Cummins CMI

> ## EARNINGS BANNER — REPORTED THIS MORNING

> Cummins released second-quarter 2026 results on 2026-08-04, the day of this dive, and filed both the 8-K and the 10-Q the same day.

> Diluted EPS $6.73 against a $7.21 consensus — a 6.7% MISS. Revenue $9,457M against $9,330M — a 1.4% beat.

> Full-year guidance RAISED to revenue growth of +10% to +13% and EBITDA of 18.0-18.5%. Quarterly dividend RAISED from $2.00 to $2.20, the 17th consecutive annual increase.

> The shares opened at $596.50 (−8.1%), traded down to $577.72 (−11.0%), and closed at $635.94 (−1.99%) — a 10.1% recovery off the low on 2.15x normal volume.

> The verdict below is set with that timing as an explicit input. A dive struck on the day of a violent, two-directional repricing carries more uncertainty than one struck in a quiet week, and the Watch designation reflects it.

Reference table

1. What the business is, and today's segment detail

Five reportable segments. Chief executive Jennifer W. Rumsey; approximately 67,400 employees; headquartered in Columbus, Indiana; founded 1919.

June 2026 quarter, from the earnings release filed 2026-08-04 ($M):

Read that table and the whole dive is visible in it. Four of five segments grew sales; three of five saw EBITDA margin fall. The one that expanded is Power Systems, and it expanded by 170 basis points to 24.5% — nearly double the margin of Engine or Components — on 19% growth. That segment is 19.3% of total segment sales and it is the entire reason this stock trades at 18.6x forward earnings rather than the 11-14x a heavy-duty engine manufacturer has historically commanded.

Management's explanation for the consolidated margin decline is specific and self-limiting, and we quote it because it is the pivot of the second-half guidance: "EBITDA percentage for the company and some of the individual segments declined year-over-year. This was primarily due to higher incentive compensation, tied to expected record full year results. EBITDA percentage for the company is expected to be higher in the second half and full year 2026 compared to the same periods in 2025." That is management framing and is weighted at half. It is also falsifiable within one quarter.

The segment drivers, verbatim from the release:

• Power Systems — "Revenues in North America increased 19% and international sales increased 19% driven primarily by increased power generation demand, particularly for data center markets in the United States, China and Asia Pacific."
• Distribution — "Revenues in North America increased 13%... driven by increased demand for power generation products, particularly for data center applications."
• Engine — "Revenues in North America increased 1% and international sales increased 23% due to stronger construction demand in China." North America up ONE percent.
• Components — "Revenues in North America increased 6% and international sales increased 8%."

Two of the five segments are being driven by data centres and the release says so in both cases. The North American truck business grew 1%.

The corporate actions the payload does not contain

The 8-K scan produced four material items, none of them in the vendor data.

First, the fuel cell disposal. From the release's segment footnote: "In the first quarter of 2026, we sold our low pressure fuel cell business to a customer, cancelled future commitments and resolved certain claims against us with that customer resulting in a net payment by us of $175 million. These transactions resulted in a net charge of $199 million which is reflected in other operating expense, net." Cummins paid $175 million to be rid of a business and took a $199 million charge for which no tax benefit was recognised. It sits inside the Accelera segment's first-half EBITDA loss of $346 million and inside first-half "Other operating expense, net" of $301 million against $74 million a year earlier. The vendor's inc_q reports the resulting first-quarter figures without any indication of what caused them.

Second, the dividend increase. "Cummins announced an increase in the quarterly common stock cash dividend from $2.00 to $2.20 per share. The company has increased the quarterly dividend to shareholders for 17 consecutive years." $8.80 annualised, a 1.38% forward yield. The vendor's dividendPerShareTTM of $8.00 and lastDividend of 8 are trailing and now stale.

Third, the May 2026 Analyst Day. "Cummins hosted its 2026 Analyst Day and raised its 2030 financial targets... The company also announced disciplined capacity and product investments to further strengthen its position in mining and power generation." Half-weighted as management framing, and noted as a gradeable commitment.

Fourth, and most consequential, the Model Year 2027 change. "Cummins announced updated Model Year 2027 North American on-highway product launch plans following the EPA's proposed changes to upcoming emissions regulations. The company will introduce its new X10 and X15 engines through a measured production ramp while maintaining select legacy product availability in 2027." This is discussed in Section 6c because it is the principal variant perception in this dive.

The 8-K filed 2026-05-14 is an Item 5.07 annual-meeting vote report — eleven directors elected, say-on-pay, PwC ratified, a 2026 Omnibus Incentive Plan approved, and two shareholder proposals (separation of Chairperson and CEO roles; a report on charitable support). No corporate action.

2. The trajectory, and what the trailing numbers hide

The GAAP series is unusable without the events behind it and this is why the consensus row is shown alongside. FY2023's $5.15 reflects the emissions settlement charge; FY2024's $28.37 includes a $1,333 million gain on the Atmus divestiture (visible in the vendor's own FY2024 cash-flow reconciliation); FY2025's $20.50 is the cleanest of the three. est.epsAvg tracks an adjusted basis and diverges from GAAP by up to 165% in FY2023 and 28% in FY2024 — which is exactly why this dive values on epsAvg and shows the trailing GAAP multiple separately rather than mixing them.

Quarterly, with the actual-versus-estimate record:

Four consecutive beats of decreasing magnitude — 22.9%, 15.7%, 13.9%, 9.2% — and then a miss. That is the classic shape of consensus catching up to and then overshooting a re-rating story, and it is the most useful pattern in the earnings calendar. The stock's 11% intraday drawdown and 10% recovery is the market working out which of those two readings is correct.

Note that the Q3 2025 through Q1 2026 rows carry two EPS figures — the vendor's inc_q GAAP diluted EPS and the earn_cal actual — and they differ by $1.44 to $1.74. The earn_cal figures are on the adjusted basis consensus uses. The June 2026 quarter is the exception: $6.73 appears in both, because the quarter contained no adjusting item large enough to separate them, only the $0.21 of discrete tax. That is worth knowing before comparing $6.73 to $7.21: the miss is a genuine miss, not a basis mismatch.

3. Cash flow, capital and the checks

The capital-expenditure check: VERIFIED CLEAN

The 10-K states capital expenditures of $1,235 million (FY2025), $1,208 million (FY2024) and $1,213 million (FY2023) on the face of the consolidated cash-flow statement, and separately in the MD&A: "Capital expenditures were $1.2 billion each year in 2025, 2024 and 2023." The vendor's investmentsInPropertyPlantAndEquipment reads −$1,235M, −$1,208M and −$1,213M. Identical to the dollar in all three years.

The trailing recomputation:

The vendor's freeCashFlowYieldTTM of 3.8369% is reproduced exactly, as are capexToOperatingCashFlowTTM (0.2754 against our 1,280 ÷ 4,647 = 0.2754) and capexToDepreciationTTM (1.1429 against our 1,280 ÷ 1,120 = 1.1429). Following the MPC lesson, the diagnostic and the datum were checked independently against the filing; both pass. No repair applied. This is the second clean capex verification in this batch and it is reported as a finding.

Forward capital intensity is rising and the company says so: the 10-K guides FY2026 capital expenditure to $1.35-1.45 billion, up 9-17% on FY2025, "with over 65 percent of these expenditures expected to be invested in North America." That is the capacity being built for the power-generation demand, and it will compress free cash flow before it expands it.

Where the filing wins: total debt

The 10-K's working-capital table states total debt of $7,552 million at 31 December 2025 (against $7,059 million a year earlier), and total debt as 36.0% of total capital. The vendor's totalDebt for the same date is $8,114 million — 7.4% higher. The reconciliation:

This is the lease-inclusion defect class again, at a milder 7.4% than Sherwin-Williams' 19.1%. The filing wins and every leverage figure in this dive uses $7,552 million.

At 30 June 2026 the equivalent build is: loans payable $459M + commercial paper $348M + current maturities $151M + long-term debt $6,737M = $7,695M, against cash of $3,179M and marketable securities of $745M — $3,924M of total liquid assets, for net debt of approximately $3,771M. That is well under one turn of EBITDA and is the strongest single feature of this balance sheet.

The enterprise value carries two errors that nearly cancel

enterpriseValueTTM reads $92,268,767,268 against a market capitalisation of $87,752,767,268 — an implied net debt of exactly $4,516 million. That figure is $7,695M of debt less $3,179M of cash and cash equivalents. Two things follow:

1. The $745 million of marketable securities is omitted from the cash deduction. This is the known short-term-investments omission. bal_a carries shortTermInvestments of $764M at 2025-12-31, so the data to make the deduction is in the payload and was not used.

2. The $1,059 million of noncontrolling interests is omitted from enterprise value. Cummins carries $1,059 million of NCI on the June 2026 balance sheet, unchanged from December, and it generated $36 million of net income in the quarter and $62 million in the half. This is the confirmed NCI-omission class — KKR at 35.6%, MPC at $6,772M, FCX at 11.5% — and Cummins' instance is small at 1.15% of enterprise value.

Rebuilt: $87,753M market cap + $7,695M debt − $3,924M cash and securities + $1,059M NCI = $92,583M. Against the vendor's $92,269M, a difference of $314 million, 0.34%. The two errors run in opposite directions and very nearly cancel — the payload arrives at approximately the right enterprise value for the wrong reasons, and both errors are reported rather than the small net difference being used to wave them through. Because the residual is well under the 15% withholding threshold, EV-based multiples are usable here, and evToEBITDATTM of 18.6x is quoted with the correction noted.

netDebtToEBITDATTM of 0.909x is computed on the mis-specified $4,516M. On the correct net debt of $3,771M the ratio is approximately 0.76x. Either number supports the same conclusion — this is an under-levered balance sheet — and the corrected one is used.

Capital return, and the pause that just ended

Cummins repurchased no stock at all in FY2023, FY2024 or FY2025 — the cash-flow statement has no line for it — and resumed in the first half of 2026 with $468 million. Annualised that is roughly 1.07% of the market capitalisation; with the dividend at 1.38% the total shareholder yield is approximately 2.45%. That is low against most of this batch and it is a real difference from the buyback-driven return profiles elsewhere.

Management's stated policy, quoted from today's release, is more ambitious: "Cummins plans to continue generating strong operating cash flow and returns for shareholders and is committed to our long-term strategic goal of returning 50% of operating cash flow back to shareholders." On trailing operating cash flow of $4,647 million that would be roughly $2.3 billion, or 2.65% of the market capitalisation — against the roughly $1.6 billion run-rate the first half implies. Half-weighted as management framing, and named as a gap between policy and practice.

4. Geography — a defect that removes the fastest-growing lines

The FY2025 seg_geo block does not tie to revenue, and the missing amount is the part of the business that is growing.

The vendor reports for FY2025: UNITED STATES: $18,974M and Non-US: $9,677M, summing to $28,651M against inc_a revenue of $33,670M. $5,019 million — 14.9% of revenue — is unaccounted for.

The 10-K settles it. Its disaggregation-of-revenue note gives the complete series:

Three findings.

First, the FY2024 vendor row is complete and ties exactly — China $2,948M + India $1,779M + Non-US $9,953M + US $19,422M = $34,102M. So the block is not structurally broken; the FY2025 row specifically dropped two lines.

Second, the surviving Non-US label is wrong. $9,677M is the 10-K's "Other international" line, not total non-US revenue. Total non-US revenue in FY2025 was $14,696 million, 43.6% of the total. A reader taking the vendor at face value would conclude Cummins derives 66.2% of revenue from the United States when the correct figure is 56.4%, and would understate international exposure by 5,019 million dollars.

Third, and this is why it matters more than a normal omission, the two dropped lines are the growth. China grew 11.9% in FY2025 to $3,298 million while the United States FELL 2.3% and "Other international" fell 2.8%. The single geography the vendor deleted is the only one that grew — and today's earnings release leads its revenue commentary with "international revenues, led by growth in China, increased 12%" and attributes the Engine segment's 23% international growth to "stronger construction demand in China." The vendor's FY2025 geography block deletes precisely the line that explains the quarter.

All geographic figures in this dive come from the 10-K.

seg_prod, by contrast, ties exactly. FY2025: Engine $10,875M + Components $10,149M + Distribution $12,405M + Power Systems $7,463M + Accelera $460M = $41,352M, less the eliminations line of $7,682M = $33,670M, matching revenue to the dollar. The eliminations line is mislabelled Total Segment, which is the opposite of what it contains, but the value is correct and the block does not over-count in the MRSH manner. The FY2022 and earlier rows correctly show the segment under its former name, "New Power", before the 2023 rename to Accelera.

5. Valuation — priced in or room?

At $635.94 (market cap $87.753B, 137.6M shares outstanding):

Estimate coverage is thin on earnings and better on revenue. 7 analysts on FY2026 EPS, 10 on FY2027, 7 on FY2028 — against 51 analysts issuing ratings. The FY2028 EPS range is $31.823 to $46.638, a 46.6% spread, which is too wide to anchor on and is used only as a cross-check. The FY2027 range of $32.711 to $37.433 is a 14.4% spread and is the anchor for all three fair values.

est.ebitdaAvg and est.ebitAvg are REJECTED with the fabrication signature stated exactly. In every year from FY2023 to FY2028, ebitdaAvg is exactly 19.812% of revenueAvg and ebitAvg is exactly 15.884%, to three decimal places. Cummins' actual consolidated EBITDA margin in the June 2026 quarter was 17.5% and in the first half 16.5%, and the company's own full-year guidance is 18.0-18.5%. The fabricated 19.812% sits above all of them and, more damningly, is identical in FY2023 — a year in which the emissions settlement crushed reported profitability. All forward valuation uses epsAvg.

Note the FY2026 consensus revenue of $37,240M is +10.6%, at the LOW end of the guidance range management raised this morning (+10% to +13%). That is a genuine, dated, falsifiable gap: if guidance is met at the midpoint, FY2026 revenue is approximately $37.5 billion and consensus is roughly 0.8% light; at the top end it is $38.0 billion and consensus is 2.1% light.

5a. What today's price assumes (the inversion)

At $635.94 — 21.7x FY2026 consensus and 18.6x FY2027 — the price embeds:

• Revenue grows 10.6% in FY2026, requiring a second half of $19.4 billion against $17.9 billion in the first half and $16.9 billion a year ago. (Consensus, cross-checked against the guidance raised this morning.) Implied second-half growth of roughly 15%, after a first half that grew 6.2%. That is the steepest single assumption in the price and it is management's own.
• Adjusted EPS reaches $34.20 in FY2027, 16.8% above FY2026. (Consensus; 10 analysts.) On revenue growth of 8.3%, that requires roughly 250 basis points of operating leverage — plausible if Power Systems keeps mixing up, and not if North American truck volume falls.
• The Model Year 2027 emissions transition does NOT produce a revenue decline. (Our derivation; consensus has FY2027 revenue at +8.3%.) This is the most fragile assumption in the price. The company disclosed this morning that it will run a "measured production ramp while maintaining select legacy product availability in 2027" in response to proposed EPA changes. Every prior North American emissions step-change has pulled purchases into the pre-transition year. A flat rather than growing FY2027 revenue line, with the mix unchanged, would put adjusted EPS nearer $30 than $34 and the stock at 18.6x that is $558.
• Power Systems holds a 24-27% EBITDA margin while capacity is added. (Our derivation; the June quarter was 24.5%, the first half 26.8%.) The 10-K guides FY2026 capital expenditure up to $1.35-1.45 billion with over 65% in North America. Capacity expansion into a demand surge is the normal way high-margin segments become average-margin segments, and it is not visible yet.
• The market keeps paying 18-22x forward earnings. (Our number.) Cummins has spent most of the last decade at 11-15x forward earnings as a cyclical engine manufacturer. The current multiple is the data-centre re-rating, and it is roughly 40% of the twelve-month return. At 14x FY2027E the stock is $479; at 22x it is $752.

5b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: adjusted EPS growth (+16.8%, FY2026E $29.277 to FY2027E $34.198) + multiple drift (COMPRESSION, from 21.7x on today's forward year to roughly 19.0x on the then-forward year, approximately −12%) + shareholder yield (+2.45%) ≈ +7%.

Our base assumes modest multiple compression and the reason is the segment mix, not scepticism about the demand. Power Systems is genuinely worth a high multiple; it is 19.3% of segment sales. Engine and Components, 49% of segment sales between them, are worth a cyclical multiple and both just contracted margin. A 19x blended forward multiple prices the mix roughly correctly today: it implies the market is paying something like 30x for the fifth of the company that is data-centre power and 14x for the rest. We think that is fair rather than cheap.

Note what that makes the return: essentially all earnings growth, with the multiple giving back most of it and a shareholder yield of only 2.45% providing little cushion. This is the opposite of the Capital One structure elsewhere in this programme, where half the return was yield and none of it required the crowd to change its mind. Here, most of the return requires the earnings to arrive on a steep second-half ramp AND the multiple to hold.

At 14x FY2027E the price is $479 (−24.7%). At 22x it is $752 (+18.3%). Our bull case of $855 is 22.8x the FY2027 consensus high — that is, it requires both the top of the estimate range and no de-rating.

5c. Variant perception (where we differ, what would surprise)

• We think the Model Year 2027 emissions pre-buy is the clearest unpriced item in this file, and the company put it in writing this morning. Consensus has FY2027 revenue growing 8.3%. The company has just said it will run a "measured production ramp" of the new X10 and X15 "while maintaining select legacy product availability in 2027" — language that describes managing a demand pull-forward, not a growth year. Watchable event: North American on-highway order and backlog commentary in the November 2026 and February 2027 releases, and the Engine segment's North American revenue line, which grew 1% this quarter. A FY2027 revenue guide below +3% would take roughly $4 off consensus EPS and, at an unchanged multiple, roughly $75 off the share price.
• We agree with the knowledge base on the direction and disagree on the timing. jordi_visser's September 2025 claim that "Cummins is an AI company" has been comprehensively validated — Power Systems margin up 170 basis points on 19% growth, a five-year Texas data-centre supply agreement, Distribution's North American revenue up 13% on the same driver. Our disagreement is that the market has now agreed too: the stock is up 78.7% in twelve months against SPY's 24.3%, and the multiple has moved from cyclical to structural. The claim was an edge in September 2025. It is consensus in August 2026. That distinction is the difference between our $650 and the street's $779.70.
• We think the miss matters more than the recovery suggests, because of its position in a sequence. Beats of +22.9%, +15.7%, +13.9% and +9.2% and then a −6.7% miss is a decaying surprise series terminating in a negative. The June-quarter miss is a clean one: the $6.73 appears identically in the vendor's inc_q and the earnings calendar, so it is not a GAAP-versus-adjusted artefact, and even adding back the $0.21 of unfavourable discrete tax gives $6.94 against $7.21. Watchable number: the November print against consensus, and specifically whether consolidated EBITDA margin reaches the 18.0-18.5% full-year guide, which requires a substantial step from the first half's 16.5%.
• Positive surprise that would force a re-rating upward: a second five-year data-centre supply agreement of Circe scale, or the disclosure of a Power Systems backlog figure — Cummins does not currently publish one, and a multi-year backlog would convert the segment from a demand story into a contracted annuity; or Power Systems passing 25% of total segment sales; or the second half delivering the top of the raised guidance at +13% revenue growth.
• Negative surprise that would break the thesis: FY2027 revenue guided flat or down on the emissions transition; Power Systems EBITDA margin below 22% as capacity is added; the incentive-compensation drag failing to reverse, leaving full-year EBITDA below the 18.0% guidance floor; a Chinese demand reversal, with China at $3,298 million of FY2025 revenue and rising; or any tariff or trade action against the Chinese and Indian joint ventures that produced $252 million of first-half equity income.

Synthos fair values

All three anchors are multiples of the FY2027 consensus EPS distribution (mean $34.198, low $32.711, high $37.433; 10 analysts), cross-checked against the recomputed free-cash-flow yield.

• Bear ~$440 — 13.5x the FY2027 consensus LOW of $32.711. Cross-check: 15.0x FY2026E; a 7.7% free-cash-flow yield; 21.7% above the 52-week low of $361.59. The scenario: the Model Year 2027 pre-buy reverses, North American truck volume falls, Power Systems margin normalises toward 20% as capacity arrives, and the multiple returns to the 13-15x band Cummins occupied for most of the last decade. −30.8%.
• Base ~$650 — 19.0x the FY2027 consensus MEAN of $34.198. Cross-check: 22.2x FY2026E; 16.0x FY2028E; a 5.2% free-cash-flow yield. Sensitivity, stated openly: 17x gives $581 (−8.6%) and 22x gives $752 (+18.3%). The entire answer lives in a 17-22x band on one estimate row supported by ten analysts. The scenario: the raised guidance is met, Power Systems keeps compounding at high margin, the 2027 transition is managed without a decline, and the multiple settles modestly below today's. +2.2%.
• Bull ~$855 — 22.8x the FY2027 consensus HIGH of $37.433. Cross-check: 21.1x FY2028E; 17.5% above the 52-week high of $727.59. The scenario: data-centre generator demand proves to be a multi-year contracted annuity rather than a cycle, Power Systems passes a quarter of the company, the 2030 targets raised at the May Analyst Day prove conservative, and the market values Cummins as a power-infrastructure business. +34.4%.

Base is 2.2% above spot; asymmetry roughly 1.11:1 (30.8% down, 34.4% up), before a shareholder yield of approximately 2.45%. A payoff ratio barely above one, on a name that has risen 78.7% in twelve months and missed this morning, does not clear the bar for a Buy. It is a clean Watch with a pre-registered level, set out in Section 10.

6. Knowledge base — two claims, one voice, and the filing agrees with both

Raw hits: 2. After a case-sensitive entity re-run: 2. Entity matches: 2. Name-level claims on Cummins: 2, both BULLISH. Discarded: 0.

The entity sweep ran CMI and Cummins across all 52,021 distilled claims and returned two hits. A free-text sweep on "cummins" returned the identical two. There is no homograph contamination on this name at all — a notable result for a three-letter ticker in a store where CEG matched Constellation Software on 26 of 29, PWR matched Quanta Computer at conviction 85, and EMR's only entity match was an unrelated LLC. CMI is clean, and that is worth recording because it means the two claims below can be taken at face value.

Both claims, verbatim:

> 2025-09-14 · BULLISH · conviction 68 · skill 2.0 · horizon: thesis · entities: CAT, CMI, PCAR · categories: datacenter power, AI infrastructure, electrical equipment · channel: jordi_visser · no named speaker · role: independent

> "The AI power buildout is a macro theme spanning transformers, switchgear, power distribution — Caterpillar, Cummins, eventually Paccar and electrical-component names benefit as demand shows up in earnings."

> 2025-09-21 · BULLISH · conviction 70 · skill 2.0 · horizon: thesis · entities: CMI · categories: datacenter power, generators · channel: jordi_visser · no named speaker · role: independent

> "Cummins is an AI company — surging generator and power-systems demand is making it money even amid a PMI recession."

What this lane is, and it is unusually good.

Both claims carry skill 2.0 — double weight, the highest multiplier in the store. Both are eleven months old. Both are directionally correct and specifically correct. The second claim's mechanism — "surging generator and power-systems demand" — is reproduced in today's filing almost word for word: Power Systems sales up 19%, EBITDA margin up 170 basis points to 24.5%, driven by "increased power generation demand, particularly for data center markets in the United States, China and Asia Pacific", plus Distribution's North American revenue up 13% "driven by increased demand for power generation products, particularly for data center applications", plus the June 2026 Circe Energy agreement for generator sets to a Texas high-performance-computing data centre with deliveries scheduled 2026 through 2030.

The second claim's qualifier — "even amid a PMI recession" — is also visible in the data and it is the more useful half. The June quarter shows exactly that split: Power Systems +19% at expanding margin, while Engine's North American revenue grew 1% and its margin fell 130 basis points, and Components' margin fell 150 basis points. A single company containing both a boom and a stall is precisely what the claim described.

Over the twelve months since the claims were made, CMI has returned +78.72% against SPY's +24.26% and QQQ's +30.80%. We record that the call was right, and we record how it was made — a thematic macro claim about AI power buildout, mapped to names, before the earnings showed it.

The concentration test is the whole caveat and it is severe. Both claims come from a single channel. Removing jordi_visser empties the lane completely. Two claims from one voice is breadth 1, not breadth 2, and the conviction rating is set on that basis: positive-moderate, not positive-high. No independent voice in 52,021 claims corroborates it. Both carry speaker: null and are channel-attributed — the weakest sourcing the four-lane policy admits — with speaker_role: independent, which under the standing rule that speaker_role is unreliable is noted rather than relied upon. Neither carries a management voice.

One further honest observation. The claims are bullish on a thesis that has now substantially played out in the price. A knowledge-base edge is the gap between what credible thinkers argue and what the market has priced. In September 2025 that gap was large. After a 78.7% advance and a re-rating from roughly 13x to 18.6x forward earnings, it is small. The claim being right is not the same as the stock being cheap, and this dive separates the two.

Conclusion. Breadth 1, claim count 2, net conviction positive-moderate. The knowledge base is bullish on Cummins for the right reason, from one voice, eleven months early, and the filing has now confirmed it. The verdict is Watch despite that, because the price has confirmed it too.

7. Data integrity — what we rejected, and what came back clean

Six findings. Cummins' file is mid-range: two fields are wrong, one estimate row is fabricated, and three checks pass cleanly including the capex check.

1. capitalExpenditure — CHECKED AND VERIFIED CLEAN. FY2025 −$1,235M, FY2024 −$1,208M and FY2023 −$1,213M all match the 10-K consolidated cash-flow statement to the dollar, and the MD&A's "Capital expenditures were $1.2 billion each year in 2025, 2024 and 2023" corroborates. The recomputed trailing free cash flow of $3,367M reproduces freeCashFlowYieldTTM (3.8369%), capexToOperatingCashFlowTTM (0.2754) and capexToDepreciationTTM (1.1429) exactly. Both the diagnostic and the datum were tested separately against the filing and both pass. No repair applied.

2. seg_geo FY2025 drops the China and India lines and mislabels the residual — REJECTED, filing substituted. The vendor reports United States $18,974M and Non-US $9,677M, summing to $28,651M against revenue of $33,670M — a $5,019M, 14.9% shortfall. The 10-K's disaggregation note gives United States $18,974M, China $3,298M, India $1,721M and Other international $9,677M. The vendor has kept the "Other international" line and relabelled it "Non-US". The FY2024 row is complete and ties exactly, so this is a FY2025-specific deletion, not a structural failure. The aggravating detail: the two deleted lines are the growth — China grew 11.9% while the United States fell 2.3% — and today's earnings release leads with "international revenues, led by growth in China, increased 12%." True non-US revenue is $14,696M, 43.6% of the total, not the 28.7% the block implies. All geographic figures come from the 10-K.

3. totalDebt overstates borrowings by 7.4% through lease inclusion — REJECTED, filing substituted. Vendor $8,114M at 2025-12-31 against the 10-K working-capital table's $7,552M. The $562M gap is capitalLeaseObligationsCurrent ($138M) plus capitalLeaseObligationsNonCurrent ($424M). Milder than Sherwin-Williams' 19.1% in the same batch but the same defect class. All leverage figures use the filing.

4. enterpriseValueTTM carries TWO errors running in opposite directions — both reported, neither used to excuse the other. The vendor's $92,269M implies net debt of exactly $4,516M, which is $7,695M of June-2026 debt less cash and cash equivalents only. (a) The $745M of marketable securities is omitted from the cash deduction — the known short-term-investments omission, and the payload's own bal_a carries the field. (b) The $1,059M of noncontrolling interests is omitted from enterprise value — the confirmed NCI class, at 1.15% of EV here against KKR's 35.6%. Rebuilt EV is $92,583M, 0.34% above the vendor's. The residual is far inside the 15% withholding threshold, so EV multiples are usable, but the corrected net debt of $3,771M (against the vendor's implied $4,516M) changes netDebtToEBITDATTM from 0.909x to approximately 0.76x, and the corrected figure is used.

5. est.ebitdaAvg and est.ebitAvg carry a fixed-ratio fabrication signature — REJECTED. In every year from FY2023 to FY2028, ebitdaAvg is exactly 19.812% of revenueAvg and ebitAvg is exactly 15.884%. Cummins' actual June-quarter consolidated EBITDA margin was 17.5%, the first half 16.5%, and full-year guidance is 18.0-18.5% — the fabricated ratio exceeds all of them, and is identical in FY2023, a year distorted by the emissions settlement. All forward valuation uses epsAvg.

6. inc_q EPS and earn_cal EPS use different bases in three of the last four quarters — noted, and the current quarter is confirmed NOT affected. For Q3 2025, Q4 2025 and Q1 2026 the vendor's GAAP diluted EPS ($3.85, $4.27, $4.71) differs from the earnings-calendar actual ($5.59, $5.81, $6.15) by $1.44 to $1.74 — the calendar is on the adjusted basis consensus uses. For the June 2026 quarter both read $6.73, which the earnings release confirms is GAAP diluted EPS attributable to Cummins. This matters because it means the 6.7% miss against $7.21 is a genuine miss and not a basis mismatch, and this dive states that explicitly rather than leaving the reader to reconcile two EPS series.

Also correct and worth recording: seg_prod ties exactly — FY2025 segments summing to $41,352M less an eliminations line of $7,682M equals $33,670M, matching revenue to the dollar, with no over-counting of the MRSH class (the eliminations row is mislabelled Total Segment, which is the opposite of its content, but the value is right); operating cash flow of $3,621M (FY2025), $1,487M (FY2024) and $3,966M (FY2023) matching the 10-K exactly; acquisitionsNet of −$122M for FY2025 against the 10-K's acquisition line of −$12M plus other investing items — close enough to note but not to rely on, and not used; dividendPerShareTTM of $8.00 as a correct trailing figure now superseded by today's $2.20 quarterly declaration; and the share count, with 137.6M outstanding per the release against 137.99M implied by market capitalisation ÷ price, a 0.28% difference attributable to the timing of the count.

Non-equity tripwire — checked and passed. CMI is common stock, $2.50 par value, NYSE-listed, per the balance sheet in the earnings release (500 million shares authorised, 222.5 million issued, 84.9 million in treasury). Price of $635.94 is not par-like; beta is 1.246; the dividend is a declared and repeatedly raised quarterly rate, not a coupon; volume was 2.23M shares (~$1.4 billion of turnover); the 52-week band of $361.59 to $727.59 is a 101% range. No preferred stock is outstanding. Note the separate $1,059 million of noncontrolling interests, which are NOT this security and which are addressed in finding 4. This is common equity.

Vendor composite rating — noted, and not used. B / 3 overall, with 5 on return on equity, 4 on return on assets, and 2, 2 and 1 on debt-to-equity, price-to-earnings and price-to-book. The debt-to-equity sub-score is generated partly by the overstated debt of finding 3. The composite adds nothing this dive has not derived directly.

8. Technicals

• Price $635.94. −12.60% from the 52-week high of $727.59; +75.87% above the 52-week low of $361.59. Position within the annual range: 75th percentile.
• The moving averages have crossed against the price for the first time in this move: 4.8% BELOW a 50-day average of $668.26; 9.3% ABOVE a 200-day of $581.77. The 50-day sits far above the 200-day, which is the shape of a strong uptrend beginning to consolidate.
• RSI 41.0 — the weakest reading in this batch and the only one below 50. MACD −10.42, also the batch's most negative. Both are consistent with a stock that has stopped going up, not one that is breaking down.
• Maximum drawdown from peak over the trailing year: −12.60%, identical to the distance from the high, meaning the current level IS the low of the drawdown.
• Relative performance, and the twelve-month figure is the story: 3-month −3.17% against SPY +7.59% and QQQ +7.67% — eleven points behind; 6-month +6.86% against SPY +11.09%; 12-month +78.72% against SPY +24.26% and QQQ +30.80% — a 54-point and 48-point outperformance. This is the largest twelve-month relative gain in this batch by a wide margin, and the last three months have given a little of it back. Both facts belong in the same sentence.

Today's move, in detail, because it is unusual

CMI closed 2026-08-04 at $635.94, down 1.99% or $12.91 from $648.85. But the path matters more than the endpoint:

From the low to the close is a 10.08% recovery. Volume was 2,226,922 shares against a 1,035,613 average — 2.15 times normal. The intraday range of $577.718 to $640.745 is 10.9% of the closing price.

The honest read: this was a full, high-volume, two-sided argument about the same release, and it ended close to neutral. The bears had the miss — $6.73 against $7.21, EBITDA margin down 90 basis points, three of five segments contracting. The bulls had the raised guidance, the raised dividend, and Power Systems at +19% with expanding margin. Neither side won. We do not think a research note written the same afternoon should claim to have settled what the market spent a full session and twice its normal volume failing to settle, and the Watch verdict says so.

What this does to the entry: it improves it, but not enough. The shares are now 12.6% below the 52-week high and below the 50-day average for the first time in this advance, with RSI at 41. That is a materially better entry than the one available a month ago. It is still 18.6x FY2027 consensus with a 1.11:1 payoff and a 2027 regulatory hazard the company disclosed this morning.

9. Insiders — eight transactions, all small, none informative

Total open-market sales: 1,372.8 shares, approximately $967,000, across two officers. Total gifts: 973 shares. One director award of 306 shares.

The reading, and it is thin. Three of the eight transactions are G-Gift — charitable or family transfers at a zero recorded price, which carry no directional information. One is a director's routine A-Award. Four are open-market sales, all in the week of 11-14 May 2026, at prices between $702.66 and $712.70 — roughly 11% above today's close, and near what was then the high.

Two qualifications, both reducing the signal. The amounts are trivially small: Donald Jackson, VP Treasury & Tax, sold 730 shares in three tranches on a single day at rising prices — a single order executing — and retains 8,315.6. Bonnie Fetch's sale of 642.8 shares leaving a recorded balance of zero looks alarming in isolation and is not: she separately shows 11,524 shares held after a gift two days later, so the zero applies to one specific holding, not to her total position. The entire insider file amounts to under one million dollars of sales across two officers in one week.

What the file does not contain: any transaction by chief executive Jennifer Rumsey or the chief financial officer, and no purchases by anyone. On a stock that rose 78.7% in twelve months, the absence of large executive selling is mildly notable; the absence of any purchase is normal. The file is close to uninformative and is reported as such.

10. Verdict, kill-criteria and flip conditions

Watch.

The thesis is right and it is not ours. In September 2025 a single voice in the Synthos knowledge base — jordi_visser, skill 2.0 — argued that "Cummins is an AI company" because "surging generator and power-systems demand is making it money even amid a PMI recession." Eleven months and a 78.72% return later, today's filing states it in the company's own words: Power Systems sales +19% to $2,255 million with EBITDA margin expanding from 22.8% to 24.5%, driven by "increased power generation demand, particularly for data center markets"; Distribution's North American revenue +13% on the same driver; a June 2026 agreement to supply generator sets to a Texas high-performance-computing data centre with deliveries scheduled from 2026 through 2030. Meanwhile Engine's North American revenue grew 1% and three of five segments contracted margin. A boom and a stall inside one company, exactly as described.

We are not buying it at $635.94, and the reasons are four numbers.

First, the payoff ratio is 1.11:1. Base $650 (+2.2%), bear $440 (−30.8%), bull $855 (+34.4%). A bar of 2:1 exists precisely so that a correct thesis at the wrong price does not become a position.

Second, the multiple has done roughly 40% of the work already. Cummins spent most of the last decade at 11-15x forward earnings. It is now 18.6x FY2027. The re-rating from cyclical to structural has happened. What remains is the earnings.

Third, the company disclosed a dated hazard this morning that consensus does not appear to model. The Model Year 2027 North American emissions transition, a "measured production ramp while maintaining select legacy product availability in 2027", against a consensus FY2027 revenue line growing 8.3%. Every prior emissions step-change pulled demand into the preceding year.

Fourth, the shareholder yield is only about 2.45% — a 1.38% dividend and roughly 1.07% of buyback — so there is very little cushion under a return that depends on both the earnings and the multiple.

And one counterweight we will not minimise. Guidance was RAISED this morning, not cut, to +10-13% revenue growth and 18.0-18.5% EBITDA. The dividend was raised for the seventeenth consecutive year. The balance sheet carries net debt under one turn of EBITDA. If the second half delivers the 15% growth the guidance implies and Power Systems holds 24%-plus margins, the street's $779.70 will look conservative and this Watch will look timid. We accept that risk explicitly, because the alternative — buying a 1.11:1 payoff on the afternoon of a 6.7% miss — is worse.

Pre-registered BUY trigger — the specific level at which this becomes Buy — Tactical:

• A price at or below approximately $545 with the FY2027 consensus of $34.20 intact. That is 16x FY2027E, restores roughly 2:1 asymmetry against the $440 bear and the $855 bull, and is 6.3% below the 200-day moving average of $581.77. This is a standing, gradeable level.
• Or, at any price, two consecutive quarters of Power Systems growth above 15% with EBITDA margin at or above 24%, AND a FY2027 revenue guide above +5% — which would resolve the pre-buy question in the company's favour and justify paying up.

Pre-registered KILL criteria — what would take this to Avoid rather than merely leaving it unowned:

• FY2027 revenue guided flat or down, confirming the emissions pre-buy reversal.
• Power Systems EBITDA margin below 22% in any quarter, indicating the capacity build is diluting the one segment carrying the multiple.
• Full-year 2026 EBITDA below the 18.0% guidance floor, which would mean the incentive-compensation explanation for the June quarter's 90-basis-point decline was not the whole story. The first half was 16.5%.
• Engine segment North American revenue declining year on year. It grew 1% this quarter.
• A second consecutive earnings miss, extending a surprise series that has decayed from +22.9% to −6.7% in four quarters.
• A material China disruption — tariff, trade action or demand — against $3,298 million of FY2025 China revenue and $252 million of first-half equity income from Chinese and Indian joint ventures.

Where CMI fits in the Synthos Framework Portfolio. No position today. Tracked on the industrials/AI-infrastructure watch list at the $545 buy trigger above, with a 2.5% initial size and a 4% target if triggered. Sizing note: this is the rare name where the knowledge base, the filing and the price all agree on the direction and only the price argues against acting — which is exactly the situation a pre-registered trigger exists for, and exactly the situation in which improvising an entry is most tempting and least disciplined. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $635.94, with the fair-value anchors, the buy trigger, the kill criteria and the 6a assumptions all gradeable.

Single biggest risk: the Model Year 2027 emissions pre-buy, disclosed by the company on the morning of this dive. Cummins has changed its North American on-highway launch plan for the 2027 model year in response to proposed EPA changes, opting for a "measured production ramp while maintaining select legacy product availability in 2027." North American on-highway is the historic profit cycle of this company; Engine external sales of heavy-duty truck engines already fell 23% from $3,320 million in FY2024 to $2,544 million in FY2025. Consensus has FY2027 revenue growing 8.3% and FY2028 growing 9.5%, with no visible allowance for a pull-forward reversal. If FY2027 revenue is flat rather than up 8.3%, adjusted EPS lands nearer $30 than $34.20, and at an unchanged 18.6x multiple the shares are approximately $558 — 12% below today's price with nothing else going wrong. Power Systems can be everything the knowledge base says it is and this can still happen, because Power Systems is 19.3% of the company and North American trucks are not.

Provenance & disclosures

• Traceability: 2 name-level knowledge-base claims name Cummins out of 52,021 distilled claims, both BULLISH, and both from the same channel (raw hits 2, surviving a case-sensitive entity re-run 2, name-level 2, discarded 0; breadth 1 — two claims from one voice is breadth 1 — claim count 2, net conviction positive-moderate). Both are jordi_visser, skill 2.0 (double weight, the highest in the store), speaker: null, speaker_role: independent: 2025-09-14, conviction 68, entities CAT/CMI/PCAR — "The AI power buildout is a macro theme spanning transformers, switchgear, power distribution — Caterpillar, Cummins, eventually Paccar and electrical-component names benefit as demand shows up in earnings."; and 2025-09-21, conviction 70, entity CMI — "Cummins is an AI company — surging generator and power-systems demand is making it money even amid a PMI recession." Both have been validated in detail by the 2026-08-04 filing (Power Systems sales +19%, EBITDA margin 22.8% to 24.5%, attributed to data-centre power generation; Distribution North America +13% on the same driver; a June 2026 Circe Energy agreement with deliveries scheduled 2026-2030), and the stock has returned +78.72% against SPY's +24.26% in the eleven months since. The concentration test is decisive and is stated as a limitation rather than buried: removing jordi_visser empties the lane entirely, and no independent voice in 52,021 claims corroborates the thesis. A free-text sweep on "cummins" returned the identical two hits — there is NO homograph contamination on this three-letter ticker, which is recorded because CEG, PWR, EMR, USB and MMM all failed the same test. Under the standing rule that speaker_role is unreliable, the "independent" tag is noted rather than relied upon. Quotes are verbatim from the stored claim text.
• Data as-of: fundamentals — income statement, balance sheet, cash flow and segment results through 2026-06-30, all filing-verified from the 8-K earnings release and 10-Q both filed 2026-08-04, the day of this dive; annual figures for FY2025, FY2024 and FY2023, the geographic disaggregation, the total-debt figure and the capital-expenditure series from the 10-K filed 2026-02-10 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873598 = 2026-08-04T19:59:58Z ($635.94, −1.99%; open $596.50; low $577.718; 50-DMA $668.26; 200-DMA $581.77; RSI 41.0; MACD −10.42) · knowledge-base claims 2026-08-04. Cummins' fiscal year is the calendar year. All figures come from the Synthos vendor data file for CMI or from the SEC filings in the CMI archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-10 (fiscal 2025 — the source of total debt of $7,552M, the capital-expenditure series, the FY2026 capex guidance of $1.35-1.45bn, the complete geographic disaggregation and the dividend history); 10-Q filed 2026-05-05 (March 2026 quarter); 10-Q filed 2026-08-04 (June 2026 quarter, filed the day of this dive); 8-K filed 2026-05-05 (first-quarter earnings release); 8-K filed 2026-05-14 (Item 5.07 — 2026 annual meeting results: eleven directors elected, say-on-pay, PwC ratified, a 2026 Omnibus Incentive Plan approved, and shareholder proposals on Chair/CEO separation and charitable-support reporting; no corporate action); 8-K filed 2026-08-04, which carries the complete second-quarter earnings release with the segment EBITDA table, the raised guidance, the dividend increase, the Circe Energy agreement, the Model Year 2027 launch-plan change, the fuel-cell disposal footnote and the tax-rate detail. All carry preserved [TABLE] statement data (tables: true).
• Where the filings contradicted or corrected the vendor (detailed in Section 7): seg_geo FY2025, which reports only United States $18,974M and a mislabelled Non-US $9,677M, summing to $28,651M against revenue of $33,670M — a 14.9% shortfall caused by the deletion of the China ($3,298M) and India ($1,721M) lines, with the 10-K's "Other international" residual relabelled as total non-US; the FY2024 row is complete and ties, so this is a FY2025-specific deletion, and the deleted China line is the only geography that GREW; totalDebt, at $8,114M against the 10-K's stated $7,552M, a 7.4% overstatement from $562M of lease obligations; enterpriseValueTTM, which omits $745M of marketable securities from the cash deduction AND omits $1,059M of noncontrolling interests from enterprise value — two errors in opposite directions netting to a 0.34% understatement, with the corrected net debt of $3,771M changing netDebtToEBITDATTM from 0.909x to approximately 0.76x; and est.ebitdaAvg/ebitAvg, rejected for a fixed-ratio fabrication signature — exactly 19.812% and 15.884% of revenueAvg in every year from FY2023 to FY2028 — against an actual June-quarter EBITDA margin of 17.5% and full-year guidance of 18.0-18.5%. Where vendor and filing AGREED — recorded, because a clean verification is a real finding: the capital-expenditure check PASSED (FY2025 −$1,235M, FY2024 −$1,208M and FY2023 −$1,213M all matching the 10-K to the dollar, with the recomputed trailing free cash flow of $3,367M reproducing freeCashFlowYieldTTM of 3.8369%, capexToOperatingCashFlowTTM of 0.2754 and capexToDepreciationTTM of 1.1429 exactly); operating cash flow ($3,621M / $1,487M / $3,966M, matching the 10-K); seg_prod (FY2025 segments summing to $41,352M less eliminations of $7,682M equals $33,670M, exactly inc_a revenue — no over-counting, though the eliminations row is mislabelled Total Segment); and the share count (137.6M outstanding per the release against 137.99M implied by market capitalisation ÷ price, a 0.28% difference).
• Basis note: consensus epsAvg is on an adjusted basis that excludes items such as the FY2023 emissions settlement and the FY2024 Atmus divestiture gain, and it diverges from GAAP diluted EPS by up to 165% (FY2023) and 28% (FY2024). Forward multiples in this dive use epsAvg; the trailing multiple of 32.5x is computed on GAAP diluted EPS of $19.56 and is shown separately. Critically, the June 2026 quarter is NOT affected by this divergence: $6.73 appears identically in the vendor's inc_q, the earnings calendar and the earnings release, so the 6.7% miss against the $7.21 estimate is a genuine miss rather than a basis mismatch, and even adding back the $0.21 of unfavourable discrete tax gives $6.94. In the three preceding quarters the two series differ by $1.44 to $1.74 and the dive shows both.
• Estimate coverage: 7 analysts on FY2026 EPS, 10 on FY2027 — the anchor for all three fair values, with an FY2027 range of $32.711 to $37.433 (14.4%) — and 7 on FY2028, whose $31.823-$46.638 range is a 46.6% spread and is used only as a cross-check. Revenue coverage is broader at 13, 15 and 12. Against 51 analysts issuing ratings, EPS coverage of 7-10 is thin and is stated as such.
• Peer note: the vendor peer set — AMETEK, Canadian National, Canadian Pacific Kansas City, CSX, FedEx, Illinois Tool Works, Norfolk Southern, PACCAR, Roper Technologies and United Rentals — contains four railroads and a parcel carrier, which are freight demand proxies rather than manufacturing comparables. PACCAR is the only genuine on-highway peer and Caterpillar, the closest analogue for the power-generation business and a named entity in one of the two knowledge-base claims, is absent. No peer-multiple comparison is drawn.
• Fair-value caveat: the $440 / $650 / $855 anchors are multiples of the FY2027 consensus EPS distribution — 13.5x the low of $32.711, 19.0x the mean of $34.198, and 22.8x the high of $37.433 — each cross-checked against the recomputed free-cash-flow yield (7.7%, 5.2% and 3.9% respectively). Stated arithmetic, not a discounted cash flow. The base is sensitivity-disclosed: 17x gives $581, 22x gives $752. The whole answer lives in a 17-22x band on one estimate row supported by ten analysts, and no greater precision than that is available from this file.
• Timing — and this dive is unusual in this respect: second-quarter 2026 results were released on the MORNING of 2026-08-04, the day of this dive, with the 8-K and 10-Q both filed the same day. Diluted EPS of $6.73 MISSED the $7.21 consensus by 6.7%; revenue of $9,457M beat by 1.4%; full-year guidance was RAISED to +10-13% revenue and 18.0-18.5% EBITDA; and the quarterly dividend was RAISED from $2.00 to $2.20. The shares opened at $596.50 (−8.1%), traded to $577.718 (−11.0%) and closed at $635.94 (−1.99%) on 2.15x average volume. A dive struck on the day of a violent two-sided repricing carries more uncertainty than one struck in a quiet week, the Watch verdict reflects that explicitly, and a reader should weight the pre-registered $545 trigger above more heavily than the same-day fair-value estimate. The next print is the third quarter, expected early November 2026 on the company's recent cadence, approximately 94 days away. The most recent insider filing is dated 2026-05-19.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.