding them it is +$188.9M.
Set against the gigawatt headlines, the disclosed AI segment is about 250 MW across roughly two dozen customers — less than one quarter of shipments at the basis rate. This case holds the named frameworks flat and grows only the book Bloom has actually quantified, which is the shape the disclosure supports today rather than the shape the press releases imply.
Latest: $1.77B (2031Q2E)
| Period | Value |
|---|---|
| 2023Q1 | $275M |
| 2023Q2 | $301M |
| 2023Q3 | $400M |
| 2023Q4 | $357M |
| 2024Q1 | $235M |
| 2024Q2 | $336M |
| 2024Q3 | $330M |
| 2024Q4 | $572M |
| 2025Q1 | $326M |
| 2025Q2 | $401M |
| 2025Q3 | $519M |
| 2025Q4 | $778M |
| 2026Q1 | $751M |
| 2026Q2 | $1.07B |
| 2026Q3E | $1.12B |
| 2026Q4E | $1.17B |
| 2027Q1E | $1.22B |
| 2027Q2E | $1.27B |
| 2027Q3E | $1.31B |
| 2027Q4E | $1.36B |
| 2028Q1E | $1.40B |
| 2028Q2E | $1.44B |
| 2028Q3E | $1.48B |
| 2028Q4E | $1.51B |
| 2029Q1E | $1.54B |
| 2029Q2E | $1.58B |
| 2029Q3E | $1.61B |
| 2029Q4E | $1.63B |
| 2030Q1E | $1.66B |
| 2030Q2E | $1.69B |
| 2030Q3E | $1.71B |
| 2030Q4E | $1.73B |
| 2031Q1E | $1.75B |
| 2031Q2E | $1.77B |
What drives each segment
Product — Energy Servers
Capacity × utilisation × priceThe engine, and 88% of the basis quarter. Bloom sells solid-oxide fuel cell modules to whoever pays for the box — increasingly a financier, utility or strategic partner rather than the end user, who is buying power. What caps this line is not orders: management says every major US hyperscaler and a dozen-plus neoclouds, AI labs and colocation operators have validated the product, and that capacity is not the constraint. What caps it is how fast megawatts get built, shipped and accepted against customer site-ready dates. So the projection is a deployment schedule — quarterly deliverable megawatts, the share actually converted, and what a megawatt sells for — not a demand forecast.
Latest: $1.45B (2031Q2E)
| Period | Value |
|---|---|
| 2023Q1 | $194M |
| 2023Q2 | $215M |
| 2023Q3 | $305M |
| 2023Q4 | $262M |
| 2024Q1 | $153M |
| 2024Q2 | $226M |
| 2024Q3 | $234M |
| 2024Q4 | $472M |
| 2025Q1 | $212M |
| 2025Q2 | $297M |
| 2025Q3 | $384M |
| 2025Q4 | $638M |
| 2026Q1 | $653M |
| 2026Q2 | $935M |
| 2026Q3E | $982M |
| 2026Q4E | $1.03B |
| 2027Q1E | $1.07B |
| 2027Q2E | $1.10B |
| 2027Q3E | $1.14B |
| 2027Q4E | $1.17B |
| 2028Q1E | $1.21B |
| 2028Q2E | $1.24B |
| 2028Q3E | $1.26B |
| 2028Q4E | $1.29B |
| 2029Q1E | $1.31B |
| 2029Q2E | $1.33B |
| 2029Q3E | $1.35B |
| 2029Q4E | $1.37B |
| 2030Q1E | $1.39B |
| 2030Q2E | $1.40B |
| 2030Q3E | $1.41B |
| 2030Q4E | $1.43B |
| 2031Q1E | $1.44B |
| 2031Q2E | $1.45B |
Assumptions & reasoning
- Product gross margin has run 33-37% for two years with no trend break, so margin is not what decides this line - the megawatt cadence is. Q2'26 36.50% against a 35.76% trailing-twelve-month figure.
- Megawatt volumes here are back-solved from an assumed price, not disclosed. At $3.0M/MW the implied quarterly deployment runs 128, 213, 218 and 312 MW across the last four quarters - a coherent ramp, but arithmetic on an assumption.
- The Oracle warrant is a contra-revenue item, not a cost. Its $324.4M aggregate fair value is 'recognized as a reduction of revenue as the underlying Energy Server systems sold under the Oracle arrangement are delivered'; $306.5M of it still sits on the balance sheet as a customer consideration asset ($91.0M current, $215.5M non-current) and will reduce reported product revenue on the Oracle 1.2 GW as it ships.
- Bloom's own definition of product backlog 'includes both expected Bloom product revenue and reflects anticipated ITC and other tax incentives', so a per-MW figure quoted off backlog is a system value, not recognised revenue per MW. The two are different units.
- The 5 GW figure is not current capacity. The 10-K says the Fremont building 'can accommodate additional capacity expansion of up to approximately 5 gigawatts' at 'approximately six to nine months to install and capital expenditure of approximately $100 million to $150 million' per incremental gigawatt. The committed, funded target is 2 GW by end-2026.
Installation
Units × priceSite work that turns delivered boxes on. Economically a pass-through Bloom runs at or below cost, and structurally a shrinking claim on the P&L: as deployments move to AI-scale campuses where a financier or EPC contracts the civil and electrical work directly, less of it lands in Bloom's revenue at all. Installation was 17.1% of product revenue in 2025 Q3 and 4.0% and 5.4% in the two most recent quarters. It is driven off the same megawatts as product, at a falling dollar-per-megawatt.
Latest: $61M (2031Q2E)
| Period | Value |
|---|---|
| 2023Q1 | $21M |
| 2023Q2 | $24M |
| 2023Q3 | $22M |
| 2023Q4 | $26M |
| 2024Q1 | $11M |
| 2024Q2 | $43M |
| 2024Q3 | $32M |
| 2024Q4 | $36M |
| 2025Q1 | $34M |
| 2025Q2 | $37M |
| 2025Q3 | $66M |
| 2025Q4 | $67M |
| 2026Q1 | $26M |
| 2026Q2 | $51M |
| 2026Q3E | $51M |
| 2026Q4E | $51M |
| 2027Q1E | $51M |
| 2027Q2E | $51M |
| 2027Q3E | $51M |
| 2027Q4E | $52M |
| 2028Q1E | $52M |
| 2028Q2E | $52M |
| 2028Q3E | $53M |
| 2028Q4E | $54M |
| 2029Q1E | $54M |
| 2029Q2E | $55M |
| 2029Q3E | $56M |
| 2029Q4E | $56M |
| 2030Q1E | $57M |
| 2030Q2E | $58M |
| 2030Q3E | $59M |
| 2030Q4E | $60M |
| 2031Q1E | $60M |
| 2031Q2E | $61M |
Assumptions & reasoning
- Gross margin has been negative in eleven of the fourteen reported quarters, ranging -35.3% to +8.9%. Modelling this line at a positive margin would misstate how Bloom actually books site work.
- The ratio to product revenue is the real variable and it is falling hard: 17.1%, 10.5%, 4.0%, 5.4% over the last four quarters. That is a mix shift toward AI-scale sites, not a pricing change, and it is why this line does not scale with product.
Service — O&M on the installed fleet
Subscribers × ARPUContracted operation and maintenance on every Energy Server deployed, on terms Bloom says run 5 to 20 years with roughly full attach on product. It is the annuity under the hardware and the one line whose margin genuinely turned: gross margin went from -26% in 2023 Q1 to +18.65% in the basis quarter as stack life improved and the fleet aged into cheaper service. It grows with the installed base rather than the quarter's shipments, which is why it compounds slowly while product triples. The fleet is measured in kilowatts and the fee in dollars per kilowatt-month.
Latest: $257M (2031Q2E)
| Period | Value |
|---|---|
| 2023Q1 | $41M |
| 2023Q2 | $42M |
| 2023Q3 | $48M |
| 2023Q4 | $53M |
| 2024Q1 | $56M |
| 2024Q2 | $53M |
| 2024Q3 | $51M |
| 2024Q4 | $54M |
| 2025Q1 | $54M |
| 2025Q2 | $54M |
| 2025Q3 | $59M |
| 2025Q4 | $62M |
| 2026Q1 | $62M |
| 2026Q2 | $69M |
| 2026Q3E | $77M |
| 2026Q4E | $85M |
| 2027Q1E | $94M |
| 2027Q2E | $104M |
| 2027Q3E | $113M |
| 2027Q4E | $123M |
| 2028Q1E | $133M |
| 2028Q2E | $143M |
| 2028Q3E | $153M |
| 2028Q4E | $163M |
| 2029Q1E | $173M |
| 2029Q2E | $183M |
| 2029Q3E | $193M |
| 2029Q4E | $202M |
| 2030Q1E | $212M |
| 2030Q2E | $221M |
| 2030Q3E | $231M |
| 2030Q4E | $240M |
| 2031Q1E | $249M |
| 2031Q2E | $257M |
Assumptions & reasoning
- This is the line the $14B 'service backlog' points at, and it is also the line that most overstates the backlog: the 10-K says O&M terms are 'subject to termination for convenience on an annual basis', so most of that figure is not an enforceable obligation.
- Service revenue grew 26.8% year over year in the basis quarter while product grew 215%. Modelling service off the quarter's shipments rather than the cumulative fleet would overstate it by several multiples.
- The installed-fleet megawatt figure is the weakest number in this brief. It is assumed, and the ARPU it produces ($153k per MW per year, about 5.1% of an assumed $3.0M/MW capital cost) is a cross-check that makes it credible rather than a disclosure.
Electricity — Bloom-owned systems
Growth pathLegacy power sales from systems Bloom owns and operates itself, sold under PPAs. Bloom is not adding to this fleet — it now sells the box or lets a financier own it — so the line is in slow runoff and was 0.9% of revenue in the basis quarter. It is carried because it reconciles the reported total, not because it moves fair value.
Latest: $5M (2031Q2E)
| Period | Value |
|---|---|
| 2023Q1 | $20M |
| 2023Q2 | $20M |
| 2023Q3 | $26M |
| 2023Q4 | $16M |
| 2024Q1 | $14M |
| 2024Q2 | $14M |
| 2024Q3 | $14M |
| 2024Q4 | $11M |
| 2025Q1 | $27M |
| 2025Q2 | $13M |
| 2025Q3 | $10M |
| 2025Q4 | $10M |
| 2026Q1 | $10M |
| 2026Q2 | $10M |
| 2026Q3E | $10M |
| 2026Q4E | $9M |
| 2027Q1E | $9M |
| 2027Q2E | $8M |
| 2027Q3E | $8M |
| 2027Q4E | $8M |
| 2028Q1E | $8M |
| 2028Q2E | $7M |
| 2028Q3E | $7M |
| 2028Q4E | $7M |
| 2029Q1E | $7M |
| 2029Q2E | $6M |
| 2029Q3E | $6M |
| 2029Q4E | $6M |
| 2030Q1E | $6M |
| 2030Q2E | $5M |
| 2030Q3E | $5M |
| 2030Q4E | $5M |
| 2031Q1E | $5M |
| 2031Q2E | $5M |
Assumptions & reasoning
- Quarterly revenue is lumpy because it includes periodic true-ups: 2025 Q1 was $26.95M against $12.81M the following quarter. The trend is down; the quarter-to-quarter path is noise.
- This is the only vertical using a growth driver, and the reason is that the underlying owned fleet is never disclosed in megawatts and is not being added to.
Where each case comes from
Backlog is not an obligation case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the Backlog is not an obligation column is what happens if they are taken at face value.
GAAP remaining performance obligations at YE2025
- Feb 9, 2026 As of December 31, 2025, we have unsatisfied performance obligations of $ 394.4 million, primarily related to product sales and installation services.
- Feb 9, 2026 The terms of the contracted operations and maintenance services range from 5 to 20 years, subject to termination for convenience on an annual basis.
Oracle concentration at a single campus
The Oracle warrant reduces reported revenue as systems ship
Frameworks convert case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the Frameworks convert column is what happens if they are taken at face value.
Oracle partnership expanded to 2.8 GW
Brookfield financing framework raised fivefold
AEP takes a substantial portion of its 900 MW option
The AI book is still small case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the The AI book is still small column is what happens if they are taken at face value.
From cash flow to fair value
| Present value of free cash flow, 20 quarters | $3.00B |
| Terminal-year revenue | $6.96B |
| Terminal-year EBITDA | $1.36B |
| Exit multiple, on revenue | 3.5x |
| Terminal value | $24.36B |
| Discounted at 11.0% a year, terminal value becomes | $14.46B |
| Enterprise value | $17.45B |
| Net cash | -$18M |
| Equity value | $17.44B |
| Diluted shares | 0.32B |
| Fair value per share | $53.93 |
| Against the current price of $200.69 | -73% |
GAAP-profitable for two quarters, one product line, and heavy concentration in a single named counterparty — 11% is the rate that pays for that. The 4.5x exit is power-equipment hardware at maturity; the stock trades near 16x forward revenue on the FY26 guide midpoint today.
Read the other way round: at $200.69 the market is paying 15.0x terminal-year revenue, holding every other assumption on this page fixed. That is the number to argue about.
Capex outside the verticals
Each vertical's capex is a share of its own revenue, which is the right shape for capacity that scales with sales. These are not that: fixed-size programmes with their own schedule that spend whether or not any segment grows. They are added on top, and they are why free cash flow dips in the years below.
Fremont capacity expansion, each incremental gigawatt beyond 2 GW
2027 Q1 → 2027 Q3The 10-K prices each incremental GW of run rate at $100-150M and six to nine months to install, up to a 5 GW ceiling the building can accommodate. This is the midpoint of one such step, and it belongs to no vertical.
The projected path
| Quarter | Product — Energy Servers | Installation | Service — O&M on the installed fleet | Electricity — Bloom-owned systems | Revenue | YoY | EBITDA | Capex | FCF | R40 | PV of FCF |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026 Q3E | $982M | $51M | $77M | $10M | $1.12B | +116% | $202M | $49M | $138M | +128 | $134M |
| 2026 Q4E | $1.03B | $51M | $85M | $9M | $1.17B | +50% | $214M | $50M | $147M | +63 | $140M |
| 2027 Q1E | $1.07B | $51M | $94M | $9M | $1.22B | +62% | $225M | $93M | $119M | +72 | $110M |
| 2027 Q2E | $1.10B | $51M | $104M | $8M | $1.27B | +19% | $236M | $94M | $128M | +29 | $115M |
| 2027 Q3E | $1.14B | $51M | $113M | $8M | $1.31B | +17% | $246M | $95M | $136M | +28 | $119M |
| 2027 Q4E | $1.17B | $52M | $123M | $8M | $1.36B | +16% | $256M | $54M | $182M | +29 | $155M |
| 2028 Q1E | $1.21B | $52M | $133M | $8M | $1.40B | +15% | $266M | $55M | $189M | +28 | $158M |
| 2028 Q2E | $1.24B | $52M | $143M | $7M | $1.44B | +13% | $274M | $56M | $196M | +27 | $159M |
| 2028 Q3E | $1.26B | $53M | $153M | $7M | $1.48B | +12% | $283M | $57M | $203M | +26 | $161M |
| 2028 Q4E | $1.29B | $54M | $163M | $7M | $1.51B | +11% | $291M | $58M | $210M | +25 | $162M |
| 2029 Q1E | $1.31B | $54M | $173M | $7M | $1.54B | +10% | $298M | $58M | $216M | +24 | $162M |
| 2029 Q2E | $1.33B | $55M | $183M | $6M | $1.58B | +10% | $305M | $59M | $222M | +24 | $162M |
| 2029 Q3E | $1.35B | $56M | $193M | $6M | $1.61B | +9% | $312M | $59M | $227M | +23 | $162M |
| 2029 Q4E | $1.37B | $56M | $202M | $6M | $1.63B | +8% | $318M | $60M | $232M | +22 | $161M |
| 2030 Q1E | $1.39B | $57M | $212M | $6M | $1.66B | +8% | $324M | $60M | $237M | +22 | $160M |
| 2030 Q2E | $1.40B | $58M | $221M | $5M | $1.69B | +7% | $329M | $61M | $241M | +21 | $159M |
| 2030 Q3E | $1.41B | $59M | $231M | $5M | $1.71B | +6% | $334M | $61M | $246M | +21 | $158M |
| 2030 Q4E | $1.43B | $60M | $240M | $5M | $1.73B | +6% | $339M | $62M | $249M | +20 | $156M |
| 2031 Q1E | $1.44B | $60M | $249M | $5M | $1.75B | +5% | $343M | $62M | $253M | +20 | $154M |
| 2031 Q2E | $1.45B | $61M | $257M | $5M | $1.77B | +5% | $346M | $62M | $256M | +19 | $152M |
Every row is projected. A year-over-year change is shown only where the quarter it compares against exists — an em dash means there is no comparable quarter, not a flat year.
Model revisions
Assumptions are marked to reality as each quarter prints. Every change is appended here, with the fair value the model produced at the time, so the model's own history stays visible.
| Date | Changed | Fair value then | Note |
|---|---|---|---|
| 2026-08-23 | — | First cut, built from the scope-model intake brief at data/models/intake/be.json on the 2026 Q2 basis. |