International Business Machines IBM

Reference table

1. What the business is, and the segment detail that matters

Chief executive Arvind Krishna; 286,800 employees; NYSE-listed; CIK 0000051143; reporting in USD (profile.currency and inc_a[0].reportedCurrency both USD — no currency-mixing trap).

Four reportable segments. The vendor's seg_prod labels one of them wrongly. The vendor calls the third segment "Infrastructure Services"; the 10-Q's segment tables call it Infrastructure. The distinction matters because IBM had a separate "Infrastructure Services" business that was spun off as Kyndryl in 2021, and a reader could reasonably mistake the vendor's label for that. We use the 10-Q's vocabulary.

The full-year 2025 picture, and the June 2026 quarter that broke the trend:

The segment lines sum to $67.472B against reported 2025 revenue of $67.536B — a $64M, 0.09% residual. This tie is clean and we record it as such. The vendor also supplies a "Segment Reconciling Items" line of −$2M for 2025 and +$243M for 2024, which is the bridge.

Inside Infrastructure, the detail from the earnings release is the single most important number in this dive:

> "Infrastructure — revenues of $3.8 billion, down 7 percent: Hybrid Infrastructure down 10 percent · IBM Z down 42 percent · Distributed Infrastructure up 37 percent · Infrastructure Support down 1 percent."

IBM Z down 42% is a mainframe product cycle rolling over. This is normal, recurring and well understood — mainframes sell in three-year waves and the trough is always ugly. But it is 42%, it is the highest-margin revenue IBM has, and it lands in a quarter when Consulting grew 0.2%. The offset is genuinely encouraging: Distributed Infrastructure up 37%, with the release stating "Power and Storage grew at a record pace in the second quarter, now having built up an order backlog of nearly $500 million."

Geography (seg_geo) ties exactly. Americas $33.342B, EMEA $22.189B, Asia Pacific $12.004B — $67.535B against reported revenue of $67.536B. Verified clean. The EMEA line is the growth story nobody discusses: $18.492B in 2023 to $19.429B in 2024 to $22.189B in 2025, up 20.0% in two years, against the Americas' +5.3%.

Note that seg_geo and seg_prod both carry deep history that is not comparable. The 2022 seg_prod entry reports categories called Sales, Services, Financing. and Extended Warranty; the 2020 and earlier entries use the pre-Kyndryl segment names (Global Technology Services, Cloud And Cognitive Software); the 2017, 2016 and 2012 seg_geo entries carry a nonsense field called "J [P]"; and the 2014 entry reports a single line of $192.945B against a company whose revenue that year was under $93B. Everything before fiscal 2021 in these two blocks is discarded and nothing in this dive is computed from it.

2. The June quarter, and what management said

Note the Q4 2025 row: GAAP diluted EPS of $5.88 EXCEEDS operating non-GAAP EPS of $4.52 by $1.36. For almost every company in this batch the relationship runs the other way, because non-GAAP excludes charges. Here it inverts because a discrete tax benefit landed in GAAP and not in the operating measure — roughly $1.3B on 952.4M shares. This is the META-class defect and it is why the FY2025 GAAP figures should not be treated as run-rate (Section 6, finding 3).

And the June 2026 quarter is where the deceleration is. From the 8-K's own income-statement summary: revenue $17.2B, GAAP gross profit $9.9B at a 57.7% margin, GAAP pre-tax income $2.5B at a 14.4% margin, GAAP net income $2.2B, GAAP diluted EPS $2.27. The 10-Q adds: "Operating (non-GAAP) pre-tax income from continuing operations of $3.3 billion increased 2.9 percent... and the operating (non-GAAP) pre-tax margin from continuing operations increased 0.3 points to 19.2 percent."

Margin expanded. Revenue did not. That is the entire quarter in one line, and it is why the operating EPS came in exactly on consensus while revenue missed by 1.7%.

Management's own framing, half-weighted per house rule:

> "Although we faced revenue headwinds late in the second quarter, we continued to focus on the fundamentals of our business, including driving productivity, strengthening our portfolio, and generating free cash flow" — James Kavanaugh, chief financial officer.

> "The company now expects full-year constant currency revenue growth in the range of four-to-five percent. At current foreign exchange rates, currency is expected to be neutral to growth for the year."

> "The company continues to expect full-year free cash flow to increase by about $1 billion year-over-year."

The arithmetic of that guidance deserves to be shown. First-half 2026 revenue was $33.079B against $31.518B — +5.0%. The second quarter was +1.1%. To reach 4-5% for the full year with currency neutral, the second half has to run at roughly 3.5-5%, i.e. it has to accelerate from the quarter just reported by two to four percentage points. Consensus for the September quarter is $16.842B, or +3.1%. That is the falsifiable claim the price rests on and it is 78 days from being tested.

3. Balance sheet — three corrections, and the largest is a double-count

The vendor's most recent balance sheet is 2025-12-31. The 10-Q filed 2026-07-23 carries 2026-06-30, and between the two dates IBM closed an $11.5B acquisition.

Correction 1 — the vendor's total debt is $5,894M too high, and the mechanism is exact. The 10-Q states plainly: "Total debt of $61,987 million at June 30, 2026 increased $727 million compared to" December, and the debt table gives $61,260M at 2025-12-31. The vendor reports $67,154M. The difference decomposes without a remainder:

> Vendor shortTermDebt $7,224M = filing short-term debt $6,424M + current operating lease liability $800M

> Vendor longTermDebt $57,383M = filing long-term debt $54,836M + non-current operating lease liability $2,547M

> Vendor capitalLeaseObligations $2,547M = the same non-current operating lease liability, counted a second time

> Vendor totalDebt = $7,224M + $57,383M + $2,547M = $67,154M

> Overstatement = $800M + $2,547M + $2,547M = $5,894M

Treating operating leases as debt is a defensible convention. Counting $2,547M of them twice is not. The vendor reports total debt of $67,154M; the 10-Q shows $61,260M at the same date and $61,987M at 2026-06-30; we use $61,987M.

Correction 2 — netDebt omits marketable securities, and is also seven months stale. The vendor's netDebt of $53,513M is totalDebt less cash and equivalents only, ignoring $830M of marketable securities. Corrected to the same date on the filing's debt figure, net debt at 2025-12-31 was $61,260M − $14,471M = $46,789M — the vendor overstates it by $6,724M, or 14.4%. The current figure at 2026-06-30 is $61,987M − $8,178M = $53,809M. So the vendor's number happens to be within 0.6% of the right answer, for the wrong date, by two offsetting errors. We state this plainly because a reader checking only the headline would conclude the field was fine.

Correction 3 — the balance sheet predates Confluent. IBM completed the acquisition of Confluent, Inc. on 2026-03-17, paying $11,268M of cash for the outstanding common stock plus $269M for equity awards, against total assets acquired of $13,647M and total liabilities assumed of $2,045M. The effects are all over the June balance sheet: goodwill +$6,882M to $74,599M, intangibles +$2,564M to $13,955M, and cash down $6,293M. The 8-K states: "The company invested $10.5 billion in acquisitions this year." None of this is in the vendor's balance sheet, and any screen using vendor goodwill, intangibles, cash or equity for IBM is reading a pre-deal company.

The rebuilt enterprise value:

> Market cap $221.543B (942M shares × $235.15)

> + total debt $61.987B (10-Q, 2026-06-30)

> + noncontrolling interests $0.089B

> − cash, restricted cash and marketable securities $8.178B

> = corrected enterprise value $275.441B

The vendor reports $279.643B, an overstatement of $4.202B (1.5%).

Our EBITDA is trailing operating income of $12.343B (the four quarters to 2026-06-30) plus FY2025 depreciation and amortisation of $5.021B. The vendor's ebitdaMarginTTM of 23.48% implies $16.22B, which against $12.343B of operating income implies only $3.88B of D&A — below the $5.021B the cash-flow statement reports and falling in a year when Confluent added $2.564M of intangibles to amortise. **We use $17.36B and treat it as a conservative floor.

Leverage of 3.10x is still real and it is the highest in this batch outside the financial names. Interest coverage is 6.4x — the second-tightest here. Goodwill and intangibles of $88,554M are 58.2% of total assets and 2.6x total equity of $34,541M. The vendor's priceToBookRatioTTM of 6.415x and debtToEquityRatioTTM of 1.895x are arithmetically correct and economically uninformative for the same reason: book equity is what remains after a $170.9B treasury-stock balance and decades of acquisition accounting. returnOnInvestedCapitalTTM of 10.45% is the honest measure and is what we use — and it is the number the knowledge base's most testable claim about IBM turns on.

4. Cash flow and capital returns

Free cash flow is the strongest single fact about IBM and it deserves the emphasis. $11.575B on revenue of $67.536B is a 17.1% free-cash-flow margin achieved with capital expenditure of $1.617B — 0.4% of revenue and 32% of depreciation. The company converts more than 100% of GAAP net income to operating cash flow. freeCashFlowYieldTTM of 6.58% is materially better than the 4.85% earnings yield, and it is the reason the sell-side low target of $175 sits alongside a high of $350: on cash flow this is cheap, on growth it is not.

But the first half of 2026 shows the strain. Operating cash flow was $7.8B, up $1.7B year on year, and free cash flow was $4.8B, flat. The 8-K is explicit: second-quarter free cash flow was $2.5B, down $0.3B year to year. Management guides full-year free cash flow to rise about $1B off $11.575B, which requires the second half to do the work the first half did not.

Capital allocation has shifted decisively from buybacks to acquisitions. Buybacks were $407M, $402M, ZERO and $1.018B across 2022-2025 — effectively nothing, against a $221.5B market capitalisation. Acquisitions were $1.076B, $5.086B, $2.591B, $8.293B, and $10.5B in the first half of 2026 alone. Diluted share count has RISEN from 912.3M (2022) to 952.7M (Q2 2026) — up 4.4% — because stock compensation is not offset. IBM is not returning capital through the share count; it is diluting slowly and buying companies.

Dividend. The vendor reports lastDividend: 6.73 and dividendPerShareTTM: 6.73, yielding 2.86%. That $6.73 is the trailing-twelve-month total, not the last declared dividend — the same mislabelling seen across this batch. The 8-K gives the actual rate: "The IBM board of directors approved a regular quarterly cash dividend of $1.69 per common share... With payment of the September 10, 2026 dividend, IBM will have paid consecutive quarterly dividends every year since 1916." The forward run-rate is $6.76 — a 2.87% yield. Dividends paid in the first half were $3,166M against net income of $3,381M — a 94% GAAP payout, but only about 66% of first-half free cash flow.

Total shareholder yield is roughly 2.9%, and it is almost entirely the dividend. The buyback contributes nothing net.

5. Valuation — priced in or room?

At $235.15 (market cap $221.543B, corrected enterprise value $275.441B):

Two estimate-block problems.

The FY2029 and FY2030 rows are non-monotonic and thin, and are excluded. Consensus operating EPS is $14.290 in FY2028 (6 analysts), $14.110 in FY2029 (4) and $12.560 in FY2030 (2) — a 12% decline over two years while revenue rises from $76.909B to $81.471B. No conclusion in this dive touches FY2029 or beyond, and the FY2028 row is used only for the bull case, with its six-analyst coverage disclosed.

est.ebitdaAvg and est.ebitAvg carry the fixed-margin fabrication signature and are rejected. Across all eight years, ebitdaAvg / revenueAvg = 20.45% and ebitAvg / revenueAvg = 12.23%, to four significant figures. IBM's actual FY2025 operating margin was 18.5% and its actual FY2025 EBITDA of $17.284B was 25.6% of revenue. The rows understate EBITDA by roughly 20% and EBIT by roughly 34%, in every year, by construction. Per the data contract, all forward valuation runs on epsAvg.

Peer context. The vendor peer set is unusable in aggregate: Micron ($1.008 trillion) and Magic Software ($853M) cannot both be IBM comparables, and Information Services Group at $210M is a rounding error. The defensible names are Accenture ($104.3B at $170.43), SAP ($227.8B at $195.49), Cisco ($479.8B at $121.74) and Salesforce ($156.4B at $190.99) — and IBM at $221.5B sits between SAP and Salesforce in size while growing far more slowly than either. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.

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

At $235.15 — 19.1x FY2026 consensus operating EPS and 17.9x FY2027 — the price embeds:

• Revenue re-accelerates from +1.1% in the June quarter to +3.1% in September and 4-5% constant currency for the full year. (Consensus and company guidance; 17 analysts on the FY2026 revenue line.) This is the most fragile assumption in the price. The first half ran +5.0% and the second quarter +1.1%; the guidance requires the second half to run at 3.5-5%. It is tested on 2026-10-21.
• Operating EPS reaches $13.163 in FY2027 and $14.290 in FY2028 — about 8% compound. (Consensus; 14 and 6 analysts.) Since revenue grows about 4.4%, roughly half the earnings growth must come from margin. Management has guided to "improved pre-tax income margin expansion for the full year" — management voice, half-weighted. The June quarter delivered it: non-GAAP pre-tax margin +0.3 points to 19.2% on 1.1% revenue growth.
• The IBM Z decline is cyclical, not structural. (Our framing; the −42% is the company's own disclosure.) Mainframe revenue moves in product waves and always troughs. The falsifier is whether the next IBM Z cycle, when it comes, restores the prior peak — and the knowledge base contains a specific, dated claim that AI-assisted COBOL modernisation is the threat to exactly that base.
• Free cash flow rises about $1B to roughly $12.6B. (Company guidance, half-weighted.) First-half free cash flow was flat at $4.8B and the second quarter was down $0.3B. The whole increase has to come in the second half.
• The market keeps paying 17-19x on the forward year. (Our number.) At 15x FY2027E the stock is $197; at 21x it is $276. IBM traded at 26.7x FY2026E at its $329.23 high nine months ago and at roughly 16x when the highest-skill independent voice in our knowledge base flagged it as cheap in March 2025. The multiple has traversed that entire range inside eighteen months.

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

Expected return over roughly two years decomposes as: operating EPS growth (+16.1%, from FY2026E $12.313 to FY2028E $14.290) + multiple drift (COMPRESSION, from 19.1x to 17.5x on the respective forward year, −8.4%) + shareholder yield (+2.9% a year, ~5.9% compounded) ≈ +13% to +14% total, or roughly +6.5% annualised.

Our base assumes mild compression and the reason is growth, not risk. Today the market pays 19.1x on the next year. If in two years it pays 17.5x on the then-next year, that is a genuine de-rating of about 8%. We assume it because a business growing revenue 4% and earnings 8% does not sustain a high-teens multiple indefinitely once the AI-adjacency narrative that took the stock to $329 has been tested and found wanting twice — first by the −25% six-month move, then by a 1.1% revenue quarter.

Almost all of the expected return is earnings growth plus a 2.9% dividend, and the multiple is a modest headwind. If the multiple simply HELD at 19.1x on FY2028E the price would be $273 (+16.1%); if it compressed to 14x the price would be $200 (−14.9%). That 14x is not a stress case — it is roughly the multiple IBM carried for most of the decade before 2024.

The bull case at $347 requires an earnings beat AND a re-rating — 21x against the FY2028 consensus high of $16.544 — and it lands just below the street's own $350 high target. The bear case at $177 is 14x against the FY2028 consensus low of $12.625, and it sits just above the street's $175 low target. Both ends of our range are independently corroborated by the published target distribution, which is unusual and worth noting.

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

• On the business we are close to consensus, and the consensus itself is not one thing. 1 strong buy, 23 buy, 22 hold, 5 sell across 51 analysts, with targets from $175 to $350 — a 2.0x spread, the widest in this batch. Our $250 is 2.3% below the $255.92 mean. When the street cannot agree within a factor of two, "consensus" is a statistical artefact rather than a view, and our sitting near its centre is not agreement — it is the same uncertainty expressed differently.
• We differ materially on the balance sheet. Every screen shows IBM with $67.154B of total debt and 3.58x net-debt-to-EBITDA. The 10-Q says $61,987M and our corrected figure is 3.10x. The gap is $3,347M of operating leases treated as debt plus $2,547M of those same leases counted twice. That is not a rounding error on a name where interest coverage is 6.4x.
• We differ on the quality of the 2025 earnings base. FY2025 GAAP net income of $10.593B exceeded pre-tax income of $10.328B because the tax provision was minus $242M. In the fourth quarter, GAAP EPS of $5.88 exceeded operating non-GAAP EPS of $4.52 — an inversion worth roughly $1.3B — and the 2026 quarters carry positive provisions of $313M and $484M. Anyone comparing 2026 GAAP earnings to 2025 GAAP earnings is comparing a normal tax year to a benefit year. Our forward work runs entirely on the operating non-GAAP basis for this reason.
• We differ from the knowledge base's most cited claim on this name, and the divergence is now gradeable. The highest-skill independent voice in the file argued in March 2025 that IBM traded near a 16 P/E, half the Mag 7's, and "should re-rate to a market multiple as Mag 7 multiples compress." What happened is that IBM re-rated to roughly 26.7x forward at its $329.23 high and then gave the entire re-rating back, and is now at 19.1x. The claim was directionally right for months and is currently underwater. Our reading is that the re-rating was an AI-adjacency bid rather than a durable revaluation, and a 4%-growth business does not hold a market multiple without the growth.
• Positive surprise that would force a re-rate: September-quarter revenue at or above $17.0B (i.e. +4% or better) with Software above +5%, which would kill the deceleration thesis in one print; or an IBM Z announcement that starts a new cycle; or free cash flow tracking clearly toward the +$1B guide in the third quarter after a flat first half.
• Negative surprise that would break the thesis: third-quarter revenue below $16.5B (+1% or less), which would put the full-year 4-5% guide out of reach; Software growth below 4%; a further large acquisition funded by debt at 3.1x leverage; or free cash flow guidance being withdrawn. At 19.1x with 6.4x interest coverage there is a real de-rating available.

Synthos fair values

All three anchors are multiples of the FY2028 consensus operating-EPS distribution (mean $14.290, low $12.625, high $16.544, 6 analysts — coverage disclosed as thin), cross-checked against FY2027E of $13.163, which carries 14 analysts.

• Bear ~$177 — 14x the FY2028 consensus LOW of $12.625. Cross-check: 13.4x FY2027E; within 1.1% of the street's own low target of $175. The scenario: revenue growth stays near 1%, the IBM Z decline turns out to be partly structural, Consulting shrinks, and the multiple returns to the low-teens level IBM carried for most of the last decade. −24.7%.
• Base ~$250 — 17.5x the FY2028 consensus MEAN of $14.290. Cross-check: 19.0x FY2027E; 20.3x FY2026E; corrected EV of $290.7B on FY2028E revenue of $76.909B is 3.78x sales. Sensitivity, stated openly: 16x gives $229 — below spot — and 19x gives $272. The whole answer lives in a 16-19x band. The scenario: revenue re-accelerates to the guided 4-5%, margin expansion continues, free cash flow reaches roughly $12.6B, and the multiple drifts modestly lower as the market prices 4% growth as 4% growth. +6.3%, plus about 5.9% of compounded shareholder yield.
• Bull ~$347 — 21x the FY2028 consensus HIGH of $16.544. Cross-check: 26.4x FY2027E; just below the street's $350 high. The scenario: the Software segment carries the company, Confluent and HashiCorp compound inside it, quantum produces a commercial milestone, and IBM re-rates back toward where it traded nine months ago. +47.6%.

Base is 6.3% above spot; asymmetry roughly 1.9:1 to the upside (24.7% down, 47.6% up). The asymmetry is genuinely good and the base case is genuinely thin, and both facts are true at once. What holds this at Watch is that the base sits below the street, the June quarter grew 1.1%, and the entire near-term case rests on a re-acceleration that has not happened yet and is 78 days from being tested.

6. Knowledge base — the deepest lane in this batch, and the worst attribution

Raw hits: 32 (23 entity, 9 text). Used: 11. Discarded: 21. Searches covered IBM, International Business Machines, Red Hat, Arvind Krishna and watsonx across all 51,928 distilled claims.

The discards, by reason, because they are more instructive than the keeps:

• METAPHOR — 10 claims discarded. Six of nine text hits and four entity hits use IBM as a historical analogy and say nothing about the security: "Ethereum is the IBM/Microsoft of crypto"; "Reinventing the computer for the first time since IBM System/360 (1964)"; "the only such vertically integrated computer company since IBM in the 60s/70s"; "Nvidia alone accounts for ~40% of S&P 500 gains — more than 1960s IBM/AT&T monopolies"; "biotech... it's IBM-vs-HP with no aggressive young founders"; "OS markets naturally consolidate to a few standard winners (IBM, Windows, Android)". IBM is one of the most frequently used metaphors in investment conversation, and any entity-matched sweep on this name will be dominated by it. This is the single most important collision class for IBM and it is analogous to the Visa/"visa" homograph problem, but semantic rather than orthographic.
• PRODUCT-LEVEL, NOT EQUITY — 4 claims discarded. Two 2026 claims about GPU-accelerated data processing list IBM Watsonx among ecosystem names; a 2017 claim about voice assistants lists IBM Watson; a 2025-09-16 claim about asset tokenisation lists IBM incidentally. None is a claim about the security.
• DUPLICATE — 1 claim discarded. Two 2025-01-31 claims from the same channel and date make the identical argument (enterprise software as the second and third derivative of AI); we count one.
• STALE BASKET — 1 claim discarded. A 2021-05-27 rotation call listing IBM among eight global value names is five years old and basket-level.
• HELD SEPARATELY AS MANAGEMENT VOICE — 3 claims. See below.

The attribution defect, reported in full because it is exactly the failure the 4-lane policy exists to prevent. Three claims dated 2023-08-07 are filed under a channel slug named for the chief executive of a large pharmaceutical company. Their content is:

> "IBM's focus on hybrid cloud, AI, and quantum ('brutal clarity on strategy') positions it to grow by shedding low-value work and doubling down on differentiators." (bullish, conviction 80)

> "Quantum computing complements (not replaces) classical compute; solves intractable materials/chemistry/finance problems, a market that will explode like AI within 5 years." (bullish, conviction 65)

> "A quantum computer will do something that astonishes people within the next five years; value potential in the hundreds of billions across tens-of-trillion-dollar industries." (bullish, conviction 60)

These are IBM's own strategy, IBM's own quantum programme and IBM's own five-year framing, attributed to a channel named for a different company's chief executive. The speaker is, on the evidence of the content, IBM's chief executive appearing as a guest. They are excluded from the conviction pool entirely and quoted here at half weight as management framing. Their aggregate conviction of 80, 65 and 60 would otherwise have contributed the three largest bullish weights in the lane. This is the CRM/LLY failure mode in a new form: not one speaker under four channel labels, but one speaker under a channel named for someone else.

The eleven claims that remain, in date order:

What this lane actually says, and it is more useful than most.

First, the two highest-weight claims are from the same independent voice and they bracket the argument. In March 2025 he called IBM cheap at ~16x with a re-rating case; in April 2026 he called the 13% COBOL-disruption sell-off "biased and overdone, though some markdown is warranted." That second claim is the most balanced statement about IBM in the entire knowledge base — it concedes the threat is real and argues the market over-priced it — and it is our single best external cross-check.

Second, one claim is directly testable against this file and it fails IBM. The 2025-11-27 claim asserts that firms "below cost of capital (IBM)" should pay out rather than reinvest. IBM's returnOnInvestedCapitalTTM is 10.45%. Whether that clears cost of capital is arguable at the margin — but IBM's revealed behaviour is the opposite of the claim's prescription: buybacks of effectively zero across 2022-2024 against $8.293B of acquisitions in 2025 and $10.5B in the first half of 2026 alone, with the diluted share count RISING 4.4% since 2022. The claim is a criticism, IBM's conduct is the thing criticised, and the criticism is well-founded on the evidence in this file.

Third, the bear and bull cases in the lane are the same fact seen twice. The COBOL-modernisation claim and the quantum claim are both statements that IBM's installed base is exposed to a technology shift — in one direction it is the victim, in the other the beneficiary.

Net conviction: mixed. Breadth 11, with two claims carrying a named independent speaker and nine channel-attributed. No promoter-class voice appears. Three management-voice claims are held out. This is a real lane and it does not resolve to a direction — which is consistent with the Watch verdict rather than an argument against it.

7. Data integrity — what we rejected from the vendor file and why

Ten findings. IBM's file has the largest absolute balance-sheet error in this batch and one of the cleanest segment ties.

1. totalDebt of $67,154M at 2025-12-31 overstates the 10-Q's $61,260M by $5,894M, and $2,547M of the gap is a DOUBLE-COUNT — REJECTED and replaced. Mechanism, exact to the dollar: vendor shortTermDebt $7,224M = filing $6,424M + current operating lease $800M; vendor longTermDebt $57,383M = filing $54,836M + non-current operating lease $2,547M; vendor capitalLeaseObligations $2,547M = the same non-current operating lease, a second time. The vendor reports $67,154M; the 10-Q shows $61,260M at that date and $61,987M at 2026-06-30; we use $61,987M, and every leverage and enterprise-value figure in this dive is computed on it.

2. netDebt of $53,513M omits $830M of marketable securities AND is seven months stale — REJECTED. Corrected to the same date on the filing's debt, net debt at 2025-12-31 was $46,789M, so the field overstates by 14.4%. The correct current figure, at 2026-06-30, is $61,987M − $8,178M = $53,809M. The vendor's number is within 0.6% of the right answer for the wrong date, through two offsetting errors, and would pass a naive sanity check. We state this explicitly.

3. FY2025 GAAP net income exceeds pre-tax income because the tax provision is NEGATIVE — the run-rate is REJECTED. incomeTaxExpense for FY2025 is −$242M on pre-tax income of $10.328B, producing net income of $10.593B and an effective tax rate of −2.3%. ratios_ttm.effectiveTaxRateTTM is −2.53%. The effect is concentrated in the fourth quarter, where GAAP diluted EPS of $5.88 EXCEEDED operating non-GAAP EPS of $4.52 — an inversion worth roughly $1.3B on 952.4M shares. The 2026 quarters carry positive provisions of $313M (Q2) and $484M (six months), so the benefit has not repeated. The available filings in this archive do not disclose the driver — the 10-K text in the IBM archive covers the risk factors and exhibits but not the income-tax note, so we can size the effect but not attribute it, and we say so rather than guessing. All forward valuation in this dive runs on the operating non-GAAP basis, where the item does not appear. Note that IBM's tax provision has been negative in four of the six years in the file (2020, 2022, 2024, 2025), so this is a recurring irregularity rather than a one-off — which makes the 2026 reversal more, not less, significant.

4. The balance sheet predates an $11.5B acquisition — stated, with the filed figures substituted. IBM completed the Confluent acquisition on 2026-03-17, paying $11,268M for the common stock plus $269M for equity awards, against total assets acquired of $13,647M and liabilities assumed of $2,045M. Goodwill rose $6,882M to $74,599M, intangibles $2,564M to $13,955M, cash fell $6,293M to $8,178M including restricted cash and securities, and the 8-K states the company "invested $10.5 billion in acquisitions this year." None of this appears in the vendor's balance sheet. Every goodwill, intangible, cash, equity and total-asset figure in this dive uses the 10-Q.

5. est.ebitdaAvg and est.ebitAvg carry a fixed-margin fabrication signature — REJECTED outright. ebitdaAvg / revenueAvg = 20.45% and ebitAvg / revenueAvg = 12.23% in all eight years from 2023 to 2030, to four significant figures. Actual FY2025 EBITDA was $17.284B (25.6% of revenue) and actual operating margin 18.5%. The rows are constructed, not estimated. Per the data contract, all forward valuation runs on epsAvg.

6. The FY2029 and FY2030 estimate rows are non-monotonic and thin — EXCLUDED. Operating EPS of $14.290 (FY2028, 6 analysts), $14.110 (FY2029, 4) and $12.560 (FY2030, 2) declines 12% while revenue rises from $76.909B to $81.471B. No conclusion touches them. The FY2028 row's six-analyst coverage is disclosed wherever it is used.

7. Vendor trailing EBITDA is understated — recomputed. ebitdaMarginTTM of 23.48% implies $16.22B, which against trailing operating income of $12.343B implies D&A of only $3.88B — below the $5.021B the FY2025 cash-flow statement reports, in a year when Confluent added $2.564M of amortisable intangibles. We compute EBITDA as trailing operating income $12.343B + FY2025 D&A $5.021B = $17.364B and treat it as a floor. Consequential: evToEBITDATTM 17.24x → ~15.9x; netDebtToEBITDATTM 3.582x → ~3.10x.

8. seg_prod mislabels the Infrastructure segment and carries unusable deep history — corrected in presentation. The vendor calls the segment "Infrastructure Services"; the 10-Q calls it Infrastructure. The distinction matters because IBM spun off a business called Infrastructure Services (as Kyndryl) in 2021. The 2025 and 2024 lines tie well — the four segments plus reconciling items sum to $67.472B against reported revenue of $67.536B, a 0.09% residual. But every seg_prod entry for fiscal 2022 and earlier uses a different and non-comparable taxonomy (2022: Sales, Services, Financing., Extended Warranty; 2020 and prior: the pre-Kyndryl segment names), and seg_geo for 2017, 2016 and 2012 carries a field literally named "J [P]" while the 2014 entry reports a single line of $192.945B against a company whose revenue that year was under $93B. Everything before fiscal 2021 in both blocks is discarded and no figure in this dive comes from it.

9. seg_geo ties exactly for the years used — a clean check. Americas $33.342B + EMEA $22.189B + Asia Pacific $12.004B = $67.535B against reported revenue of $67.536B. Verified clean for 2021-2025.

10. lastDividend: 6.73 is the trailing-twelve-month total, not the last declared dividend — corrected from the 8-K. The board "approved a regular quarterly cash dividend of $1.69 per common share" on 2026-07-22. The forward run-rate is $6.76 — a 2.87% yield.

Share count — checked and verified clean. The 10-Q balance sheet gives 2,296M shares issued less 1,354M in treasury = 942M outstanding. Market cap of $221,542,810,100 divided by $235.15 gives 942,133,000 — agreement to within 0.01%. No TSLA-class share-count error. Note separately that the diluted weighted-average count used for EPS is 952.7M, 1.1% higher, which is the normal dilution wedge and not a defect.

incomeQualityTTM of 2.607 is not reconcilable and is not used. Operating cash flow per share of $15.84 against 942M shares implies TTM operating cash flow of roughly $14.9B, against TTM net income of $10.725B — a ratio of 1.39, not 2.607. We use the ratio we can derive and flag the field.

quote.yearHigh/yearLow ($332.46 / $199.19) disagree with tech.hi52/lo52 ($329.23 / $205.77) by 1.0% and 3.2%. We use tech throughout, per house rule.

Vendor composite rating — mechanism disclosed. rating gives B / 3, with 1 out of 5 on debt-to-equity, 1 on price-to-book and 2 on price-to-earnings, against 5 on return on equity and 4 on return on assets. The debt-to-equity and price-to-book 1s are driven entirely by a book equity of $34,541M that sits under $170.9B of treasury stock and $88.6B of goodwill and intangibles — and the debt-to-equity input is itself computed on the total-debt figure we have just shown is $5,894M too high. Both are rejected as lenses. returnOnEquityTTM of 33.5% is the same distortion in the opposite direction and is also rejected. returnOnInvestedCapitalTTM of 10.45% is the honest measure.

Non-equity tripwire — checked and passed. IBM is common stock, par value $0.20, NYSE-listed. Price $235.15 is not par-like; beta 0.705; the dividend is variable and was raised to $1.69; volume 7.21M shares (~$1.7B of turnover); the 52-week band of $205.77 to $329.23 is a 60% range. This is common equity.

8. Technicals

• Price $235.15. −28.6% from the 52-week high of $329.23; +14.3% above the low of $205.77. Position within the annual range: 24th percentile — the bottom quartile.
• Moving averages, both overhead: −9.6% below a 50-day average of $260.11 and −12.9% below a 200-day average of $270.01. Both are above the price and the 200-day is above the 50-day — the classic broken-trend configuration.
• RSI 69.07 — the highest in this batch, and it is the interesting number. A stock 28.6% below its high with RSI at 69 has rallied hard and recently. MACD is still −10.31, so the medium-term momentum has not turned; the short-term has.
• Maximum drawdown from peak over the trailing year: −28.6% — the current price is not the worst point; the low was $205.77.
• Relative performance: 3-month +2.5% vs SPY +7.6% and QQQ +7.7%; 6-month −25.3% vs SPY +11.1% and QQQ +15.7%; 12-month −6.0% vs SPY +24.3% and QQQ +30.8%. A thirty-six point deficit over six months and a thirty-point deficit over twelve. This is one of the worst relative performances in this batch.
• Sentiment: 1 strong buy, 23 buy, 22 hold, 5 sell across 51 analysts — the deepest coverage in this batch. Consensus target $255.92 (+8.8%), median $260, high $350 (+48.8%), low $175 (−25.6%). The $175-to-$350 spread is a factor of two and it is the most honest thing in the sentiment block: the street genuinely does not know.

Today's move

IBM closed 2026-08-04 at $235.15, up 3.91% or $8.84 from $226.31. It opened at $224.76 — BELOW the previous close — traded a $223.30 to $235.28 range, and closed within 0.06% of the day's high on 7.21M shares. That is a 5.4% intraday reversal from low to close, which is a strong-close pattern. No company-specific news is in this file for 2026-08-04; the last company event was the 2026-07-22 release, thirteen days earlier. The session was a broad technology up-day.

The honest read. A 3.9% gain that opened lower and closed at the high, in a name 28.6% off its peak with RSI at 69, is a short-covering or bottom-fishing pattern. It is not confirmation of anything. The 50-day average sits 10.6% overhead at $260.11 and the 200-day 14.8% overhead at $270.01; the stock has to travel a long way before the chart says the downtrend is over.

What we are therefore telling the reader. Buying today means paying a 3.9% one-day move for a 6.3% base-case return, thirteen days after a revenue miss, seventy-eight days before the print that decides whether the guidance holds. The setup that changes this is a price near $200 — roughly 15x FY2027E — where our $250 base is +25% and the bear case is only 12% away, or an October print that puts revenue growth back above 4%.

9. Insiders

All eight transactions in the file belong to one officer, Anne Robinson (Senior Vice President), filed 2026-07-02 for transactions dated 2026-07-01, at a reference price of $286.725 where priced.

The pattern is a standard annual vest-and-withhold: M-Exempt acquisitions of 5,883 and 1,412 shares (units converting to common stock), followed by F-InKind dispositions of 781 and 3,254 shares (shares retained by the issuer to satisfy the tax liability on the vest), with further M-Exempt dispositions completing the mechanics. Holdings after the sequence remain above 5,884 shares.

F-InKind shares are withheld by the company and never reach the market — they are not sales. The file therefore contains zero open-market purchases and zero open-market sales, from one officer only. That is a neutral reading, and the more informative fact is what is absent: with the stock 28.6% below its high and 14% off the low, not a single insider in this file bought a share on the open market. We report that as an absence, not as a signal.

Note the reference price of $286.725 on 2026-07-01 against $235.15 today — the stock has fallen 18.0% in the thirty-four days since.

10. Verdict, kill-criteria and flip conditions

Watch.

What is genuinely strong: $11.575B of free cash flow on $1.617B of capital expenditure, a 17.1% free-cash-flow margin and a 6.58% free-cash-flow yield; a Software segment at 44% of revenue growing 5.1% at a 32.2% margin with Red Hat, watsonx, HashiCorp and Confluent inside it; $20.4B of deferred income on the balance sheet; Distributed Infrastructure up 37% with a ~$500M Power and Storage backlog; a dividend paid every year since 1916 and just declared at $1.69; a 0.705 beta; and the deepest analyst and knowledge-base coverage in this batch, including the batch's most balanced independent claim.

What we are declining to pay for: 0.6% segment revenue growth and a 1.7% revenue miss in the June quarter; IBM Z down 42% and Consulting up 0.2%; a chief financial officer describing "revenue headwinds late in the second quarter"; corrected net debt of $53.8B at 3.10x EBITDA with 6.4x interest coverage; a balance sheet 58.2% composed of goodwill and intangibles after $10.5B of first-half acquisitions; a diluted share count that has RISEN 4.4% since 2022; a FY2025 earnings base flattered by a negative tax provision; and a base fair value of $250 that is 6.3% above spot and 2.3% below the street's $255.92.

The distinction that matters. IBM is not broken and it is not cheap. It is a 4%-growth business at 19x forward with a real free-cash-flow yield, a real balance-sheet constraint, and an unresolved question about whether the June quarter was a stumble or a trend. We have no independent basis for answering that question ahead of the 2026-10-21 print, our fair value sits below the street, and the knowledge base — for once genuinely deep on a name — resolves to "mixed" rather than to a direction. That is a Watch by construction.

Pre-registered KILL criteria — what would take this to Avoid:

• Third-quarter revenue below approximately $16.5B (+1% or less) on 2026-10-21, or the full-year 4-5% constant-currency guide being withdrawn or lowered. That is the deceleration confirmed as a trend.
• Software segment growth below 4%. It is the only segment growing and 44% of the company; below 4% there is nothing carrying the business.
• Free cash flow guidance withdrawn, or full-year free cash flow tracking below $11.6B. The first half was flat and the entire +$1B increase is back-half loaded.
• A further debt-funded acquisition above $5B at 3.10x net leverage with 6.4x interest coverage.
• Consulting revenue turning negative for two consecutive quarters — that is the AI-substitution thesis becoming visible in the numbers rather than in the commentary.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price at or below approximately $200 (roughly 15x FY2027E, 14x FY2028E) with the fundamental picture unchanged. This is the most likely path to a Buy and it is 15% below today's close, inside the range this stock has traversed in a fortnight. At $200 the base case is +25% and the bear case is 12% away.
• The 2026-10-21 print with revenue at or above $17.0B (+4%) and Software above +5%, which would kill the deceleration thesis and validate the guidance in a single number.
• An IBM Z product announcement starting a new mainframe cycle off a −42% trough.
• A reclaim of the 50-day average at $260.11 together with one of the above — on its own this would be a technical flip only.
• The COBOL-modernisation threat visibly not materialising in the installed base — the knowledge base has both the bear claim and the best available rebuttal of it, and the resolution is an empirical matter observable in Infrastructure Support and Software revenue.

Where IBM fits in the Synthos Framework Portfolio. The enterprise-software-and-services sleeve, at 0% today with a 1.5-2% target on a fill near $200 or a confirmed October re-acceleration. On batch overlap: IBM's nearest neighbour in this batch is nothing — it is the only enterprise IT name here. Its closest economic cousin is ADI or QCOM as a mature technology franchise trading on a cycle rather than a slope, and among those three IBM is the cheapest on free cash flow and the slowest-growing. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $235.15.

Single biggest risk: the revenue re-acceleration does not arrive. The June quarter grew 1.1%; consensus wants 3.1% in September, 4.3% for the full year, and 3.9% and 5.1% in the two years after. The whole valuation architecture — 19.1x forward, 8% consensus EPS growth, the +$1B free-cash-flow guide — rests on the second half doing what the second quarter did not. IBM Z is 42% below prior-year, Consulting is flat, and Software cannot carry a $67B revenue base on its own at 5% growth. At 19x with 3.10x leverage there is meaningful de-rating available: a return to 14x, the multiple IBM carried for most of the decade before 2024, is the $177 bear case.

Most fragile assumption in the price: that the second half accelerates. Every other assumption is a margin or cash assumption, and management has delivered on both — non-GAAP pre-tax margin expanded 0.3 points in a quarter with 1.1% revenue growth, which is genuinely creditable execution. This one is a demand assumption made against a quarter that just missed, in a company whose own chief financial officer used the phrase "revenue headwinds," and against the two segments (Infrastructure and Consulting) that together are 55% of revenue and grew −7.4% and +0.2%. If the second half runs at 2% rather than 4%, FY2027 operating EPS is nearer $12.50 than $13.16, and at 17.5x that is $219 — seven percent below today's price.

Provenance & disclosures

• Traceability: 32 raw knowledge-base hits (23 entity, 9 text); 11 used; 21 discarded. Discard reasons, itemised: 10 for METAPHOR — IBM used as a historical analogy with no claim about the security ("the IBM/Microsoft of crypto", "since IBM System/360 (1964)", "like IBM in the 60s/70s", "more than 1960s IBM/AT&T monopolies", "IBM-vs-HP", "OS markets consolidate (IBM, Windows, Android)", and four others); 4 for PRODUCT-LEVEL claims naming IBM software rather than the equity (two 2026 GPU data-processing claims naming IBM Watsonx, a 2017 voice-assistant claim naming IBM Watson, a 2025 tokenisation claim); 1 DUPLICATE (two identical 2025-01-31 claims from one channel); 1 STALE BASKET (2021-05-27, eight-name global-value rotation). Held out separately: 3 claims dated 2023-08-07 that are unmistakably IBM management voice — describing IBM's own hybrid-cloud, AI and quantum strategy — but are filed under a channel slug named for the chief executive of an unrelated pharmaceutical company. They are excluded from the conviction pool and quoted at half weight, and the mis-filing is reported as an attribution defect of the same family as the CRM and LLY cases. Of the 11 used, 2 carry a named independent speaker (Jordi Visser, skill 2.0, on 2025-03-02 bullish and 2026-04-04 neutral; Jack Farley, independent, on 2021-07-26 bearish) and 9 are channel-attributed with no named speaker — the weakest sourcing the 4-lane policy admits. No promoter-class voice appears. Net conviction mixed, breadth 11. Searches: IBM, International Business Machines, Red Hat, Arvind Krishna, watsonx. All quotes are verbatim from the stored claim text.
• Data as-of: fundamentals — annual statements to 2025-12-31 from the vendor payload; quarterly income statement to 2026-06-30; most recent filing-verified financials 2026-06-30, from the 10-Q filed 2026-07-23 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873603 = 2026-08-04T20:00:03Z ($235.15, +3.91%; 50-DMA $260.11; 200-DMA $270.01; RSI 69.07; MACD −10.31) · knowledge-base claims 2026-08-04. Fiscal note: IBM's fiscal year is the calendar year. All figures come from the Synthos vendor data file for IBM or from the SEC filings in the IBM archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-24 (fiscal 2025); 10-Q filed 2026-04-23 (March 2026 quarter); 10-Q filed 2026-07-23 (June 2026 quarter); 8-K filed 2026-06-23; 8-K filed 2026-07-14; 8-K filed 2026-07-22 (second-quarter results, including Exhibit 99.1 with the full release). All six carry preserved financial tables (tables: true), which is why the balance-sheet and segment work in this dive is filing-sourced rather than vendor-sourced. The one gap: the extracted 10-K covers the risk factors and exhibits but not the income-tax note, so the FY2025 negative tax provision can be sized from the statements but not attributed to a disclosed driver — stated rather than guessed.
• Where the filings contradicted or corrected the vendor (detailed in Section 7): total debt restated from the vendor's $67,154M to the 10-Q's $61,260M at 2025-12-31 and $61,987M at 2026-06-30 — a $5,894M vendor overstatement whose mechanism is $3,347M of operating leases counted as debt plus $2,547M of those same leases counted a SECOND time in capitalLeaseObligations; net debt rebuilt from $53,513M to $53,809M at the correct date, with the vendor's figure shown to be 14.4% too high for the date it actually carries; enterprise value rebuilt from $279.643B to $275.441B and every EV multiple recomputed (EV/EBITDA 17.24x → ~15.9x; net debt/EBITDA 3.582x → ~3.10x); the balance sheet identified as predating the $11.5B Confluent acquisition completed 2026-03-17, with goodwill ($67,717M → $74,599M), intangibles ($11,391M → $13,955M), cash ($14,471M → $8,178M) and equity ($32,740M → $34,541M) all taken from the 10-Q; the FY2025 negative tax provision of −$242M identified and sized at roughly $1.3B of Q4 GAAP earnings, with the Q4 GAAP-above-non-GAAP inversion ($5.88 vs $4.52) as the evidence and the 2026 positive provisions ($313M and $484M) as the reversal; est.ebitdaAvg/ebitAvg rejected outright on a fixed-margin fabrication signature (20.45% and 12.23% of revenue in all eight years, against an actual FY2025 EBITDA margin of 25.6%); trailing EBITDA recomputed from the vendor's implied $16.22B to $17.364B; seg_prod relabelled from "Infrastructure Services" to the 10-Q's Infrastructure, with all pre-2021 history in both segment blocks discarded as non-comparable (including a 2014 seg_geo line of $192.945B against a sub-$93B revenue year and a field literally named "J [P]"); and lastDividend of $6.73 identified as the TTM total against the 8-K's declared $1.69 quarterly rate, a $6.76 forward run-rate. Where vendor and filing AGREED — worth recording: the share count (10-Q's 2,296M issued less 1,354M treasury = 942M, matching the implied market-cap count to 0.01%); the segment revenue tie ($67.472B vs reported $67.536B, 0.09%); the geographic tie ($67.535B vs $67.536B); FY2025 gross margin (vendor 58.2%, 8-K's Q2 2026 GAAP figure 57.7%); and the quarterly revenue and EPS series against the earnings calendar.
• Basis note: analyst estimates in est.epsAvg are operating (non-GAAP) EPS — verified because the four earn_cal actuals to 2026-06-30 sum to $12.01 against GAAP diluted EPS of $11.30, and because the 10-Q's own reconciliation reports Q2 2026 diluted EPS of $2.27 GAAP against $2.93 operating (non-GAAP), matching earn_cal exactly. The GAAP-to-non-GAAP gap is 6.3% on the trailing year but inverted and 30% in the fourth quarter of 2025 because of the tax item. Forward multiples use consensus operating non-GAAP EPS, and the trailing multiple is shown on both bases (19.6x operating, 20.8x GAAP).
• Estimate coverage: 13 analysts on FY2026 EPS and 17 on revenue; 14 and 18 on FY2027; only 6 and 15 on FY2028; 4 and 11 on FY2029; 2 and 7 on FY2030. FY2029 and FY2030 are excluded entirely as non-monotonic and thin. The FY2028 EPS row, used for all three fair-value anchors, rests on 6 analysts and that is disclosed at every point of use.
• Peer note: the vendor peer set is not usable in aggregate — it pairs Micron ($1.008T) with Magic Software ($853M) and Information Services Group ($210M). The defensible comparables are Accenture ($104.3B), SAP ($227.8B), Cisco ($479.8B) and Salesforce ($156.4B). No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.
• Fair-value caveat: the $177 / $250 / $347 anchors are multiples of the FY2028 consensus operating-EPS distribution — 14x the low of $12.625, 17.5x the mean of $14.290, and 21x the high of $16.544, on 6 analysts. Stated arithmetic, not a discounted cash flow. The base is sensitivity-disclosed: 16x gives $229, below spot; 19x gives $272. The whole answer lives in a 16-19x band. Both tails are independently corroborated by the published target distribution — our bear of $177 sits 1.1% above the street's $175 low, and our bull of $347 sits 0.9% below the street's $350 high — which is coincidence rather than confirmation but is worth recording.
• Timing: second-quarter 2026 results were released 2026-07-22, thirteen days before this dive, with revenue missing consensus by 1.7% ($17.162B against $17.464B) and operating EPS landing exactly in line at $2.93. The next print is 2026-10-21, 78 days away, and it is the referee for the full-year guidance. 2026-08-04 was a broad technology up-day; IBM's +3.91% carried no company-specific news in this file.
• 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.