Bloom Energy BE

Reference table

1. What they actually sell — and why it matters right now

Bloom Energy Corporation (NYSE: BE, founded 2001 as Ion America Corp., renamed September 2006; headquartered at 4353 North First Street, San Jose, California; CEO and founder K. R. Sridhar; 2,214 full-time employees; IPO'd 2018-07-25) makes, in its own words, "cutting-edge solid-oxide fuel cell systems designed for on-site electricity generation."

The core product is the Bloom Energy Server — "an advanced power platform capable of converting various fuels, including natural gas, biogas, hydrogen, or a blend of these, directly into electricity using an electrochemical process that eliminates the need for combustion."

The customer list, per the company's own description, spans "data centers, hospitals, healthcare manufacturing and biotechnology facilities, grocery and hardware stores, banks, and telecommunication centers." The data centers are the reason the stock is up 484%.

Revenue by line (FY25, 2025-12-31):

Two things to take from this mix, and one of them is a caution.

The good: the split ties almost exactly to the reported FY25 income-statement revenue of $2.024B (a $2M rounding difference), so unlike three other names in this batch the segment data here is trustworthy.

The caution: 75.6% of revenue is Product — a one-time hardware capital sale. The recurring lines, Service (11.3%) and Electricity (3.0%), together are just 14.3% of revenue and grew far more slowly (+6.2% and +14.2%) than Product (+41.1%). This is a capital-equipment business, not an annuity business, and that fact is the single strongest argument for why the exponential score is 8 and not 10. Every dollar of revenue has to be re-won next year.

Data gap, stated: there is NO geographic segment disclosure in this dataset — the field is empty. Bloom does not report geographic segments here, and none is analysed, despite the company describing customers "both within the United States and internationally."

Note also: the FY22 and FY21 product-segment rows contain only a "Financing Obligations" line ($3.3M and $10.0M), which is clearly not a revenue breakdown. Those two years are excluded from the segment analysis.

2. The inflection — verified, not promised

This is the section that justifies the verdict, and it is worth laying out in full because the change is unusually stark.

The quarterly sequence:

Read the transition. Through Q3'25 this was a company losing money in most quarters on sub-$550M revenue. By Q2'26 it did over a billion dollars in a single quarter and earned $196.3M of GAAP net income — against a $42.6M loss in the same quarter a year earlier. H1'26 revenue of $1.816B was +149.7% on H1'25's $727.3M, and H1'26 GAAP net income of $267.0M compares to an H1'25 loss of $66.4M.

And the beats have been enormous:

Five consecutive beats, the last two by 3.6x and 2.0x, with revenue beating by 39% and 29%. The sell-side is chasing this company, not leading it. That is a genuine, verifiable, dated fact and it is the strongest single piece of evidence in this dive.

Now the honest counterweight — gross margin. Q2'26 gross margin was 33.4% against 36.1% in Q2'25 — down 270 basis points while revenue rose 165%. Over the FY series: 20.3% (FY20), 20.1% (FY21), 12.4% (FY22), 14.8% (FY23), 27.5% (FY24), 29.0% (FY25). So the multi-year trend is up, and the year-over-year quarterly comparison is down. The most likely reading is mix — Installation revenue grew 59.1% in FY25 and installation carries lower margin than product — but this dataset does not break out margin by line, so that is inference and is labelled as such. Either way, margin is the metric that determines whether this is a durable compounder or a capex-cycle beneficiary, and it is currently going the wrong way.

Note on EPS definitions: the income statement reports Q2'26 diluted EPS of $0.62 while the earnings calendar records an actual of $0.78, consistent with a non-GAAP versus GAAP distinction. Both are shown throughout; the $0.78 is used when comparing to consensus, since consensus is stated on the same basis.

3. The financials — a strong balance sheet, a weak cash-conversion record, and heavy dilution

Annual series:

The multi-year trajectory is a company grinding from deep losses to breakeven — gross margin more than doubling from 12.4% to 29.0%, operating income crossing zero in FY24 — and then, in 2026, inflecting hard.

TTM profitability: gross margin 31.2%, EBITDA margin 11.2%, operating margin 11.2%, net margin 7.87%. Return on equity 24.8%, return on invested capital 7.51%. R&D is 7.08% of revenue and SG&A 12.93% — a real but not bloated cost base. Stock-based compensation is 2.75% of revenue. Interest coverage 8.22x.

Balance sheet (FY25, 2025-12-31) — genuinely strong, and the reason the risk score is 8 rather than 10:

Cash tripled to $2.454B in FY25 while net debt FELL to $537.8M, because the company raised $1.450B of net new debt (the cash-flow statement's netDebtIssuance line) and held the proceeds. Current ratio 4.09x, cash ratio 2.37x, cash per share $9.46. Net-debt/EBITDA is just 0.41x. There is no near-term solvency question here.

But note the retained-earnings deficit of negative $3.987B — the accumulated cost of two decades of losses. Book value per share is $5.71 against a $218.32 share price — a price-to-book of roughly 38x (the reported TTM ratio is 38.9x). This is a company whose value is entirely forward-looking.

Cash flow — the weakest part of the picture:

FY25 free cash flow of $57.2M against a $64.30B market capitalisation is a 0.09% free cash flow yield. That is the number the bull case has to grow into, and it is a very long way from the equity value. The TTM figures are better — freeCashFlowYield 0.98%, priceToFreeCashFlow 102.4x — reflecting the 2026 inflection, but they still describe a company whose cash generation lags its earnings recognition badly. Cash conversion is where this story is thinnest, and it should be the first thing checked in each new print.

Dilution — the hidden cost, and it is large:

Up 42.2% versus FY24, and up 34.5% versus FY25. The $1.450B of net debt issuance in FY25 points to convertible instruments as the likely mechanism, and the share count jumped 33% between Q4'25 and Q1'26 — consistent with conversion or with dilutive-security inclusion once the company turned profitable (accounting rules include convertibles in diluted share counts once they are dilutive). Whatever the mechanism, the effect is unambiguous: enterprise growth of 165% translates into per-share growth of substantially less, and every forward EPS estimate in §4 is stated on a share count that has been rising fast.

Insider activity — a genuine yellow flag. The eight most recent filings show five open-market sales through May and June 2026 and zero purchases:

Plus two director A-Awards on 2026-06-30 (Warner 76 shares, Jeffrey R. Immelt 85 shares, both at a reference price of $302.70).

How to read it. Every sale was executed between $288.62 and $300.37 — within roughly 13% of the 52-week high of $345.85 — and the shares now trade at $218.32, meaning the sellers were, so far, right. Officer sales at these prices could easily be routine diversification after a 500% move, and Chambers' 55,000 shares left him holding 238,333. Immelt still holds 231,243. But the absence of any open-market buying, on any day, at any price, including after a 37% drawdown, is the honest counterpoint to the conviction lane. It is flagged, weighted, and turned into a watch item rather than dismissed.

4. The knowledge base — the deepest lane in this batch, and its own bear case has been falsified

This is where BE separates itself sharply from BAM (zero claims), MPLX (one passing mention) and ET (two).

11 tagged claims across 5 tracked thinkers. Net conviction: positive-high.

Lead voice — jordi_visser (skill 2.0), five claims, all bullish, conviction 60-75:

• 2025-08-10 (conviction 62), entity ['BE']: "Bloom Energy is a company to focus on — management claims it can solve the AI power-speed challenge within 90 days; with power you need speed and size." Note the house half-weight rule applies to the management-sourced portion of this claim.
• 2025-09-14 (conviction 70), entities ['BE']: "Bloom Energy is Oracle's first direct hyperscaler client; every energy company is looking for power solutions — stock ran from ~20-30 when I flagged it to nearly 70." A dated, checkable, self-scoring claim.
• 2025-12-21 (conviction 60), entities ['Bloom Energy','BE']: "Overlooked power/infrastructure names tied to Stargate and the datacenter buildout are becoming investable as hardware spend intensity rises."
• 2026-01-11 (conviction 75 — his highest), entity ['Bloom Energy']: "Bring-your-own-generation / behind-the-meter power is a major theme; hyperscalers must self-generate as the grid can't keep pace." This is the cleanest statement of the thesis in the entire corpus.
• 2026-06-17 (conviction 68), entity ['BE']: "The energy/power layer — Bloom Energy, battery makers, natural-gas transformer companies — is one of the bottom-three cake layers worth owning for the AI buildout."

Supporting voices:

• forward_guidance, 2026-04-30 (bullish, conviction 60), entity ['BE']: "Data centers bypass grid politics using Bloom Energy's gas turbines; owning power access makes you a 21st-century gas station in secular AI-infrastructure growth."
• doomberg, 2026-02-06 (bullish, conviction 55), entity ['BE']: "Technologies unlocking cheap natural gas for electricity—Bloom Energy and gas-turbine makers—are compelling AI/datacenter enablers; doesn't own it, not advice." Note the disclosed non-ownership.
• warren_pies, 2023-03-24 (bullish, conviction 65), entities ['Bloom Energy','BE']: "Bloom Energy is well-financed with minimal external finance requirements and should break out meaningfully this year." Three years old; included for completeness.

The cautionary lane — three claims, all dated 2025-10-27, all from compound_and_friends:

• (bearish, conviction 65), entities ['D-Wave','Rigetti','RGTI','IonQ','Oklo','OKLO','Bloom Energy','BE']: "The quantum/nuclear/power froth echoes 1998-99 mania — trim and take some off, don't blow it all out, but don't fully exit."
• (neutral, conviction 55), entities ['Bloom Energy','BE']: "Bloom has never made money and doesn't scale, but now has a real product powering backup for a few data centers."
• (neutral, conviction 50), entity ['Bloom Energy']: "Bloom finally has a real product (datacenter backup power) but has never made money and doesn't scale."

And here is the analytically important part, stated plainly: two of those three cautionary claims have been factually falsified by the subsequent data.

"Has never made money" was true when written on 2025-10-27 — FY24 net income was -$29.2M and FY25 would print -$88.4M. It is no longer true: H1'26 GAAP net income was $267.0M, with $196.3M in Q2'26 alone. "Doesn't scale" was a claim about unit economics; revenue scaled 165.5% year over year in Q2'26 while operating income scaled from $34.2M to $182.2M — operating leverage of roughly 5x. The remaining bearish claim — that this is 1998-99-style froth in a quantum/nuclear/power basket — is not falsified, and given a 3.737 beta and an 83x forward multiple it deserves genuine respect. But the specific operating objections have been overtaken by events.

This matters beyond bookkeeping. A conviction lane where the bull case has been confirmed by five consecutive earnings beats and the bear case has been contradicted by the same data is materially stronger than a lane where nobody has been tested. That is the single best argument for owning BE at all, and it is why the verdict is Stage-In rather than Watch despite a base fair value only 6.3% above spot.

Adjacent thematic support, cited verbatim:

• sam_altman_mgmt, 2026-07-03 (neutral, conviction 72): "Data centers run 24/7 and can't run on renewables alone without sufficient energy storage, which is why AI companies keep invoking nuclear (and natural gas) as the baseload solve-all." Half-weighted per house rule as management-sourced.
• all_in, 2026-06-10 (bullish, conviction 68), entity ['natural gas']: "Can't win the AI race without energy; Pennsylvania data centers are driving enormous investment—Homer City's 4.4GW coal plant converting to gas, 3.4GW feeding a data center complex."
• forward_guidance, 2026-03-06 (bullish, conviction 50): "Natural gas/LNG is a key funding driver of the energy needed for the AI datacenter buildout."

Read: High conviction, breadth 5, net positive-high. The panel owns this theme and names this ticker repeatedly, at rising conviction, from a skill-2.0 lead voice, over a full year. This is the strongest KB position in the batch by a wide margin — and it is deliberately not allowed to override the beta-driven sizing constraint.

5. Valuation — priced in or room?

At $218.32 (market cap $64.30B, EV $64.45B, ~323.3M diluted shares, net debt $537.8M):

Three things to note before using any of this.

First — a critical data caveat. The consensus EBITDA fields in this dataset are unusable and are NOT used anywhere in this dive. They are negative in every single forecast year — -$209.3M (FY26E), -$345.4M (FY27E), -$504.4M (FY28E), -$665.4M (FY29E), -$830.2M (FY30E) — while consensus EPS in the same rows is strongly positive and rising. A company cannot earn $12.88 per share on negative EBITDA. This is an internal contradiction in the vendor's estimate set, it is flagged rather than smoothed, and every valuation figure here is built on revenue and EPS instead.

Second, a related inconsistency. FY26E consensus netIncomeAvg of $620.3M against epsAvg of $2.61745 implies roughly 237M shares — but actual Q2'26 diluted shares are 323.3M. The estimate set appears to be running on a stale share count. This dive uses the published epsAvg figures as the headline (since that is what the street quotes and what the price is measured against) but flags that the EPS estimates may be optimistic relative to the actual, and rising, diluted share base. That is a material caution on every P/E in the table above.

Third, the coverage is genuinely deep. 18-25 analysts on revenue and 13-16 on EPS through FY28 is far more than any other name in this batch — more than double LNG's and triple ET's outer-year coverage. That does not make the estimates right, but it does mean they are not one or two people's guesses.

The de-rating curve is the whole valuation argument. At a constant $218.32, the multiple falls from 83.4x (FY26E) → 45.4x (FY27E) → 28.3x (FY28E) → 20.1x (FY29E) → 17.0x (FY30E). That is the fastest forward compression in the batch by an enormous margin. But the addendum's guardrail applies with full force: that compression is mechanical rolldown at a constant price, NOT re-rating. It only accrues to a shareholder if the estimates are delivered.

Synthos fair values — stated arithmetic on forward EPS, no DCF theater. Method: target multiple × FY28E consensus EPS of $7.72196, chosen because FY28 is the last year with double-digit analyst coverage on both revenue (25) and EPS (13), and because FY26's 83x multiple is too far out on the curve to anchor a fair value.

• Bear ~$154 = 20x FY28E consensus EPS of $7.72196 = $154.44. The de-rating case: growth decelerates from the modelled 46% FY28 rate, gross margin keeps slipping from 33.4%, dilution continues, and a high-beta name in a risk-off tape re-rates to a market-like multiple on delivered rather than projected earnings. Lands ~14% below the rising 200-DMA of $179.24. -29.3% from spot.
• Base ~$232 = 30x FY28E consensus EPS of $7.72196 = $231.66. A 30x multiple is set as base because it is roughly a market-premium multiple for a business consensus models compounding EPS ~49% a year — a PEG of about 0.6 — and because paying more than 30x on estimates two years out, for a hardware business at 33% gross margin with 42% trailing dilution, is not defensible. +6.3% from spot at the stated $232 anchor.
• Bull ~$309 = 40x FY28E consensus EPS of $7.72196 = $308.88. Requires the full case: revenue tracking to the $9.754B FY28E figure, gross margin recovering above 36%, dilution stopping, and the market continuing to pay a growth premium. Sits between the street's $285 median and $350 high targets, and just below the 52-week high of $345.85. +41.5% from spot.

Cross-check on sales: the base $232 implies a market cap of $75.0B and, holding net debt at $537.8M, an EV of $75.5B — 7.74x FY28E consensus revenue of $9.754B, against 6.61x today. For a business growing revenue 46% that year, 7.7x forward sales is not extreme. The methods agree.

Cross-check against the street: consensus is $273.73 with a $285 median — well above this dive's $232 base. We are deliberately more conservative than the street, and the reason is the two data problems above: broken EBITDA fields and an EPS estimate set apparently running on a stale share count. When the estimates themselves are internally inconsistent, the appropriate response is a lower multiple, not a higher target.

6a. What today's price assumes (the inversion)

At $218.32 (83.4x FY26E, 45.4x FY27E, 28.3x FY28E consensus EPS), today's price embeds roughly the following. Every figure is consensus-derived arithmetic, labelled as such.

• Revenue compounds ~41% a year for five straight years. Consensus: $4.047B (FY26E) → $16.054B (FY30E), which is 7.9x FY25's actual $2.024B. Growth decelerates gently — +100%, +65%, +46%, +32%, +25% — with no stumble in any year. Falsified by: any quarter where revenue growth drops below ~40% year over year, or by FY27 revenue tracking below ~$6B against the $6.680B consensus. Near-term check: Q3'26 consensus of $1.0449B is BELOW Q2'26's actual $1.0654B — the street models a sequential decline, so a sequential increase is an immediate positive surprise.
• EPS reaches ~$7.72 by FY28 (consensus, 13 analysts) from FY25's actual -$0.37. And that estimate appears to assume ~237M shares, against 323.3M actually outstanding at Q2'26. The price assumes both the earnings and a share count that has already been exceeded. This is the most fragile assumption in the price. If FY28 EPS is delivered on 350M+ shares instead, the same net income produces roughly $5.20 of EPS, and 30x that is $156 — the bear case, reached without the business missing a single revenue target.
• Gross margin stops compressing and ideally recovers. Q2'26 was 33.4% against 36.1% a year earlier. Consensus net margins implied by the FY28E estimates require substantial operating leverage, which is very hard to deliver on a falling gross margin. Falsified by: a Q3'26 gross margin below ~32%.
• The market keeps paying 28-45x forward earnings for a hardware business with 33% gross margins, 75.6% one-time Product revenue, a 3.737 beta and a 0.09% FY25 free cash flow yield. Falsified by: any broad de-rating of high-beta AI-adjacent names — which is exactly the scenario compound_and_friends (2025-10-27, bearish, conviction 65) described as "the quantum/nuclear/power froth echoes 1998-99 mania."

6b. The return bridge (why the multiple moves)

> expected return ≈ EPS growth + multiple drift + shareholder yield

• EPS growth: +83.5% (consensus FY26E $2.61745 → FY27E $4.80386). Over FY26-FY30 the compound rate is 48.9%. This is overwhelmingly the entire return, and unusually so.
• Multiple drift: substantially NEGATIVE, and deliberately so. This dive's base assumes the market pays 30x FY28E where it currently pays 28.3x FY28E — so on a like-for-like FY28 basis the assumption is a marginal +6%. But measured against the 83.4x the market pays on FY26E today, the base case embeds massive multiple compression as the growth matures. That is the correct treatment and it is stated explicitly: we are assuming the multiple falls a long way, and the return still works because the earnings grow faster than the multiple contracts. The addendum's guardrail matters here more than anywhere else in the batch — 83.4x becoming 28.3x at a constant price is rolldown, not re-rating, and it is not counted as return.
• Shareholder yield: NEGATIVE. There is no dividend (lastDividend 0, dividendYieldTTM 0) and no buyback (FY25 commonStockRepurchased $0). Worse, the diluted share count rose 42.2% since FY24, so the shareholder yield is meaningfully negative. This is the only name in this batch where the capital-structure leg subtracts from the return rather than adding to it — compare LNG's +5.9%, MPLX's +7.8% and ET's +6.58%.

Say it plainly: 100% of the base-case return is earnings growth, fighting against multiple compression and negative shareholder yield. There is no yield floor, no buyback support and no valuation cushion. That is the definition of a name you stage into rather than buy in size, and it is the arithmetic behind the 1-1.5% starter.

6c. Variant perception (where we differ, what would surprise)

• Where we differ from the street: we are more conservative, and on the estimates rather than the story. Consensus is $273.73 with a $285 median; this dive's base is $232. The gap is not a disagreement about Bloom's growth — we accept the 41% revenue path — it is a refusal to pay street multiples on an estimate set that contains negative EBITDA in every forecast year alongside positive EPS, and that implies ~237M shares against 323.3M actually outstanding. When the model set is internally inconsistent, the right adjustment is a haircut. That is the variant perception: same business, lower multiple, because the numbers underneath the consensus do not tie.
• Where we differ from the KB's own bear: it has been falsified, and we say so. compound_and_friends (2025-10-27): "Bloom has never made money and doesn't scale." H1'26 GAAP net income was $267.0M on revenue up 149.7%, with operating income scaling 5x. We treat the falsification as a positive signal about the thesis and record it rather than quietly dropping the claim — a bear case that was tested and failed is worth more than one that was never tested.
• Where we side with the KB's remaining bear: the froth risk is real. The unfalsified part of that same 2025-10-27 cluster — "the quantum/nuclear/power froth echoes 1998-99 mania — trim and take some off" (conviction 65) — describes a market-structure risk that a 3.737 beta and an 83x forward multiple make entirely live. The stock is already -36.9% from its July high, which is what that claim looks like when it is partially right.
• The under-discussed structural question: does the window stay open? Bloom's advantage is speed to power while grid interconnection is slow and gas turbines are supply-constrained. Both constraints are being actively worked on across the industry. We think the market is underwriting a five-year revenue ramp against a bottleneck whose duration nobody has priced. This is the single most important long-horizon uncertainty and it is not visible in any estimate in this dataset.
• Positive surprise that would force a reprice up: a Q3'26 print (2026-10-27) that beats the low $0.659 consensus and shows revenue rising sequentially above Q2's $1.0654B and gross margin recovering above 35%. That combination — growth plus margin plus a beaten low bar — validates 40x and puts the $309 bull anchor in play. Watchable numbers: Q3 revenue against $1.0449B, and gross margin against 33.4%.
• Negative surprise that would force a reprice down: gross margin below ~32%, or diluted share count above ~340M, or the first missed quarter after five consecutive beats. In that case 20x FY28E and the $154 bear anchor becomes operative — a 29% fall from here, which for a 3.737-beta stock is an ordinary month. Watchable number: diluted shares against the Q2'26 323.3M.

6. Technicals — a violent correction inside a violent uptrend

• Price $218.32 on 2026-08-04, +6.08% on the day, on 18.27M shares against a 12.54M average — a genuine volume bounce, with an intraday range of $191.00 to $224.64 (a 17.6% swing in one session, which tells you everything about the 3.737 beta).
• -36.9% from the 52-week high of $345.85 (quote-based high $351.28). +504.8% above the 52-week low of $36.10. Maximum drawdown from peak: -36.9%.
• 15.8% BELOW a 50-DMA of $259.43, and that average is now falling — the intermediate trend is broken.
• 21.8% ABOVE a rising 200-DMA of $179.24. This is THE level in the name. It is the primary-trend line, it sits 17.9% below spot, and it is where the bear-case fair value zone ($154) begins to come into view. Holding it keeps the secular thesis intact; losing it changes the character of the chart entirely.
• RSI 44.3 — below the midline but nowhere near oversold. MACD -19.17 — deeply negative, the weakest momentum reading in this batch.
• Relative performance: 12-month +483.9% against SPY +19.9% and QQQ +23.9%; 6-month +44.2% against SPY +9.5%; 3-month -24.9% against SPY +5.1%. A 484% twelve-month return with a -25% three-month return is the signature of a name in the digestion phase of a parabolic move — and digestion of that magnitude usually takes longer than four weeks.

The tactical read. This is neither a bottom nor a breakdown. A stock 37% off its high, below a falling 50-DMA, above a rising 200-DMA, with RSI at 44 and MACD at -19, is mid-correction. The honest position is that further downside toward the 200-DMA at ~$179 is entirely ordinary for a 3.737-beta name and would not damage the thesis at all — which is precisely why tranche two is set there rather than here.

7. Moat and competitive position

The moat is technological and temporal, and the temporal half is the part that matters.

Technologically, solid-oxide fuel cells at commercial scale are genuinely hard. Bloom has been at it since 2001 — a 25-year head start, embodied in $3.987B of accumulated retained-earnings deficit, which is a fair proxy for what the learning curve cost. R&D runs 7.08% of revenue. The product converts multiple fuels — natural gas, biogas, hydrogen or blends — without combustion, which is a real engineering differentiator and gives long-run optionality if hydrogen economics ever arrive.

Temporally, and more importantly: Bloom sells speed. A hyperscaler that needs 100MW can wait years in a grid interconnection queue, or it can put Bloom servers on its own land and have power in months. That is the moat, and it is a moat made of somebody else's bottleneck. The knowledge base names the mechanism precisely — jordi_visser (2026-01-11, conviction 75): "hyperscalers must self-generate as the grid can't keep pace"; forward_guidance (2026-04-30, conviction 60): "Data centers bypass grid politics."

And that is the vulnerability, stated plainly. A moat built on a queue erodes when the queue clears. Three things could shorten it: grid interconnection reform, gas-turbine supply catching up with demand (Bloom competes for the same customers), or utilities building dedicated generation for hyperscalers directly. This dataset contains no information on any of those trajectories, which is itself the point — the most important variable in the long-run thesis is unmeasurable from here.

The peer set in this dataset is unhelpful and worth flagging as such: EME (EMCOR, $36.1B), FIX (Comfort Systems, $62.3B), HEI (HEICO, $50.2B), HUBB (Hubbell, $25.2B), IR (Ingersoll Rand, $34.4B), OTIS ($27.5B), RKLB (Rocket Lab, $40.8B), UAL (United Airlines, $41.7B), WAB (Wabtec, $50.1B) and XYL (Xylem, $27.8B). These are size-matched industrials, not competitors — Bloom has no listed pure-play solid-oxide peer in this dataset, and no peer-multiple comparison is offered because none would be meaningful.

One competitive fact from the KB worth recording: jordi_visser (2025-09-14, conviction 70) states "Bloom Energy is Oracle's first direct hyperscaler client." If accurate, a named hyperscaler relationship is exactly the reference customer this kind of business needs. This dataset does not independently confirm it, and it is reported as a tracked claim rather than as a fact.

8. Verdict, kill-criteria and flip conditions

Stage-In (small, staged). Bloom Energy has done something rare: it delivered the inflection the narrative promised. Revenue +165.5% year over year in Q2'26. GAAP net income from -$42.6M to +$196.3M in twelve months. Five consecutive earnings beats, the last two by 3.6x and 2.0x. $2.454B of cash, a 4.09x current ratio and only $537.8M of net debt. Consensus modelling revenue to $16.054B by FY30E on 12-25 analysts. And the deepest knowledge-base conviction lane in this batch — 11 claims, 5 thinkers, a skill-2.0 lead voice at conviction 60-75 across a full year — whose own bear case has been factually falsified by the results.

Four things cap this at a small, staged position rather than a core one:

1. Beta 3.737. Five times ET's, three times BAM's. The stock traded a 17.6% intraday range on the day of this dive. It is already -36.9% from its high and can easily be -29% again to the bear case without anything breaking.

2. The estimate set is internally broken. Consensus EBITDA is negative in every forecast year while consensus EPS is strongly positive — an impossibility. And FY26E consensus EPS implies ~237M shares against 323.3M actually outstanding. When the numbers underneath the target price do not tie, you buy less.

3. Dilution of 42.2% since FY24 with no dividend and no buyback — the only name in this batch where the capital-structure leg subtracts from the return.

4. Gross margin is compressing — 33.4% against 36.1% a year earlier — and insiders sold between $288.62 and $300.37 through June 2026 with zero buying, including a director's 55,000 shares at $297.69.

Base fair value of $232 is only +6.3% from spot. The reason this is Stage-In rather than Hold is not the fair value — it is the option value in the bull case ($309, +41.5%) combined with a conviction lane that has been repeatedly tested and repeatedly confirmed. You are buying a real exponential at a fair-ish price with a wide distribution of outcomes, and the correct response to a wide distribution is a small first position.

Staged entry — build toward 2.5-3% in three tranches:

1. Tranche 1 — now (roughly one third of target, so 1-1.5% of the portfolio), at ~$218.32. You are buying a verified inflection, the deepest conviction lane in the batch, at 28.3x FY28E, 37% below the high, above a rising 200-DMA. Size it so a 40% drawdown is survivable without a decision.

2. Tranche 2 — a test of the 200-DMA (~$179) that holds, OR a Q3'26 print (2026-10-27) that beats the low $0.659 consensus with sequential revenue growth and stable gross margin. Either a better price or a confirmed margin picture.

3. Tranche 3 — only if BOTH the margin recovers above ~35% AND the diluted share count stabilises near ~325M. This tranche is contingent on the two things most likely to go wrong actually going right.

Pre-registered KILL / avoid-adding criteria:

• Gross margin falls below ~32% (from Q2'26's 33.4%) — the metric that separates a compounder from a capex-cycle beneficiary breaking down.
• Diluted share count rises above ~340M (from 323.3M) — continued dilution at scale, which would confirm the FY28 EPS estimates are structurally too high.
• The first missed quarter after five consecutive beats, or Q3'26 revenue below the $1.0449B consensus — the sell-side has been chasing this company; if it stops having to, the multiple has no support.
• The 200-DMA (~$179.24) is decisively lost on company-specific news rather than sector beta — the primary uptrend breaking, which opens the $154 bear zone.
• Any further insider selling below ~$220 — selling near the high is diversification; selling into a 37% drawdown is a different signal.
• A broad de-rating of high-beta AI-adjacent names — the unfalsified half of the compound_and_friends 2025-10-27 froth warning becoming operative.

Pre-registered FLIP TO HIGH-CONVICTION (upsize toward the top of the 2.5-3% range):

• Q3'26 (2026-10-27) beats the $0.659 consensus with revenue ABOVE Q2's $1.0654B and gross margin above 35% — growth, margin and a beaten low bar together.
• Or the shares reach ~$179-190 (the 200-DMA zone, roughly 23-25x FY28E) with FY28E consensus revenue still at or above $9.754B — a 22-30% gap to base and the margin of safety a 3.737-beta name requires.
• Or any insider BUYING appears after the June 2026 selling cluster, which would resolve the sharpest qualitative flag in the dive.
• Or the consensus estimate set is corrected — positive forward EBITDA and an EPS series built on the actual ~323M share count — which would let a higher multiple be defended honestly.

Where BE fits in the Synthos Framework Portfolio. The AI-infrastructure / power sleeve, as the high-beta exponential leg — 1-1.5% starter, 2.5-3% maximum, built in three tranches. This is deliberately the smallest maximum weight in the batch despite carrying the highest conviction rating, because conviction and volatility are separate inputs and the 3.737 beta is the binding one. On overlap: BE is complementary, not redundant to the three energy names in this batch. ET and MPLX sell the molecules and the transport — low beta, high yield, contracted, 6-7% distributions. LNG (Cheniere) sells the export capacity. BE sells the conversion device that turns those molecules into electrons at the customer's fence line — no yield, no contract cover, and a beta an order of magnitude higher. A sensible sleeve pairs BE's torque against MPLX's or ET's income, so that the blended volatility is tolerable and the gas-demand thesis is expressed at both ends of the chain. Logged as a tracked Synthos call (Stage-In) as of 2026-08-04 at $218.32.

Single biggest risk: the window closes. Bloom's entire competitive advantage is speed to power while grid interconnection queues are long and gas turbines are supply-constrained. That is a moat made of somebody else's bottleneck, and bottlenecks get fixed — by interconnection reform, by turbine capacity catching up, or by utilities building dedicated hyperscaler generation. If the queue shortens, an 83x-forward-earnings, 33%-gross-margin, 75.6%-one-time-hardware business with no yield and 42% trailing dilution has nothing left to hold the multiple up. Nothing in this dataset measures the duration of that bottleneck, which is exactly why the position is staged and capped.

Single most fragile assumption in the price: that FY28 EPS of $7.72196 arrives on the share count the consensus appears to assume. That estimate implies roughly 237M shares while 323.3M are already outstanding — a 36% gap. Do the arithmetic: $7.72196 × ~237M implied shares ≈ $1.83B of net income; deliver that same $1.83B on 350M shares and EPS is ~$5.23, and 30x that is ~$157 — the bear case, reached with the company hitting every single revenue target. Dilution, not demand, is the most likely way this thesis disappoints.

Provenance and disclosures

• Traceability: kb_claim_count 11 across 5 tracked thinkers — breadth 5, net conviction positive-high. Lead voice jordi_visser (skill 2.0), five bullish claims: 2025-08-10 (conviction 62), 2025-09-14 (70), 2025-12-21 (60), 2026-01-11 (75), 2026-06-17 (68). Supporting: forward_guidance (2026-04-30, bullish, 60), doomberg (2026-02-06, bullish, 55, discloses non-ownership), warren_pies (2023-03-24, bullish, 65). Cautionary: compound_and_friends, three claims all dated 2025-10-27 — bearish 65 (froth), neutral 55 and neutral 50 (both stating Bloom "has never made money and doesn't scale"). The two neutral claims are recorded as FACTUALLY FALSIFIED by H1'26 results ($267.0M of GAAP net income on revenue +149.7%) rather than dropped; the bearish froth claim is not falsified and is weighted in the risk score. Adjacent support cited verbatim from sam_altman_mgmt (2026-07-03, conviction 72 — half-weighted per the house management-framing rule), all_in (2026-06-10, conviction 68) and forward_guidance (2026-03-06, conviction 50). The jordi_visser 2025-08-10 claim relays a management assertion (the "90 days" figure) and is likewise half-weighted on that portion. All quotes verbatim.
• Data as-of: fundamentals 2026-06-30 (Q2'26; annual figures 2025-12-31) · estimates 2026-08-04 · prices 2026-08-04 (quote $218.32, 50-DMA $259.43, 200-DMA $179.24, 52-wk range $35.00-$351.28 per quote / $36.10-$345.85 per technicals) · KB claims swept 2026-08-04. All market data from the single pre-pulled FMP dataset for this ticker. Nothing was re-fetched, recalled from memory, or estimated.
• Data gaps and inconsistencies, stated explicitly: (1) The consensus EBITDA estimates are NEGATIVE in every forecast year (-$209.3M FY26E through -$830.2M FY30E) while consensus EPS in the same rows is strongly positive — an internal impossibility. The EBITDA estimate fields are excluded entirely; all valuation work uses revenue and EPS. (2) The consensus EPS estimates appear to be built on a stale share count — FY26E netIncomeAvg of $620.3M against epsAvg of $2.61745 implies ~237M shares versus 323.3M actual at Q2'26, a 36% gap. This is flagged as a material caution on every forward P/E and is the stated reason this dive's base fair value ($232) sits below the street consensus ($273.73). (3) There is NO geographic segment disclosure — the field is empty — so Bloom provides no geographic reporting in this dataset and none is analysed. (4) The FY22 and FY21 product-segment rows contain only a "Financing Obligations" line ($3.3M, $10.0M) and are excluded. (5) The reported annual EBITDA line is unreliable — FY25 EBITDA of $20.1M against operating income of $72.8M, and FY24 EBITDA of $89.3M against operating income of $22.9M, do not reconcile through depreciation; operating income, gross profit and net income are used as the trustworthy series. (6) Q2'26 income-statement diluted EPS ($0.62) differs from the earnings-calendar actual ($0.78), consistent with a GAAP versus non-GAAP distinction; both are shown, and $0.78 is used for consensus comparisons since consensus is stated on that basis. (7) FY25 stockBasedCompensation is reported as $0 in the cash-flow statement despite the TTM ratio showing SBC at 2.75% of revenue — an inconsistency; the TTM ratio is used. (8) The Q3'26 consensus EPS of $0.659 sits below the Q2'26 actual of $0.78 and Q3 consensus revenue below Q2 actual — recorded as-is, and treated as a low bar rather than corrected.
• Fair-value caveat: the $154 / $232 / $309 anchors are scenario multiples on FY28E consensus EPS of $7.72196 (20x / 30x / 40x), chosen because FY28 carries the deepest coverage (25 revenue and 13 EPS estimates), cross-checked against FY28E consensus revenue of $9.754B on an EV/Sales basis. They are scenario arithmetic, not a discounted cash flow. They are also highly sensitive to the share-count assumption flagged above — the same FY28 net income (~$1.83B, implied by $7.72196 × ~237M) delivered on 350M rather than ~237M shares moves the base anchor from $232 to approximately $157.
• Timing caveat: this dive is written one week after the Q2'26 print (2026-07-28) and roughly twelve weeks before the Q3'26 print (2026-10-27), into a 36.9% drawdown with the stock 15.8% below a falling 50-DMA and 21.8% above a rising 200-DMA. The correction is explicitly assumed to be incomplete, and the staged entry is structured accordingly.
• Not investment advice. Independent research, educational and informational only, never personalized.
• Version: 2026-08-04-full.