Built by Grok 4.6 High

Bloom Energy Corporation reported third-quarter 2026 results on Thursday, October 22, 2026. This analysis was prepared Monday, October 26, 2026. Quarter ended September 30, 2026. Fiscal year ends December 31, so this print is 2026Q3. Next print is Q4 2026, expected February 4, 2027.

BE | Earnings Review — Q3 2026

Verdict: new highs on the level, deceleration on the rate. Q3 is the high of the twelve-quarter window on revenue, non-GAAP gross margin, adjusted EBITDA, and non-GAAP EPS. It is also the first 2026 print in which the year-over-year growth rate fell. Revenue went +130.4% → +165.5% → +138.1%. Product, which did 96% of the dollar growth, went +208% → +215% → +179%. The raised full-year range embeds another step down in Q4, toward roughly +73% to +99%. A company still growing this fast is not a broken quarter. It is no longer an accelerating one.

Print. Revenue was $1,235.8 million, +138.1% versus $519.0 million, and +$165.1 million / +15.4% versus FMP. Non-GAAP diluted EPS was $0.94, +526.7% versus $0.15, and +$0.249 / +36% versus FMP $0.691. Non-GAAP gross margin was 34.5%, +410 bps versus 30.4%. Adjusted EBITDA was $299.5 million, +407% versus $59.0 million, a 24.2% margin (+1,286 bps). GAAP diluted EPS was a profit of $0.78 against a $0.10 loss a year ago. Product was $1,073.6 million, +179.4% and 86.9% of the quarter. Service was only +15.3%.

Why EPS beat by more than revenue. Non-GAAP operating expenses were $140.5 million, +26%, against revenue +138%. Non-GAAP operating income was $285.7 million, +518% versus $46.2 million, and operating margin was 23.1% versus 8.9%, +1,420 bps. Product gross margin 38.4% expanded +250 bps while they were ramping lines and expediting. Service margin 21.4% expanded +700 bps on a small base. Installation margin went the other way, to −3.5%.

Guidance. Third raise this year. Each new low clears the prior high. FY2026 revenue is $4.4 billion–$4.6 billion, midpoint $4.50 billion, +122% versus FY2025 $2,024.0 million. Non-GAAP EPS is $3.10–$3.40, midpoint $3.25. Non-GAAP operating income is $960 million–$1,030 million, midpoint $995 million, ~22% margin. Gross margin was held at ~34%. There is no quarterly guide and no FY2027 P&L guide. The incremental operating-income raise is getting thinner: 56% → 39% → 32% drop-through on each revenue-midpoint step. Nine-month revenue is already $3,052 million, so the entire $450 million revenue raise sits in Q4. FMP’s Q4 revenue of $1,406.9 million (~+81%) sits inside the implied $1.35–$1.55 billion range, about $41 million under the midpoint. The street is not above the company. It is camped in the low third.

Beat pattern. Last twelve quarters are mixed (8/12) on both revenue and non-GAAP EPS. Last four are 4/4. Dollar surprise expanded into Q2 (+$239 million revenue, +$0.37 EPS) and narrowed this quarter (+$165 million, +$0.25). Still the third-largest dollar beat of the twelve. The street is catching a business it under-modeled, not watching a miss.

Cash is the line that is not on the P&L slope. Quarterly operating cash flow swung from a source of $19.7 million to a use of $27.1 million. Current contract assets are $792.8 million. Edwards would not give a collection window. Year-to-date operating cash flow is still a source, and cash of $2,486.7 million is roughly in line with year-end 2025 $2,454.1 million despite quarterly capex of $153.5 million. Judge the year, as he asked, and do not ignore that the quarter’s free cash flow (calculated) flipped negative while revenue made a high.

Tone. Wider commercial narrative (Army, AEP/Google carbon capture, Texas, front-of-the-meter utilities, international) and a hotter demand claim. The pins the street asked for were refused: no gigawatt timeline, no backlog growth rate, no confirmation that delivery is still inside 90 days, no collection window. The live catalyst into February is the still-unpublished backlog versus FMP’s $7.16 billion FY2027 average, not another easy revenue beat.

Contradictions, four live. Capacity numbers given in 2025 and April were disowned. The water comparison does not tie to itself. The opening operating-income guide exists as three spoken or filed ranges; model the filing. The Oracle clock is 9 months on the July call and 15 months on this one. Two older transcript defects (Q1 “+13.4%” and a reset service-margin streak) are not live.

What this is not. Not a price note. Bloomberg and Visible Alpha were not connected; consensus is FMP, pulled 2026-10-26. SharePoint, OneNote, Outlook, and Excel were unavailable and were skipped. No FY2026 annual FMP row exists; nothing was backfilled.

Product is the growth engine. Installation is project timing. Service is a mid-teens fleet annuity. Electricity is a shrinking residual under $10 million. Same-quarter year-over-year only. The eight quarters below are the regime that matters: exit from decline, trough, breakout, peak, first deceleration. The twelve-quarter window starts one year earlier, with revenue still negative through 2024Q3 (−17.5% on $330.4 million).

Revenue drivers

Product YoY +208.4% → +215.4% → +179.4% is the operating tell. The Q3-versus-Q3 stack is still much faster than last year’s +64.4%, and the last print stepped down. Mix did not shift toward the slow lines: product share is stable near 87%. Service re-accelerated to +26.8% in Q2 and gave it back to +15.3%. Service margin expansion is narrowing (+1,320 → +980 → +700 bps) even though the level is far above last year’s 14.4%. The 2024Q4 product margin of 46.9% was an outlier; do not read 2025’s decline against that quarter as a break. Q3 installation was $84.9 million, +29.0%. Electricity was $9.8 million, −5.2%.

Consolidated

Gross-margin expansion slowed from +610 bps to +410 bps and is still expansion. The level is 31.5% → 34.3% → 34.5%. The 2024Q4 margin of 39.3% is why 2025Q4 shows −740 bps; the level did not break. EBITDA margin is still up (19.0% → 23.8% → 24.2%) and the pace of expansion has flattened. GAAP has been profitable for three straight quarters. The non-GAAP premium this quarter is $0.16. The awkward gap was 2025Q4, when GAAP EPS was $0.00 against non-GAAP $0.45.

Five completed years

FY2026 is incomplete. Annual revenue growth decelerated for two years and then re-accelerated: +23.3% → +11.2% → +10.5% → +37.3%. The 2025 re-acceleration was product (+41.1%) and installation, not service (service slowed to +6.9%). Annual non-GAAP gross margin has expanded every year since 2021, and the 2025 step (+160 bps) was the smallest. Annual adjusted-EBITDA growth has slowed for two years (+171% → +96% → +69%) while the margin kept expanding by less.

Mixed over twelve quarters. A consistent beater over the last four. The latest dollar surprise is narrower than the Q1–Q2 peak.

This quarter beat FMP on both scored lines. Revenue $1,235.837 million versus $1,070.774 million is +15.4%. Non-GAAP diluted EPS $0.94 versus $0.691 is +$0.249 / +36%. Sixth straight revenue beat. Eighth straight non-GAAP EPS beat. FMP’s EPS actual matches non-GAAP diluted, not GAAP. GAAP $0.78 is also above $0.691 and is not the scored line. There is no FMP gross-margin or operating-income consensus, and Bloom does not guide the quarter.

The revenue gap is systems accepted. Edwards was explicit that product revenue is recognized when control transfers. Service (+15.3%) and installation (+29.0%) did not drive the beat. He did not call it a one-time item and he did not walk back demand. The offset he discussed was cash, not the P&L. The third raise is the tell that he treats the quarter as real, not as pull-forward he expects to give back. Gross margin was held near 34% because of mix, ramp cost, and expediting, not because product pricing faded.

Heatmap — last eight quarters

Revenue cells are the percent surprise. EPS cells are the dollar surprise, because several EPS estimates sat near zero. Green is a beat. Red is a miss. Darker green is a larger surprise (revenue at least +20%, EPS at least +$0.25). 2026 Q3 is outlined.

Eight-quarter bridge

Consensus is FMP and is not a Daloopa figure. Dollars are millions except EPS.

The early misses were real. 2023 Q4 revenue $356.9 million missed by about $119 million (−25.1%). 2024 Q3 $330.4 million missed on both lines. 2025 Q1 missed revenue by $59 million even as EPS beat a loss estimate. The pattern broke after that quarter.

Block averages are still improving. Inside the latest block the path is +$129 million / +$0.14, then +$211 million / +$0.32, then +$239 million / +$0.37, then this quarter +$165 million / +$0.25. The surprise expanded into Q2 and narrowed in Q3. It did not narrow back toward a miss. The next print is the test of whether Q3 was the street closing the gap or the start of a smaller-beat pattern.

Guidance is full-year non-GAAP only. No quarterly guide. No FY2027 revenue, margin, or EPS. No live EBITDA, capex, or cash-flow guide. The lines that moved are revenue, non-GAAP operating income, and non-GAAP EPS. Gross margin was held. The new low clears the old high on all three moving lines. That is a real raise, not a slide inside a stale range.

Full-year guided growth is still accelerating versus 2025 and versus each prior guide (about 60% → 78% → 100% → 122% at the midpoint). Quarterly YoY already peaked in Q2. Those are not contradictory. The dollar add went about $425 million → $664 million → $717 million, and the guide-implied Q4 add is about $670 million — holding the Q2 add against a much harder compare, not extending it. Margin expansion in basis points also peaked in Q2. The guide locks the gross-margin level and gives up further expansion. That is the number to watch, not the revenue raise.

Waterfall — FY2026 midpoint, vintage by vintage

Revenue in billions. The yellow bar is not a published FY2026 consensus. FMP has no FY2026 annual row. It is nine-month actuals plus the FMP Q4 estimate, labeled as a construct. EPS stops at the new guide because there is no FY2026 street EPS to put on the bridge. FY2027 street $5.29 is a different year and is not drawn here.

Guide path

Revenue figures are billions. The February operating-income range in the table is the filed range, not either spoken substitute (see contradictions). Drop-through on the revenue-midpoint raise: April +$400 million of revenue carried +$225 million of operating income (56%), July +$450 million carried +$175 million (39%), October +$450 million carried +$145 million (32%). Still above the 22% margin they are guiding, so they are not spending the raise away. They are also not letting incremental margin stay at the Q3 print.

The Q1 adjusted-EBITDA range of $650 million–$800 million was not reaffirmed. Year-to-date EBITDA is already $143.0 million + $253.4 million + $299.5 million = $696 million, through the old low with a quarter left. Stale, not a live bar. The February CFOA figure was not updated in July or October. Cash guidance was withdrawn. Do not give the P&L record to the cash line.

Q4 is implied, not guided

Nine months are booked. The whole raise drops into Q4. The low end is not a sandbag against the Q3 print: Q4 revenue at the low ($1,348 million) is already above Q3, Q4 operating income at the low ($305 million) is already above Q3’s $285.7 million, and nine-month EPS of $2.16 plus a Q4 that only repeats $0.94 equals $3.10, the new low. Hitting the bottom requires a bigger Q4 than Q3. The top is a handful of year-end acceptances. Edwards said so.

FMP Q4 revenue is about −2.8% versus the implied midpoint and inside the range. FMP Q4 EPS of $1.05 is about $0.04 under the residual midpoint. Street YoY of +81% revenue is another ~5,720 bps of deceleration from this quarter’s +138% (+138.1% to +80.9%). The implied-midpoint step, to about +86%, is the ~5,200 bps figure. FMP is underwriting that slowdown. A result inside $1.35–$1.45 billion can be “in the guide” and still miss FMP.

Holding the year at 34% after a 34.5% quarter implies Q4 gross margin a touch above 34.5% on the arithmetic of a $4.50 billion year, not a collapse. Edwards held the year because he will not underwrite more than that on Q4 mix, expediting, and new-line startup cost.

FY2027 — no company number

Sridhar refused to bracket 2027 growth and refused a backlog growth number. Both are promised for the year-start call. Edwards called the gross-margin stance “measured”: history is 28.7% in 2024, 30.3% in 2025, 34.5% in Q3, but new lines and expediting carry near-term cost. That is not a 2027 raise.

FMP’s FY2027 row (year ended 2027-12-31): revenue average $7.164 billion (20 analysts, $5.22–$8.05 billion), +59% versus the $4.50 billion 2026 midpoint. EPS average $5.291 (16 analysts), +63% versus the $3.25 midpoint. EBITDA average $1.83 billion. The street’s range is the debate. Management has not endorsed the average. Treat $7.2 billion and $5.29 as unanchored until February.

Tone versus July. The humility line on AI spend was not repeated. Confidence is now “the cycle and the adjacent markets are ours to take,” plus a company survey that a quarter of C&I sites are fully on-site by 2030. Cash tone flipped from showcase to explanation. The risk he is willing to name moved from customer identity to year-end delivery timing. The range is now only about 4% of the midpoint, down from 11% in April. Credibility on the P&L: three for three on raising, each time with the new low above the old high. The opening February number did not describe the year — the original revenue high is 33% below the new low — so the record is “early,” not “precise from the start.” From here the guide is more informative. They have stopped guiding the lines they are no longer willing to defend.

The revenue YoY rate peaked at +165.5% in Q2 2026 and decelerated 2,742 bps to +138.1% in Q3 2026. Product deliveries account for 96% of the dollar growth ($689 million of $717 million). Non-GAAP EPS YoY slowed to +526.7% for a second quarter, while the dollar EPS increase is still widening. Accel below is the change in the YoY rate versus the prior quarter’s YoY rate, in basis points. It is not sequential revenue growth.

† Small base. Q4'24 EPS sits on $0.07. Q1'26 sits on $0.03. Ignore those percents. The clean EPS read is Q4'25 at +4.7% on a real $0.43 base, then +680% → +526.7% (−15,333 bps). Unrounded revenue accel for Q3 is −2,742.0 bps. Product decelerated harder than the total, −3,601 bps.

Inflection points

The first derivative of dollars has not turned negative. The second derivative has. Q4, on both the company’s range (+73% to +99%) and FMP (+81%), is set up for another step down in the rate. Sridhar said growth is accelerating and that the year went from a 60% expectation to over 120%. That is true of the full-year step-up. It is not true of the quarterly YoY rate, which slowed this quarter and, on their own range, slows again in Q4. Edwards was more precise: they guide the year because quarterly cadence is customer readiness, and a small number of year-end deliveries decides where in the range they land.

The quarter did the easy part. What consensus believes about the next quarter is already a sharp deceleration. The live catalyst into February 4, 2027 is not another 15% revenue surprise. It is whether the still-unpublished 2026 backlog — last hard number, product $6.0 billion and service $14.0 billion at year-end 2025 — is still growing faster than revenue, and whether the first 2027 outlook looks like FMP’s $7.16 billion or like the utility, carbon-capture, and international channels management is describing and not yet numbering.

Texas is the only one of the five new commercial vectors that is a this-quarter shipment, and it has no customer name. Army leases, international partners, and front-of-the-meter utility procurement are narrative until a megawatt and a year are attached. Brookfield’s $25 billion framework was not mentioned. Oracle was a speed anecdote, not a new campus. S&P 500 inclusion already happened and is not an operating catalyst. Do not put 2.8 GW of Oracle framework, 700 MW of AEP, or 300 MW of Nebius into the 2026 revenue bridge.

FMP grades are 17 buy / 12 hold / 3 sell. Enthusiasm is already the median posture. The upside case has to be a backlog or a 2027 guide above the $7.16 billion average, not a re-rating from “nobody believes the quarter.”

Ten analysts got on. The hour was not a re-litigation of the beat. It was an attempt to turn the raise into units: factory gigawatts, backlog growth, days to power, utility TAM, and when the quarter’s working-capital use comes back as cash. Tally: 7 well answered, 8 deflected. The answers that landed describe the platform. The pins that underwrite the raise were refused.

What was answered, and is worth keeping. Gross margin stays at ~34% on purpose: mix, expedited deliveries, and cost programs that have not hit COGS yet, with a qualitative bias that scale helps later. Carbon capture was given a live project, a 10–15% cost band hyperscalers are described as willing to pay absent a mandate, and three ways to get paid (the gas customer bolts capture on, a partner splits the chain, or the power buyer keeps the credit). Site scale runs from 150 kW to the 100 MW class, with the AEP/Google project pointed at about 700 MW. Stack-line labor was described as flat at about 600 people from 100 MW through a 5 GW case. International demand was tied to higher gas prices, which makes the efficiency gap worth more, not less.

What a model still cannot update.

1. Capacity, twice. No timeline, and no yes to 10 GW. The 250–300 MW per quarter premise was retired. The plan is a dial that is not going down and will not be dated. Same refusal as July, now sitting next to a third guide raise.
2. Backlog growth. Mandloi asked if “faster than revenue” means something like this quarter’s +138%. Sridhar declined even a directional bracket. Daloopa has no 2026 quarterly backlog. The last product backlog, $6.0 billion, is nine months stale and was +140% versus $2.5 billion at year-end 2024. That +140% is last year’s product-backlog growth, not a Q3 2026 figure.
3. Cash collection, twice. The use is not mysterious in the statement: contract assets $430.3 million, inventory $170.4 million, receivables $175.2 million. He would not say in which quarter the $792.8 million bills. Year-to-date operating cash flow of $272.9 million keeps the self-funding claim intact for the year. It does not tell you whether Q4 collects.
4. Delivery speed. Servers for 300 MW at Nebius are ready. The first block’s megawatts were not given. The sub-90-day claim was not reconfirmed under the question.
5. Utility TAM. The rate-base logic is clear and, if it is real, it is a second channel the street is not modeling as a procurement cycle. Size, timing, and a second named utility were not given.
6. Token cost and water. Neither the 9% savings claim nor the gallon comparison should enter a model from this call. The water answer is internally inconsistent (see below).

Four live contradictions. Guidance raises are not contradictions; they were labeled as raises and the filed ranges match the later calls. Two older transcript defects are noted and are not the Q3 story. The print itself — revenue, margins, and the third raise — does not fight the Q2 call or the filing. The contradictions are about the record around that print.

Not live. Q1 prepared remarks said revenue was "up 13.4%" and, in the next sentence, the first quarter of greater than 100% growth. The filing is +130.4% on $751.1 million versus $326.0 million. Do not model 13.4%. The Q4 2025 call called service margin the "first" double-digit quarter after the prior call had already called it the second. Later counts match the filing: Q3 2026 at 21.4% is the sixth double-digit quarter and the second above 20%. Those are mirror defects, already contradicted by the next call and by Daloopa.

Examined and not contradictions: the three revenue raises, the gross-margin hold, 39.3% as a 2024 fourth quarter rather than the 2024 year, a 2 GW capacity ceiling versus a ~2x revenue guide, "fifth double-digit" versus "second quarter above 20%," backlog growing faster than revenue versus one customer leading a quarter, the international emphasis shift, Fremont real estate versus "we will not be the bottleneck," and the Q2 informal cash baseline versus one quarter of cash use. A single quarter of working-capital use does not break a full-year baseline they had already refused to guide. Watch Q4. Do not treat the silence on "$375 million" as a cut they announced.

This is not a rates, CPI, or consumer call. Nothing was said about the Fed, household confidence, retail spending, or tariffs. The macro is electricity scarcity: AI load the grid was not built for, a widening on-site shift, utilities trying to keep the rate base, and international markets where gas is expensive. Q3 broadens the buyer set from hyperscale campuses to AEP/Google with carbon capture, Nebius execution, the Army, and a first Texas shipment. Brookfield and the prior lighthouse names were not updated. Absence is not a downgrade. It means no new dollar and no slip was disclosed.

Self-funding is the financing comment. Edwards said the business is self-funding, which lowers dependence on external capital. He did not describe customer demand as rate-sensitive, and he did not say project-finance markets loosened or tightened. July was organized around other people’s balance sheets. October’s capital story is Bloom’s own. Pricing "remains strong" while they expedite: product gross margin 38.4% versus 35.9%, +250 bps, in a quarter of new lines. They are not cutting price to hold the higher revenue guide, and they are not passing a cost shock through. The gross-margin guide stays below the Q3 print because ramp cost has not finished flowing through COGS.

Power-demand tone versus July. The delay-cost rhetoric stepped up and is now attributed to a paraphrase of NVIDIA’s CEO at Berlin GTC ($10 billion of missed customer revenue per quarter of delay on a 1 GW data center). Do not mix that with July’s $12–24 billion per GW-year; they are different claims, and the Berlin line was not checked against an NVIDIA transcript. Utilities flipped from foil to channel: if they do not procure, they lose the customer and the rate base. That is better for a utility that copies AEP and worse for a central-station-only plan. It is not "fuel cells replace the utility." Competition got sharper against combustion on carbon capture, air permits, and water, and still no competitor was named. July’s "very high 90s" fuel-cell share claim was not repeated. Do not carry it forward as an October statement. The same week AEP bought the Bloom project it also bought Mainspring linear generators. The competitive fact pattern is dual-sourcing, not displacement.

Named last quarter, not this quarter

Cohorts with no ticker: turbine and engine suppliers are worse, directionally, on carbon capture, permits, and urban siting, and are not facing a 2026 volume cut Sridhar was willing to claim. The next regulated utility that copies AEP is better; one that waits and loses the load behind the meter is worse. "Significant wins" has no counterparty and no date. Oil, gas, and CO2 infrastructure are better if they already have a sequestration well or a pipe, because Bloom claims a concentrated CO2 stream combustion does not. No producer and no 45Q math. International partners are option value only: a near-term win was previewed and not delivered.

Do not turn 700 MW, 300 MW, or 5 GW into revenue. Do not read Brookfield’s absence as a lost project. Do not use the transcript’s combined-cycle water gallons. The only revenue facts in this section are the Daloopa figures at the top of the review.

Bottom line. Hold. The levels are records, the P&L guide was raised for the third time with the new low above the old high, and the last four quarters are a consistent beat. The trajectory that the framework weights more heavily has turned: revenue YoY −2,742 bps off the Q2 peak, product −3,601 bps, gross-margin expansion +610 bps to +410 bps, and beat dollars narrower than Q1–Q2. The guide itself embeds Q4 at about +73% to +99%, and FMP is already there. The contrarian thread — backlog growing faster than revenue, plus a utility and carbon-capture channel the 2027 model may not be built as — is exactly what management would not number. Until the February backlog or a 2027 outlook is in hand, that thread is a watch item, not a reason to treat a decelerating rate as an accelerating story.

Data sourced from Daloopa.