IBM › analyze
IBM · Analyze
Date: 2026-08-12 · Price: $235.98 · Sector: Technology (Enterprise Software / IT Services) · Beta: 0.70 Market cap $225.0B · Net debt $57.1B · EV $282.1B · Yield 2.86% · 52wk $199.19–$332.46
Verdict: 🟡 HOLD / WATCH — the business question resolved better than feared; the per-share arithmetic resolved worse. Conviction [5.5], down from [6.0].
What changed since the 2026-07-14 file
That report was written mid-crash at $219, and explicitly deferred to the Q2 print as "the referee." The print came on 2026-07-22. The referee returned a split decision, and the stock is 7.7% higher.
The moat verdict is better than the July file thought. Infrastructure −7.4% is a z17 cycle rollover on schedule, not a structural decline. IBM Consulting — which July named "the soft flank" — is the segment behaving best, expanding margins on flat revenue. The sector-wide "AI infrastructure is eating software budgets" fear that cost IBM 25% in a day has been refuted outright by ServiceNow's print.
The ownership verdict is worse. Four things the July file had wrong or did not know: the dividend grows at 0.6%/yr, not ~3% (DDM falls $139 → $92); ROIC is ~10.5%, not 12–13%, and falling; all of Q2's software growth was acquired; and IBM closed an $11.6B acquisition of Confluent on 2026-03-17 that re-levered the balance sheet a full turn and drew cash down 47%.
Net: a better-understood business, worse per-share economics, at a higher price. That is a downgrade, and it is driven by the arithmetic rather than by the moat.
0. The three findings that matter
1. All of IBM's Q2 software growth was purchased. Reported Software +5.1%. Organic: flat. Data +18.9% — the 10-Q attributes it to "recent acquisitions, primarily Confluent." Software ARR $24.6B "+8%" — the ~$1.8B increase is roughly $1.2–1.3B of acquired Confluent ARR, leaving organic ARR growth near +2.5%. Two agents reached this independently, one by derivation from the 10-Q and one from IBM's own organic disclosure. IBM paid $11.6B (of which $7.24B goodwill) for the growth line the bull case cites.
In fairness: Q2 organic software is depressed by the $400–500M of deferred ELAs. Add those back and organic software is ~+6% — respectable, still short of the "double-digit software" ambition the re-rating was built on.
2. The dividend growth is functionally dead, and this breaks the dividend thesis. Restoring Yahoo's Kyndryl-spin adjustment (see §7 pitfall), the declared quarterly dividend has risen by exactly one penny per year for six consecutive years:
| Year | 2018 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | 2026 |
|---|---|---|---|---|---|---|---|---|
| Quarterly | $1.57 | $1.63 | $1.64 | $1.65 | $1.66 | $1.67 | $1.68 | $1.69 |
0.60%/yr for five years. 0.92%/yr over eight. 1.90%/yr over ten. Before 2018 it compounded at 9.3%. The 30-year Aristocrat streak is technically alive and economically over — this payout has lost real purchasing power every year since 2018, at roughly 2pts/yr.
3. The absence of a buyback after a 25% crash is a capacity problem, not just a conviction problem. Sentiment flags that no insider bought the dip and no repurchase was authorized. Fundamentals explains why they couldn't: IBM spent $10.5B net cash on Confluent in Q1, drew cash from $13.6B to $7.2B (−47%), took net debt/EBITDA from 2.87x to 3.25x, and ran 126% of operating cash flow in FY2025. IBM did not decline to support its stock; it had already spent the firepower in March. This also means IBM cannot support the stock if it falls further — a real asymmetry for anyone underwriting a floor.
1. Fundamentals (Health)
1.1 The clean series — post-Kyndryl only
⚠️ A 5–8yr FCF CAGR, as the framework requests, is not computable. Yahoo caps at four annual periods; roic.ai's plan caps at two (pitfall-roicai-free-plan-caps-history-at-two-years — confirmed again today); and the Nov-2021 Kyndryl spin (~$19B revenue removed) invalidates anything earlier regardless (pitfall-divested-segment-corrupts-multiyear-cagr). The longest defensible window is three years. Not fabricated.
| $B | FY2022 | FY2023 | FY2024 | FY2025 | TTM |
|---|---|---|---|---|---|
| Revenue | 60.53 | 61.86 | 62.75 | 67.54 | 69.10 |
| Operating income | 8.17 | 9.38 | 10.07 | 12.49 | 12.71 |
| Operating margin | 13.50% | 15.16% | 16.05% | 18.50% | 18.39% |
| Net income (reported) | 1.64 | 7.50 | 6.02 | 10.59 | 10.73 |
| FCF (GAAP) | 8.46 | 12.12 | 11.76 | 11.46 | 13.14 |
| FCF (IBM definition) | n/a | n/a | 12.75 | 14.70 | n/a |
| Stock comp | 0.99 | 1.13 | 1.31 | 1.72 | 1.88 |
| Metric | FY22→FY25 CAGR | Read |
|---|---|---|
| Revenue | +3.72%/yr | Clean |
| Operating income | +15.20%/yr | The honest trend line — and the strongest fact in the file |
| Operating margin | +500bp | Real |
| Net income / EPS (reported) | +86% / +84%/yr | ⚠️ ARTIFACT — do not use |
| FCF (GAAP) | +10.6%/yr | ⚠️ Inflated by a depressed FY22 base; the FY23→25 trend is −2.8%/yr |
| Revenue/share | +2.33%/yr | 139bp/yr below revenue — the dilution tax |
| Shares outstanding | +1.12%/yr | Relentless, monotonic, every quarter |
Why reported EPS cannot be used. Three distortions stack: FY2022 was crushed by a ~$5.9B pension settlement; FY2024 carried $4.28B of non-operating charges; and FY2024 and FY2025 both booked negative effective tax rates (−3.8%, −2.3%). Normalising FY2025 to a 21% statutory rate gives EPS of ~$8.60 against the reported $11.17 — roughly 23% of FY2025 EPS is a tax benefit, not operations. Use operating income.
1.2 The two free-cash-flow definitions, and why the gap decides the year
IBM's own FCF measure excludes financing-receivable movements. GAAP OCF−capex does not. The gap is large and it flipped sign in 2026:
| FY2025 | H1 2026 | |
|---|---|---|
| GAAP FCF (OCF − capex) | $11.46B | $6.99B (+31.9% YoY) |
| FCF (IBM definition) | $14.70B | $4.8B (FLAT YoY) |
| Wedge | +$3.2B | −$2.2B |
The honest read of H1 2026 cash generation is flat. The +32% GAAP figure is a financing-receivables release, not an operating improvement — IBM strips exactly this out and its own measure shows 0% growth. Do not quote the GAAP H1 number as a business improvement.
1.3 The FY2026 guidance arithmetic — the single most testable claim in this file
Post-Q2, IBM cut FY26 revenue to 4–5% cc, raised the operating pre-tax margin guide to +100bp, and held FCF at "+about $1B."
| IBM-defined FCF | H1 | H2 required | Full year |
|---|---|---|---|
| FY2025 actual | $4.8B | $9.9B | $14.70B |
| FY2026 | $4.8B (flat) | $10.9B | ~$15.7B guided |
H1 delivered 0% growth. The full-year guide requires H2 to grow +10.1% — in the same half Infrastructure is running −7% and organic software is flat. Management's defence is specific and checkable: a deliberate +$600M inventory buy-ahead to secure constrained supply, which Kavanaugh calls "a tailwind in the second half," with attainment ~30% and "very similar position to the z16 mainframe cycle." If H2 demand does not arrive, that inventory is a writedown rather than a timing item.
⚠️ Correction to the record: the July file's "IBM-adjusted FCF ~$13.4B, guides ~$14.5B" was wrong on both legs — $13.4B was FY2024 operating cash flow. Correct: FY2025 IBM FCF $14.70B, FY2026 implied guide ~$15.7B.
1.4 Capital allocation — the policy is unambiguous
Four-year cumulative, FY2022–FY2025 (OCF $51.0B):
| Use | Cumulative $B | % of OCF |
|---|---|---|
| Dividends | 24.39 | 47.8% |
| Acquisitions | 19.02 | 37.3% |
| Capex | 7.21 | 14.1% |
| Buybacks, net | 0.33 | 0.6% |
| Net new debt raised | +7.71 | — |
Half the cash to the dividend, essentially all the rest to M&A, nothing to buybacks, with leverage as the marginal funding source. Acquisition intensity is escalating hard: $2.3B → $5.1B → $3.3B → $8.3B → $10.5B in H1 2026 alone. FY2025 spent 126% of OCF, plugged with $2.9B of net new borrowing.
1.5 Share count — the central per-share knock
| FY2022 | FY2023 | FY2024 | FY2025 | Q2 2026 | |
|---|---|---|---|---|---|
| Shares o/s (M) | 906.09 | 915.01 | 926.29 | 936.95 | 942.13 |
| YoY | — | +0.98% | +1.23% | +1.15% | +1.14% |
Every quarter in the series is higher than the one before. There is no buyback cycle and no offsetting quarter. The mechanism, TTM: $1,877M of stock compensation against $319M of net buyback — IBM repurchases 17 cents of every dollar it issues. SBC is compounding at +20.2%/yr against +3.7%/yr revenue and is now 15.0% of GAAP FCF.
Sized three ways: ~1.1pp/yr is subtracted from every growth rate before it reaches the owner (revenue +3.72% → revenue/share +2.33%; operating income +15.20% → +13.63%). At $235.98, the ~10.6M net new shares of the last four quarters transfer ~$2.5B of market value per year away from existing holders — 22% of GAAP FCF. Compounded, ~10.5% of per-share value per decade at unchanged fundamentals.
1.6 Balance sheet
| FY2024 | FY2025 | Q2 2026 | |
|---|---|---|---|
| Total debt | $58.40B | $64.61B | $65.27B |
| Cash + ST inv. | $14.59B | $14.42B | $8.13B |
| Net debt | $43.81B | $50.19B | $57.14B |
| Debt / assets | 42.6% | 42.5% | 42.9% |
| Net debt / EBITDA | 2.97x | 2.87x | 3.25x |
| Op income / interest | 5.09x | 6.12x | 6.59x |
| Goodwill | $60.71B | $67.72B | $74.60B |
| Goodwill + intangibles | $71.37B | $79.11B | $88.55B (58% of assets) |
| Tangible book value | −$44.06B | −$46.46B | −$54.10B |
Debt/assets is strikingly stable at 42–44% — IBM manages to a leverage target. But three years of deleveraging (3.49x → 2.87x) were given back in two quarters (3.25x) to buy Confluent. Interest coverage is comfortable and is not a near-term risk. Equity is 22.7% of assets and is $54B smaller than the intangible balance — any goodwill impairment goes straight through a thin base. Working capital is −$7.5B. Not distress; materially thinner than December.
1.7 Scorecard
| Dimension | Basis | |
|---|---|---|
| Operating margin trend | 🟢 | 13.5% → 18.5%, +500bp in three years |
| Operating income growth | 🟢 | +15.2%/yr, clean of one-timers |
| Dividend safety | 🟢 | Survives a 45% FCF cut before coverage breaks |
| Capex intensity | 🟢 | 13% of OCF — genuinely asset-light |
| Revenue growth | 🟡 | +3.7%/yr; Q2 only +1.1% |
| Leverage | 🟡 | Stable ratio, but a full turn given back in six months |
| Liquidity | 🟡 | Cash −47% in six months to $7.2B |
| ROIC | 🟡 | ~10.5%, was improving, now turning down |
| FCF trend (GAAP) | 🔴 | Peaked FY2023, down three straight years |
| Earnings quality | 🔴 | Negative effective tax rates twice; ~23% of FY25 EPS is a tax benefit |
| Share count | 🔴 | +1.1%/yr forever; net buyback 17% of SBC |
| Balance-sheet quality | 🔴 | Goodwill 49% of assets; tangible book −$54B |
| Dividend growth | 🔴 | +0.6%/yr for six years — negative in real terms |
2. Moat & Competitive Advantage
2.1 Quantitative base — ROIC is lower than we thought, and just turned down
| FY2022 | FY2023 | FY2024 | FY2025 | TTM | |
|---|---|---|---|---|---|
| NOPAT @ 21% ($B) | 6.45 | 7.41 | 7.96 | 9.87 | 10.04 |
| ROIC | 8.85% | 9.75% | 9.86% | 11.20% | 10.55% |
| ROE (reported) | 7.47% | 33.29% | 22.05% | 34.00% | — |
The honest number is ~10–11%, against the July file's "12–13%." The trend was genuinely improving on the 500bp of margin expansion, then fell to 10.55% because Confluent added ~$10.5B to invested capital with essentially no operating income attached yet. Expect further dilution until it contributes.
⚠️ The 34% ROE must not be used as a return-on-capital measure. Equity of $32.6B exists only because $170.6B of treasury stock is netted against retained earnings and paid-in capital. The denominator is an accounting residue of fifty years of buybacks, not capital employed. The 23-point gap between 34% ROE and 10.5% ROIC is the size of the distortion. ROIC of ~10.5% sits only modestly above an 8–9% cost of capital.
2.2 Segment structure — where the profit actually is
| Q2 2026 | Revenue | % rev | YoY | Gross margin | Segment margin | % of segment profit |
|---|---|---|---|---|---|---|
| Software | $7.76B | 45.2% | +5.1% (0% organic) | 82.6% (−1.3pt) | 32.2% (+1.1pt) | 61.1% |
| Consulting | $5.33B | 31.1% | flat (+1.1% cc) | 28.9% (+1.4pt) | 12.1% (+1.6pt) | 15.8% |
| Infrastructure | $3.84B | 22.4% | −7.4% | 58.4% (−3.1pt) | 21.8% (−1.5pt) | 20.4% |
| — of which IBM Z | — | — | −42% | — | — | — |
| Financing | $0.19B | 1.1% | +12% | — | 58.0% | 2.6% |
Two structural corrections to the July thesis:
(a) Consulting was never the soft flank. July weighted it by revenue (~28%) and called it the primary risk. On a profit basis it is ~16% — a Consulting shock is roughly one-fifth as damaging as the same shock to Software. And its margins are moving the right way, hard: segment profit +15.1% on flat revenue, which the 10-Q attributes to "the benefits of the productivity actions we have taken."
(b) Software is not independent of the mainframe — and this is the genuinely new finding. Transaction Processing fell −8.1% because IBM Z fell 42%, by IBM's own attribution; ~20% of Software is transactional revenue sold on mainframe ELAs. Adding Infrastructure (22.4% of revenue) to Transaction Processing (~14%), the mainframe complex is ~35% of revenue and plausibly 40%+ of segment profit. The bull case — "software is half of IBM now, so the mainframe cycle matters less" — is false to the degree the software is sold on the mainframe's purchase order. Segment diversification overstates economic diversification.
2.3 The mainframe question — cycle, not decline
| Quarter | IBM Z YoY | Cycle position |
|---|---|---|
| Q2 2025 | +70% | z17 GA (2025-06-18), launch surge |
| Q3 2025 | +59% | best Q3 in ~two decades |
| Q4 2025 | +67% | peak |
| Q1 2026 | +51% | still peak |
| Q2 2026 | −42% | laps the +70% launch quarter |
Two-year stack: 1.70 × 0.58 = 0.99. Z revenue is approximately flat versus Q2 2024 — a late-z16 trough quarter. The level has not broken below the prior cycle's floor; a structural decline would show each trough lower. And the quarter-5 rollover has happened every single cycle: z14 → −20.6%, z16 → −30%, z17 → −42%. Cadence z16→z17 was 3.1 years, so z18 lands ~mid-2028 and the z17 trough runs late-2027 into early-2028.
Verdict: ~80% cycle arithmetic, ~15% June budget displacement, ~5% unexplained. This is not a moat break.
Three honest caveats: −42% is steeper than z16's −30% at the same point (off a taller peak, but flagged); "130% program-to-program" is a management-defined, unaudited metric; and the deferral thesis is falsifiable in Q3.
2.4 Adversarial stress-test
- 🟢 Cannot be attacked — IBM Z / Transaction Processing. >140M installed MIPS, 85% of the base stable-or-growing, >70% of global transaction value. My attack as a rival requires a bank to rewrite forty years of COBOL that clears its own payments, accept migration risk on a system that cannot go down, and re-certify with regulators — for a saving that is a rounding error against the risk. Switching costs at the theoretical maximum. Nobody has done it at scale in 25 years; the hyperscalers' migration practices have produced logos, not a dent in MIPS. But immovable ≠ growing: this protects the base, not the growth rate.
- 🟡 Contested and losing — OpenShift. ARR $2.2B, inside a Hybrid Cloud line growing 11.2%, competing in a managed-Kubernetes market compounding >30%. OpenShift is growing at roughly a third of its market's rate. If a customer is committed to one cloud, EKS/AKS/GKE is cheaper and natively integrated. IBM wins where workloads genuinely span on-prem plus multiple clouds or where sovereignty forces it — a shrinking share of new workloads. RHEL underneath is a 20-year switching-cost asset; OpenShift on top is a 5-year one against three better-capitalised rivals. July treated "Red Hat +13%" as unqualified moat evidence; it is one durable asset stapled to one contested one.
- 🟡 Capped, not taken — Consulting. The obvious attack (undercut using my own AI delivery platform) is not landing: signings +6%, book-to-bill 1.05, backlog $30.8B, margins up on both lines. The successful attack is subtler — I don't take the revenue, I cap it. Signings +6% converting to +0.2% revenue means IBM is turning bookings into profit rather than volume.
- 🔴 The attack that actually worked was the budget, not a competitor. No rival took IBM's Q2; the customer's own capex allocation did. IBM has no moat against a memory/GPU scarcity that turns enterprise infrastructure into a hoarding market.
2.5 The AI question, both directions
🔴 Bear — IBM retired the scoreboard in the quarter it missed. The genAI "book of business" ran >$6B → >$7.5B → $9.5B → >$12.5B (Q4'25) and then stopped. The phrase appears zero times in the Q2 10-Q. It has been replaced by two ratios — "~50% of signings, >30% of backlog" — which cannot be compared across quarters the way a running total can. Applying ">30% of backlog" to the $30.8B Consulting backlog gives ~$9.3B of genAI Consulting backlog — the lowest-margin dollar IBM sells (28.9% GM vs Software's 82.6%), and signings, not ARR. Meanwhile IBM never separately sizes watsonx revenue. The recurring, high-margin, genuinely-moated half of the AI story is the half IBM declines to quantify. Retiring a growth metric in the quarter you miss is not proof of anything, but it removes the outsider's ability to verify the thesis — a governance and disclosure negative, arriving exactly when management's forecasting credibility took its worst hit in decades.
🟢 Bull — three positions that are structurally hard to attack. (1) Inference on Z: Telum II puts multi-model inference inside the transaction, where regulated data already lives and cannot legally or practically leave. For live fraud scoring this is not a preference — moving the data breaks latency and sovereignty at once. This is the one place mainframe lock-in converts into an AI franchise rather than a legacy tax. (2) Governance for regulated industries — watsonx sells control to buyers who cannot deploy a model they don't govern, into a 60-year trust position no startup can purchase. (3) The real-time data plane — once Kafka is a company's nervous system it is as sticky as an OS. And the most credible AI evidence IBM has is in the 10-Q rather than the press release: IBM Bob, its own AI dev system, GA March 2026, dogfooded on IBM's own developers — and Consulting's +140bp gross margin on flat revenue is what that looks like when it works.
⚖️ Judgment: AI winner in Software and on Z; AI-neutral in Consulting; AI victim in the budget — and that last one is underrated. Everyone asked "will AI eat IBM Consulting?" and the answer is no, on margin evidence. Nobody asked "will AI infrastructure capex crowd out IBM's software and mainframe budget line?" — and that was answered yes, in June 2026, at a cost of 25% in a day. IBM's AI risk is competition-for-wallet, not competition-for-capability. That is cyclical rather than moat-level, which is better news — but it recurs for as long as memory and GPU scarcity lasts, and nobody credibly dates the end of that.
2.6 Evergreen assessment
- 🟢 Genuinely evergreen: IBM Z + Transaction Processing (banks will clear payments on this in 2036), RHEL, the regulated-industry trust position.
- 🟡 Durable but not evergreen: Consulting — permanent as a business, permanently ~12% margin and ~0% growth. Automation/HashiCorp — Terraform's lock-in is real but OpenTofu proved it is forkable.
- 🔴 Not evergreen: OpenShift's competitive position. Distributed Infrastructure (+37% this quarter, −15% the same quarter last year — that is a shortage beneficiary, not a franchise).
Revised one-line moat statement: IBM is a mainframe-anchored franchise with the strongest switching-cost moat in enterprise technology, ~35% of revenue and plausibly 40%+ of profit tied to a hardware cycle that peaks and troughs every three years, wrapped in a software portfolio that is one-third acquired and whose largest line moves with the mainframe.
2.7 Moat defence — good where it is cheap, expensive where it is bought
| Defence | Risk/reward |
|---|---|
| AI-on-Z (Telum II, Spyre, compact LinuxONE) | 🟢 Excellent — cheapest, highest-leverage thing IBM does; defends 40% of profit with R&D it would spend anyway |
| Consulting Advantage / IBM Bob | 🟢 Good, and proven — the +140bp gross margin is the receipt |
| watsonx / Orchestrate | 🟢 Good — but IBM refuses to size it, which caps the credit it earns |
| HashiCorp (~$6.4B) | 🟡 Right target; Automation grew only +3.6% — synergy not yet visible |
| Confluent ($11.6B, $7.24B goodwill) | 🟡 Strategically right, expensively bought. It is now the load-bearing driver of Data +18.9% and of the ARR headline. Growth purchased at ~9x revenue is not moat evidence |
| Quantum >$10B/5yr + Anderon foundry | 🔴 Poor near-term. Real cash out, no dated revenue in, while buybacks stay suspended and share count drifts up 1%/yr. Defensible as a 10-year call option; indefensible as capital allocation for anyone underwriting the next three years |
| Buybacks | 🔴 Absent since the 2019 Red Hat close. Shareholders fund the moat defence through dilution |
3. Sentiment & Intelligence
3.1 The guidance change is not what the headline said
| Guide | Prior | New | |
|---|---|---|---|
| FY26 revenue cc | 5%+ | 4–5% | ⬇ |
| Software | ~double-digit | 6–8% | ⬇⬇ the entire cut |
| Infrastructure | down low-to-mid single | up low single | ⬆ +2pts |
| Operating pre-tax margin | — | +100bp | ⬆ raised |
| FCF | +~$1B | +~$1B | → held |
The cut is entirely a software cut, partially offset by an infrastructure raise, alongside a raised margin guide and a held FCF guide. That is a materially different picture from "IBM cut guidance."
3.2 Management held the line and quantified it
Krishna, on deferral vs. destruction — the load-bearing quote:
"The majority of what didn't happen in the second quarter was large CapEx deals at large clients… the volume or the number of those deals was in the low tens. We have been very pleased to see that about a third of those have already closed. Normally we would not expect all of them to close, but we would expect maybe two-thirds to three-fourths of them to close over the next six months. So the fact that a third have already closed in the first three weeks gives us a indication, not yet full evidence, but a good indication that this was deferral and not destruction."
The base rate is the strongest part of the bull case: normal six-month conversion is 66–75%; they hit ~33–40% in three weeks. The "not yet full evidence" hedge is the correct epistemic status.
On the mechanism — a price shock, not a preference shift:
"our clients themselves had not really thought through that some of the alternate purchase they were doing were increasing 30% in dollar value quarter to quarter. When they were faced with that issue, then they decided to move budget to those areas."
Kavanaugh on why the damage was contained: "Because these are generally purchased as operating expenditures, they were largely unaffected by the CapEx dynamics that impacted some of our ELA transactions." Only the ~20% CapEx-treated ELA book broke; the ~80% OpEx book accelerated — recurring revenue growth 7% → 8% → ~10% guided for H2.
And they guided below their own stated belief. Kavanaugh, pressed by Jefferies on why the guide doesn't credit the recovery: "we've taken a discount off of our traditional historical rates. That's why we're sitting at 8%. I think that's prudent." A team fabricating a timing story does not refuse to put the recovery in the guide.
⚠️ The one place management conceded something other than timing — Krishna on why organic Data underperformed: "clients have a certain number of dollars… in case that there is aggregate budget pressure." That is not a timing statement. It is the crack in the story, and it aligns with Morgan Stanley's argument that total 2026 IT budget growth is roughly flat.
⚠️ Citi's Boolani asked what is happening in the 15% of MIPS not stable-or-growing. It was never answered.
3.3 The sector-cannibalization fear is dead
ServiceNow reported the same day, 2026-07-22, and cRPO — precisely the metric that cracks first if enterprises reallocate away from software — accelerated to $13.2B, +21.5%, 200bp above guide. Subscription revenue +23% cc, 150bp above guide; AI ACV crossed $1B; FY26 guidance raised. McDermott, asked directly whether AI-infrastructure spend was lengthening cycles: "I have not seen any change… If I've seen any change, it's on the positive."
| 7/14 close | 8/12 | Δ | |
|---|---|---|---|
| NOW | $104.85 | $124.94 | +19.2% |
| INTU | $282.43 | $334.71 | +18.5% |
| ADBE | $220.78 | $258.75 | +17.2% |
| CRM | $167.56 | $193.32 | +15.4% |
| IBM | ~$219 | $235.98 | +7.8% |
On 7/14 the market marked down five companies for one company's problem, and has since corrected four of them. IBM is the only one that has not recovered — and it is the one that cut guidance. Caveat: CRM, ADBE and INTU have not reported since 7/14; their contribution here is price evidence only. INTU on 2026-08-25 is the first fundamental test.**
3.4 Analyst positioning — targets slashed, ratings untouched
Across a 25% one-day collapse, exactly one firm downgraded a rating (HSBC, Hold → Reduce, PT 231 → 191). Twelve firms cut targets while reiterating; RBC didn't cut at all. The Street re-priced the multiple and did not re-underwrite the thesis — a vote for "timing," but a lazy one.
Mean target $244.16 vs $235.98 spot = +3.5% implied upside. Post-event median $250. The dispersion is the story: Morgan Stanley $190 (−35%, the biggest cut) against Citi $245 (−4%, the smallest), on the same print. That spread is the fork, priced as a distribution. (Oppenheimer's 7/14 raise to $350 was published hours after the pre-announcement and is undigested — discard.)
3.5 Insider activity — the loudest bearish datapoint
⚠️ insider_purchases returned all zeros while real open-market buys sat in insider_transactions — third confirmation of pitfall-yahoo-insider-purchases-counts-rsu-grants. Open-market rows only:
| Date | Insider | Role | Type | Shares | Price |
|---|---|---|---|---|---|
| 2026-02-25 | Michael Miebach | Director | BUY | 434 | $233.33 |
| 2026-02-25 | Michelle J. Howard | Director | BUY | 50 | $237.83 |
| 2026-01-30 | David N. Farr | Director | BUY | 1,000 | $304.00 |
| 2025-03-03 | Robert David Thomas | Officer | SELL | 26,543 | $253.01 |
| 2025-02-28 | David N. Farr | Director | BUY | 1,200 | $249.00 |
| 2024-11-13 | Alfred W. Zollar | Director | BUY | 936 | $210.62 |
Total open-market buying across ~21 months is ~$913k — token, likely ownership-guideline purchases. Zero officer open-market purchases in the entire dataset; Krishna and Kavanaugh receive ~131k shares/yr in grants and gift stock away.
The key observation: nobody bought a single share after the 25% crash. The window opened ~7/24; three weeks elapsed with nothing. Farr bought 1,000 shares at $304 in January and has had four weeks to buy the same stock at $219–236 and has not. Read it as a yellow flag rather than red — director buying is calendar-clustered around February — but it is the strongest single dissent from management's deferral story.
⚠️ Q2 13F filings were due 2026-08-14, two days after this report. How institutions actually reacted to the crash is unavailable and is a real gap.
3.6 Post-Q2 events
- 2026-08-11 — Together AI, $240M multi-year. Nvidia HGX B300 inference cluster on IBM Cloud, live ~Q1 2027. This directly monetizes the same AI-infrastructure demand that caused the Q2 miss.
- Lightwell — open-source vulnerability remediation at $1M/yr; early adopters include BofA, BNY, Citi, Goldman, JPMorgan, Mastercard, Morgan Stanley, RBC, State Street, Visa, Wells Fargo. Krishna sizes the TAM at "$5 billion or $10 billion or $15 billion."
- Krishna: "we have not taken increases" on hardware pricing through July — an unexercised pricing lever while competitors raise.
- No buyback authorization. $3.2B returned in H1, 100% dividends.
4. Valuation
Inputs: price $235.98 · diluted shares 953.3M · market cap $225.0B · net debt $57.1B · EV $282.1B · dividend $6.76 · TTM diluted EPS $11.27 (reported) / ~$8.60 (normalised at 21% tax) · BVPS $36.57 · tangible BVPS −$57.42.
| Model | Output | Weight | Note |
|---|---|---|---|
| Graham √(22.5×EPS×BVPS) | $96 (reported EPS) / $84 (normalised) | ~Zero | Structurally inapplicable. Tangible book is −$57/share; goodwill+intangibles are 58% of assets. Graham punishes asset-light, goodwill-heavy businesses by construction. Reported, not used. |
| DDM (Gordon, g = 0.6% actual) | $92 at ke 8% ($106 at 7%) | High | ⚠️ The July file used g = 3.0% and produced $139 — five times the real dividend growth rate. At the actual 0.6%, the dividend alone justifies under $100. The other ~$135 of the price is the franchise and growth story, which must deliver. |
| DYT (5yr avg yield 3.85%) | $176 | Medium-High | Even at the 52-week low of $199.19 the yield reached only 3.39% — still below its own 5-year average. The stock never actually got cheap on its own dividend history. Discounted somewhat because the 5yr band spans the pre-re-rating 2021–23 era when IBM was a $110–140 value stock. |
| Reverse-DCF | Implies ~3.4% perpetual growth at 8% WACC | High | On owner FCF (GAAP FCF − SBC = $11.27B) plus after-tax interest. Revenue grows ~4%, shares grow 1.1% → ~3% per-share owner growth is roughly what is already happening. This says fairly valued — neither bargain nor trap. |
| Bogle expected return | 4.4%/yr (with mild de-rating) to 7.9%/yr (flat multiple) | Medium | 2.86% yield + ~5% per-share earnings growth − 0 to 3.5% multiple compression. Mid-single-digit expected return. |
| EV / FCF | 18.0x on IBM's FY26 guide · 21.5x GAAP TTM · 25.0x owner FCF | High | The owner figure is the honest one: SBC of $1.88B is a real cost IBM never buys back, so it belongs out of FCF permanently. |
⚠️ A note on the forward multiple. Yahoo's forwardPE of 17.91 implies forward EPS of $13.18 — a +17% jump from the verified TTM $11.27, in a year IBM just cut revenue guidance and Q2 EPS fell 1.7% YoY. That is not credible as FY2026; it is most likely FY2027 or a different basis (pitfall-vendor-forward-eps-is-the-wrong-fiscal-year). Yahoo's trailingPE of 20.98 ties exactly to the verified TTM diluted EPS of $11.27 and is the field that can be trusted. The trim is therefore written on the ttm basis — see pitfall-multiple-trim-inherits-the-broken-vendor-field.
Fair value: $195–250, central ~$220. Spot $235.98 sits in the upper third.
Synthesis. The dividend models (DDM $92, DYT $176) say expensive, and they are not wrong so much as narrow — they value IBM as the income stock it stopped being in 2018. The cash-flow models say fair: at $236 the market is asking for ~3.4% perpetual growth of owner cash flow, which is approximately what a 4%-revenue-growth business diluting 1.1%/yr actually delivers. That convergence — "priced for exactly what it does" — is the answer. There is no margin of safety and no obvious overvaluation. Consensus agrees: +3.5% implied upside.
- Entry $185–205. At $195 the yield reaches 3.47% (near its own 5yr band), the reverse-DCF requirement falls to ~2.5% perpetual growth, and Bogle clears 8%/yr even with mild de-rating. That is a real margin of safety.
- Trim 24x ttm (~$270). At $270 the yield falls to 2.50%, EV/IBM-FCF reaches ~20.5x, EV/owner-FCF ~28x, and the Bogle expected return drops to ~3%/yr. For a 4% grower diluting 1.1%/yr at 10.5% ROIC, that is a legitimate sell-into-strength level.
5. Type Overlay — Dividend Grower
| Yield | 2.86% ($6.76 annualised) |
| 10yr dividend CAGR | +1.90% |
| 8yr dividend CAGR | +0.92% |
| 5yr dividend CAGR | +0.60% |
| Pre-2018 (2014→2018) | +9.30% |
| FCF payout (GAAP) | 49.8% (FY23) → 52.3% → 54.6% (FY25) — deteriorating |
| FCF payout (IBM defn.) | 48.2% (FY24) → 42.6% (FY25) — improving |
| Streak | 30+ years of increases; consecutive quarterly dividends since 1916 |
Dividend safety: 🟢 genuinely strong. A 30% cut to GAAP FCF leaves the payout at 78% with $1.8B to spare. GAAP FCF would have to fall 45% — to $6.3B, a level not seen post-spin — before the dividend was uncovered. With beta 0.70, capex at 13% of OCF, and 6.6x interest coverage, that is a remote scenario.
Dividend thesis: 🔴 broken, and this is the overlay's real verdict. The framework prizes an 8–10yr dividend CAGR. IBM delivers 0.92%, and the forward run-rate is 0.60% — one penny per year, six years running, against ~2.5–3% inflation. This payout loses ~2pts of real purchasing power annually and the streak exists to protect the Aristocrat label, not the holder.
⚠️ A methodological note on DYT here. Normally a flat dividend makes the yield band suspect. Here it does the opposite: because the dividend barely moved, essentially 100% of the yield compression from ~5.5% (2021) to 2.86% is price. The dividend rose 3% cumulatively over five years while the price rose 97%. That makes the DYT signal cleaner than usual in the expensive direction — and it says the entire five-year return was multiple expansion, not income growth. It also inverts the usual trap in pitfall-dyt-inverts-when-price-caused-the-yield: the price caused the yield, and that is exactly why the signal is readable.
Verdict: safe dividend, dead dividend growth. Separate the two — a covered payout is not a buy case.
6. Phase 3 — Debate Round
Two genuine conflicts surfaced between agents. Both resolved into something sharper than either position.
⚔️ Conflict 1 — Is IBM Consulting a Genpact or an Infosys?
- Moat argued Genpact: gross margin +1.4pt, segment margin +1.6pt, segment profit +15.1% on flat revenue. IBM is keeping the AI productivity, not giving it back. Infosys's diagnostic signature — flat gross margin for four straight years — is absent.
- Sentiment argued Infosys: signings +6% and book-to-bill 1.05 converting to +0.2% revenue is exactly the services-deflation mechanism EPAM cut guidance for. Genpact's control-group print was +7.1% revenue on −2.6% headcount; IBM has the headcount cut without the revenue.
Resolution — both are right, and the pattern needs a third box. pattern-headcount-revenue-divergence-tests-ai-deflation currently has two states. IBM occupies a third:
| State | Margin | Revenue | Reading |
|---|---|---|---|
| Infosys | flat | stalled | Giving productivity to the client and not growing — worst |
| IBM | expanding | stalled | Keeping productivity, not converting it to volume — harvesting |
| Genpact | expanding | growing | Keeping it and growing — best |
IBM Consulting is harvesting, not compounding. It is the good branch of the pattern — no price give-back, no margin erosion — but it is not a growth engine and will not become one. Value it as a stable ~12%-margin profit annuity at ~16% of segment profit, and nothing more.
⚠️ And a caveat that invalidates the cleanest-looking number in the file. IBM-wide headcount fell 2.2% (270,300 → 264,300) against +7.6% FY25 revenue — a +9.8pt divergence, almost identical to Genpact's +9.7pt. That comparison is invalid and should not be used. It is firm-wide, so it includes Software and Infrastructure staff; FY2025 revenue was inflated by the z17 hardware surge, which consumes no consultants; and HashiCorp/Confluent added revenue and headcount on entirely different ratios. IBM does not disclose segment headcount at all, so the pattern's test cannot be run on IBM Consulting. The margin line is the valid evidence, and it points the Moat analyst's way.
→ The July file's correlation warning was wrong. It said buying IBM "adds correlated exposure to the segment that's hurting the portfolio." On a profit basis it overstated the ACN/EPAM overlap by roughly 2x, and had the direction of travel backwards — IBM Consulting's margins are expanding while ACN cut guidance and EPAM blamed "faster-than-expected shift away from task-based services."
⚔️ Conflict 2 — Is software growth 8% or zero?
- Sentiment reported: ARR $24.6B +8%, recurring revenue accelerating 7% → 8% → ~10% guided.
- Moat derived: ARR rose ~$1.8–2.0B, of which Confluent contributed ~$1.2–1.3B → organic ARR ~+2.5%, and reported Software +5.1% masks far less.
Resolution — no real conflict; Sentiment's own primary data settles it. IBM discloses Q2 Software as +5% reported, flat organic. The two agents converge: the reported software growth was purchased. ARR +8% is ~+2.5% organic; Data +18.9% is Confluent; Software +5.1% is 0% organic.
The fair adjustment: Q2 organic software is depressed by the $400–500M of deferred ELAs. Add them back and organic software is ~+6% — real, but still well short of the double-digit ambition. Both the bear framing ("zero organic growth") and the bull framing ("+8% ARR") are misleading; ~6% is the honest number, and it depends on the deferred deals actually landing.
🔗 Cross-agent synthesis neither agent made alone
Sentiment flagged the missing buyback as a conviction signal. Fundamentals independently showed IBM had $7.2B of cash left after Confluent, net debt/EBITDA back at 3.25x, and a policy that already spends 126% of OCF. Put together: IBM did not decline to buy the dip — it had no capacity to. That reframes a soft sentiment signal as a hard balance-sheet fact, and it carries a real consequence: if the stock falls again, IBM cannot support it.
7. Synthesis — Weighted Verdict
Weighting. For a mature, low-beta dividend payer, analysis_notes §5 puts Fundamentals and Valuation ahead of Moat-disruption fears. That weighting is decisive here, because the three desks split cleanly by domain:
| Desk | Read | Weight |
|---|---|---|
| Moat | 🟢 Better than July thought. Z is a cycle, not a break. Consulting is fine. Moat intact and better understood. | Medium |
| Sentiment | 🟡 70/30 timing over structural. Sector fear refuted. Management held the line, quantified it, guided below it. But no insider bought the dip and the AI scoreboard was retired. | Medium |
| Fundamentals | 🔴 The most negative desk. 0.6% dividend growth, +1.1%/yr dilution forever, ROIC 10.5% and falling, a full turn of leverage given back, cash −47%, 126% of OCF spent, growth bought at ~9x revenue, ~23% of EPS from a negative tax rate. | High |
| Valuation | 🟡 Fair, not cheap. Reverse-DCF implies ~3.4% perpetual growth — approximately what IBM delivers. Consensus upside +3.5%. | High |
The weighted call. IBM is a durable, defensive, genuinely-moated business whose competitive position survived the Q2 stress test better than the market or our own July file assumed. It is also a business that converts very little of that durability into per-share value for its owner: growth is partly purchased, the dividend has not grown in real terms since 2018, the share count rises every single quarter, and returns on capital sit only modestly above the cost of capital. At $235.98 the market is asking for roughly what IBM actually produces. That is a fair price, not an opportunity.
The July file framed this as "value versus value trap, unresolved." It is neither. It is a fairly-priced, slow, safe compounder with a cyclical earnings profile — and the honest disappointment is that after a 25% crash and a full re-underwriting, there is no dislocation left to harvest.
Conviction: [5.5] — HOLD / WATCH. Do not initiate at $236.
Down from [6.0]. The downgrade is not a moat downgrade — on the business, this file is more confident than July's. It is driven by four per-share corrections, three of which were unknown on 7/14:
- Dividend growth is 0.6%/yr, not ~3% → DDM falls from $139 to $92.
- ROIC is ~10.5%, not 12–13%, and turned down on Confluent.
- Organic software growth is ~0–6%, not 10% — the headline was acquired.
- Confluent ($11.6B, closed 2026-03-17) re-levered a full turn and drew cash down 47%, and was not in the July file at all.
Meanwhile the price is 7.7% higher. Better information, worse arithmetic, higher price.
A portfolio-specific passage was removed from the public build.
Key risks
- 🔴 The FY26 FCF guide requires H2 +10.1% after an H1 of exactly 0% — and rests on a $600M inventory buy-ahead that becomes a writedown if H2 demand does not arrive. The single most testable claim in the file.
- 🔴 Consulting deflation — signings +6% converting to +0.2% revenue, on ~31% of revenue, unverifiable because IBM discloses no segment headcount. Not a timing problem.
- 🔴 Permanent 1.1%/yr dilution with net buyback at 17% of SBC and no capacity to change it.
- 🟡 The z17 trough runs into late-2027/early-2028 before z18 — two more years of hard comps dragging Transaction Processing with them.
- 🟡 Flat aggregate IT budgets — Krishna's own "clients have a certain number of dollars," plus Morgan Stanley's flat-2026 argument. A fixed pie means the ELA recovery comes out of something else.
- 🟡 Goodwill $74.6B on $34.5B of equity, tangible book −$54B. A single impairment goes straight through.
- 🟡 Disclosure quality — the genAI book of business was retired in the quarter IBM missed, and watsonx is never sized. The AI thesis is now unverifiable from outside.
- 🟡 Quantum >$10B/5yr with a 2029 verification date, funded while buybacks stay at zero.
Golden points
Z is un-rip-and-replaceable and the −42% is cycle arithmetic (two-year stack flat) · ~⅓ of deferred deals closed in three weeks against a 66–75% six-month norm · management guided below its own stated belief · margin guide raised and FCF guide held inside a revenue cut · Infrastructure guide raised 2pts · Consulting margins +140/160bp — AI productivity retained, not given back · ServiceNow's cRPO refuted the sector cannibalization thesis outright · operating margin +500bp in three years · dividend survives a 45% FCF cut · beta 0.70 · Together AI $240M monetizes the very demand that caused the miss.
A portfolio-specific passage was removed from the public build.
9. What Would Change This
Upgrade to [6.5–7.0]: Q3 (late Oct) confirms both of Kavanaugh's own named KPIs — mainframe holding 120%+ program-to-program and software recurring reaching ~10% — with the FY26 FCF guide intact · or any officer open-market purchase · or a genuine buyback authorization, which would signal both conviction and restored capacity.
Downgrade to [4.5]: Consulting posts a second straight quarter of book-to-bill >1.0 with flat revenue (deflation confirmed, structural, and ~31% of revenue re-rates) · or the FY26 FCF guide is cut · or the H1 inventory build is written down · or organic software growth stays below 4% with the deferred ELAs already recovered.
Dated checkpoints:
| Date | Event |
|---|---|
| 2026-08-14 | Q2 13F filings — how institutions reacted to the crash (2 days after this report) |
| 2026-08-25 | INTU Q4 FY26 — first fundamental test of the software cohort since 7/14 |
| ~2026-10-21 | IBM Q3 FY26 — the referee. Three tests in order: (1) does deferred Z/ELA revenue land and does Transaction Processing recover with it; (2) does the inventory build convert or write down; (3) does Consulting hold >28% gross margin |
| ~mid-2028 | z18 launch — end of the z17 trough |
Data: fin.py / Yahoo Finance MCP (statements, holders, recommendations, dividend actions); IBM's official 2Q26 webcast transcript and prepared remarks (ibm.com/investor, downloaded in full — all management quotes verbatim from it); IBM Q2 2026 10-Q (SEC, filed 2026-07-23) and FY2025 10-K; IBM Q2 2026 press release; ServiceNow / Cognizant / Genpact / Accenture / EPAM / Infosys primary results releases. roic.ai returned only 2 annual periods (free-plan cap confirmed) and requires NYSE:IBM rather than a bare ticker. SEC.gov blocked both WebFetch (403) and agent-browser; the 7/14 8-K is secondary-sourced but fully corroborated by the official transcript. Multi-year CAGRs are capped at 3 years by the Kyndryl spin plus vendor history limits — flagged, not fabricated. Yahoo insider_purchases returned all zeros against real open-market buys in insider_transactions (known pitfall, third confirmation). Yahoo's dividend feed applies a 1.046 Kyndryl-spin factor that silently deflates all pre-Nov-2021 dividends — corrected here and cross-validated against roic.ai. Alpha Vantage, FactSet and S&P Global connectors require OAuth and were unavailable in this non-interactive session.