Bloom Energy [BE] — Criteria & Two-Horizon Valuation

Phase Space Research · 2026-07-29 · Framework: Criteria (references/criteria.md, 2026-07-29), superseding the gate framework v1.5.1 · Archetype: INFLECTION

This document supersedes the Gate block in Bloom_Energy_Trade_Construction_2026-07-28.md §8 and the Gate 4 expected-return table in Bloom_Energy_Valuation_2026-07-28.md §6.2. The research work — the Oracle warrant, the scandium-sourcing record, the backlog-quality analysis and the mention-frequency series — stands unchanged and is carried forward here in full.

There is no position verdict in this document. The previous manifest carried investment_decision: "WATCHLIST"; that field is retired.

0. What moved, and why

1. Data hygiene and the Q2 10-Q — the material update

1.1 TTM revenue was one quarter stale, and the quarter was enormous

valuation.md: "Use TTM revenue, never the last fiscal year." BE satisfies that rule and still had a stale figure, because the memo was written before its own quarter was filed.

Q2 2026 alone was $1,065.365m against $401.242m a year earlier — +165.5% YoY. H1 grew +149.8%.

A second, independent defect was found in the automated TTM and is recorded here because it affects the whole corpus. coverage_scan.ttm_revenue() sums "the last four quarterly periods" selected by end-date, without checking that they are consecutive. US registrants do not tag Q4 separately in XBRL (it is derived as FY minus 9M), so for BE the function summed 2025Q1 + 2025Q2 + 2025Q3 + 2026Q1 — a gap-riddled 15-month window — and returned $1,975m. The same defect understated BLLN by 18% and overstated SMR by 62%. Every figure in this document is computed from the filings, not from that function.

1.2 The Oracle warrant — the open item is now closed, and the answer is not the one expected

The memo's disclosed limitation read: "THE $45.2m WARRANT GAP IS UNRESOLVED. Grant-date fair value is disclosed as approximately $261.3m but the carried customer-consideration asset at June 30 is $306.5m. The Q2 10-Q, not yet filed, should reconcile this. No explanation has been assumed."

It is now filed, and it reconciles. The gap is a net-settlement inducement.

$251.6m grant-date FV + $72.3m inducement = $323.9m ≈ $324.4m; less $17.9m already amortised into revenue leaves the $306.5m asset. The $45.2m gap was the inducement, net of amortisation.

Two of the existing catalyst calendar's assumptions are now falsified, and both matter.

C2 said: "~Oct 9, 2026 — Oracle warrant expiry. Exercised: BE receives $400.0m cash, ~1.1% dilution."

Neither happened. The exercise was cashless and occurred on 2026-05-01, five months before the expiry the calendar was watching. Bloom received no cash at all — the $400.0m of expected proceeds does not exist and never will. Dilution was 0.73%, not 1.1%. C2 is closed, not pending.

The economics that remain are unchanged and were correctly identified: $306.5m of contra-revenue is queued, amortising as a reduction of revenue as the Oracle Energy Servers are delivered — roughly 7.6% of a full year of FY2026E revenue, landing precisely as the growth story is delivered. $17.9m has been taken so far.

This finding was originally surfaced by tracing an unexplained new balance-sheet line rather than from any commentary. That method stands, and it is the reason the item was on the watch list to be resolved at all.

1.3 The scandium record — preserved exactly as corrected

The memo's own correction stands and is restated so it is not lost:

• The original claim — "Zero mentions of 'scandium' or 'china' across all four sampled calls" — was FALSIFIED by the completed nine-quarter transcript series and was retracted.
• The corrected, narrower claim: scandium appears in exactly one quarter of nine (2025Q1, 3 mentions; china 6 mentions), in ZERO prepared remarks in any quarter — never volunteered by management, only surfaced under analyst questioning, in the single tariff-driven quarter.
• That quarter is fifteen months before the Hunterbrook report made scandium sourcing the centrepiece of an attack. Management then said: "we are not dependent on China for scandium… we get this from multiple geographies and multiple continents… We don't reveal those [suppliers]."
• The allegation cannot be adjudicated from public filings. BE does not name input suppliers. The company's 25GW/yr figure is a supply-chain visibility claim, not a demand or revenue-capacity claim.

Nothing in the Q2 10-Q changes this. No SEC comment letter, auditor change or restatement has appeared.

1.4 Share count and net cash

Shares used: 294,527,346, the cover-page count of the Q2 2026 10-Q as of 2026-07-22. The balance-sheet count at 2026-06-30 was 293,354,001. The existing memo used 323.3m diluted from the Q2 press release; the filed outstanding count is used here and the difference (−8.9%) reduces EV by $4.8bn.

Net cash: +$4.055m — essentially zero, so EV ≈ market cap. At 2026-06-30: cash $2,666.859m + restricted $21.649m − recourse debt $2,475.390m − non-recourse $2.583m − financing obligations $206.480m. This confirms the existing memo's ~$4m to the decimal, computed independently from the filed balance sheet.

2. The Criteria

2.1 Quality Criteria — INFLECTION standard

BE sits exactly on the archetype boundary and the boundary is worth stating rather than resolving silently. TTM operating margin is 11.32%, marginally above the 10% COMPOUNDER threshold. It is classified INFLECTION on character — margin expanding fast, growth accelerating, value in the terminal period — which is the stricter test here, because INFLECTION requires a gross-margin level that COMPOUNDER does not.

The binding limb is gross margin, and the honest reading is that BE fails a test built for a different kind of business. The ~50% standard proves software-like unit economics. Bloom Energy manufactures solid-oxide fuel-cell hardware; 31.5% is a strong hardware gross margin and it is rising (product gross margin went 33% → 37% year on year). The result is reported as FAIL because that is what the stated rule returns on a present, verified input — it is not INDETERMINATE, because nothing is missing. Whether the INFLECTION gross-margin limb should carry a hardware carve-out is a calibration question, not a licence to override the rule here.

Cheap cannot rescue a failure here — but note that BE is not cheap, so the two BINDING criteria agree.

3. Valuation Criteria — the two mandatory outputs

3.1 The implied-path test

What the price requires: a 41.3% revenue CAGR for five years.

Demonstrated: 33.1%. Margin = 33.1 − 41.3 = −8.2pp.

Demonstrated CAGR window: FY2022 $1,135.346m → TTM 2026-06-30 $3,090.770m, 3.5 years. Stated explicitly because the window choice moves the answer: measured FY2022→FY2025 only, the demonstrated rate is 20.8% and the margin widens to −20.5pp. The TTM endpoint is used because valuation.md requires TTM.

Exit multiple basis: GROWTH_MATCHED. Drawn from 33 names in the 1,433-name universe scan whose demonstrated growth brackets BE's 33.1% (band ±50%, so 16.6%–49.7%), median EV/EBIT 33.2x. This is a real comparator set with genuine growth dispersion — unlike the diagnostics anchor set that spanned 1.0–7.5% growth and was used to value 15–39% growers.

Implied compression, as a number: BE trades at 140.6x TTM EV/EBIT today. The exit multiple is 33.2x. That is a compression of 107.4 turns, −76.4%.

3.2 Sensitivity — over the exit multiple, never over scenario probabilities

The instrument this replaces ran its range across the bear weight (0% → 40%) and reported "fails at every bear weight including 0%" — a range over the parameter that could not change the answer. The multiple is the parameter that decides it.

The flip point is a 44.7x exit EV/EBIT. Above it the price is justified by what BE has already demonstrated; below it, it is not.

This is a strong independent cross-check on the retired instrument. The old Gate 4 arithmetic — different horizon, different basis (FY2029 exit P/E on EPS), different discount rate — computed a break-even base exit multiple of 40.1x. The reverse DCF computes 44.7x on EV/EBIT. Two unrelated instruments landing 4.6 turns apart is meaningful corroboration that the level of the required re-rating is real, even though the magnitude of the failure changed a lot (16.4pp → 8.2pp) once the revenue base was corrected.

Valuation Criteria: FAIL. The price requires more than demonstrated. It is not a PASS WITH ARGUMENT: the specific, evidenced arguments available — the Oracle relationship and the backlog — are both compromised as variants. The Oracle volumes arrive with $306.5m of contra-revenue attached, and the backlog was quantified once, in February 2026, with no weighted-average duration, no 12-month conversion percentage, a definition that includes anticipated ITC tax credits, and roughly 70% of the $20bn headline being multi-decade O&M. An undated, unupdated, definitionally-inflated backlog is narrative for this purpose.

3.3 The 12-month target

NTM revenue: $4,995m = 5/12 × FY2026E $4,050m + 7/12 × FY2027E $5,670m. FLAGGED: FY2026E $4,050m is the company's post-Q2 guidance midpoint, which the memo records as equal to consensus. FY2027E $5,670m is the house base case — no independent FY2027 consensus figure is documented in the file, and none has been invented. The target is sensitive to that leg.

EV / NTM revenue today: 9.85x.

Own-multiple anchor and percentile. Over 892 sessions (from 2023-04-17), BE's EV/Sales on an as-known annual revenue basis ranges:

BE trades at the 88th percentile of its own three-year multiple range.

The median is NOT used as the target anchor, and the reason is the framework's own rule. BE's own history is dominated by a period when it grew ~20% a year; it now grows 165%. An exit multiple "may only be drawn from a comparator set whose growth brackets the subject's" — and BE's own past does not bracket BE's present. Using the 4.6x median would be the same identification error as the diagnostics anchor set, committed against the name's own history instead of a peer's.

The target is therefore reported as a band across the name's own observed multiple range, with no point estimate:

12-month target: $56 – $190. Spot sits near the top of BE's own historical multiple band applied to next-twelve-month revenue.

Sanity band against the Street. Consensus target is $289.63 mean (TipRanks); JPMorgan $346, raised from $267. The house band's top ($190) sits 34% below the Street mean. The divergence is entirely a multiple disagreement and it is stated as such: there is no numbers disagreement — house FY2026E, Street FY2026E and company guidance are the same $4,050m. The Street is capitalising FY2027–28 earnings at 50–80x; this band capitalises NTM revenue at BE's own 3.3x–11.2x. Which is right is a judgement about the multiple, and this document does not claim to resolve it — it states where the disagreement lives, which is the point of the exercise.

4. Downside Criteria — MEASURED, blocks nothing

This is logged and scored, and it does not reject the name. Inverse-volatility sizing is the active protection: BE's realised volatility is 111.4% and its beta 3.14 (Blume-adjusted), so a volatility-scaled book sizes it down automatically.

5. Momentum Criteria — MEASURED, entry timing only

Momentum governs when, never whether. All veto and blocking language is deleted.

The configuration is extreme on both horizons at once, and the existing memo's reading of it is preserved and remains correct. Top-percentile 12-month momentum, combined with 111% realised volatility and a price below its own 200-day, is the exact configuration in which momentum strategies crash (Daniel & Moskowitz 2016): momentum's worst drawdowns cluster in high-volatility states following a trend break. The momentum reading is top-percentile and maximally fragile. These are not in tension — they are the same fact.

RSI-14 at 27.5 is the most oversold reading across the four names in this batch. On the reference book's observed pattern — longs cluster oversold, shorts overbought — that is a constructive entry-timing signal and nothing more.

6. What this document deliberately does not say

• No Long, Short, Watchlist or Avoid.
• No E[R] versus a 4.7% cash hurdle. The identity error the memo-critic found in that table (E[R]_reported = (E[R]_true − Ke)/(1+Ke), worth 15.0 percentage points a year running in the name's favour) is moot: the instrument is retired.
• No 5-year DCF-derived price target.
• No claim about scandium sourcing beyond what the filings support.
• No assumption that the Oracle warrant brings cash. It did not.

7. Provenance