Strip out Confluent and IBM's software business grew zero percent last quarter. The stock has already bounced 18% off the low on a recovery story that is half true. Here is what actually happens next.
Since the 22 July print, the story on IBM has consolidated into a clean narrative: mainframe weakness is cyclical, software is the engine, the 25% single-day crash was an overreaction, and at ~$236 you are getting a 7% free-cash-flow yield on a software-led compounder. The stock has obliged, running from a 52-week low of $199.19 on 23 July to $235.68 on Friday — +18.3% in a month.
Most of that narrative is defensible. One load-bearing piece of it is not.
From IBM's own second-quarter release, the segment lines:
| Segment | Q2'26 rev | Q2'25 rev | Y/Y | Segment profit | Margin |
|---|---|---|---|---|---|
| Software | $7,761M | $7,387M | +5.1% | $2,502M | 32.2% |
| Consulting | $5,327M | $5,314M | +0.2% | $647M | 12.1% |
| Infrastructure | $3,835M | $4,142M | −7.4% | $835M | 21.8% |
| Financing | $186M | $166M | +12.2% | $108M | 58.0% |
| Total | $17,162M | $16,977M | +1.1% | — | — |
Software +5.1% is the pillar. But SiliconANGLE's coverage of the earnings call states it plainly: “Excluding acquisitions, software growth was flat.”
The arithmetic is consistent. Software added $374M year over year. IBM closed the $11B Confluent acquisition on 17 March 2026, meaning Q2'26 was the first full quarter of contribution and Q2'25 contains none of it. Confluent alone plausibly accounts for essentially the whole delta. The “Data” sub-line grew 18% at constant currency — that is where Confluent lands.
Two things in the quarter were genuinely strong and are being under-weighted:
IBM Z revenue fell 42%. CFO Jim Kavanaugh's framing, per SiliconANGLE: the “mainframe stack of hardware and transaction processing software impacted IBM's growth by over five points in the quarter. We were only expecting about a point or two.” CEO Arvind Krishna: “A lot of the demand is deferred, not destroyed,” with roughly one-third of the slipped deals already closed in the first weeks of Q3.
I believe the deferral explanation. The mechanism is verifiable and it is not an IBM story at all — it is a memory-price story. Per a Bernstein note cited by CRN, Q2 conventional DRAM contract pricing rose 74% q/q to ~$17.60/GB, with NAND up ~60% q/q to ~$0.25/GB. Enterprises pulled capital forward into servers, storage and memory to beat further increases. Mainframes lost the budget fight to DRAM procurement. Kavanaugh's supporting stat — z17 revenue across the first five quarters is still ~30% ahead of the comparable z16 period — is the right rebuttal and it holds up.
Transaction Processing fell 9% at constant currency. This matters more than the hardware line. IBM prices mainframe software by capacity: no MIPS growth, no software bill growth. TPP is among the highest-margin revenue in the company, and software gross margin duly compressed 130bp to 82.6%. If MIPS installation is genuinely deferred rather than destroyed, TPP recovers with a lag of one to two quarters. If it isn't, the margin damage compounds. Kavanaugh's counter is specific and checkable: nearly 50% of z17 customers are buying the Spyre AI accelerator, and watsonx Code Assistant for Z customers are growing MIPS three times faster than non-adopters. I rate that credible and it is the single most important disclosure to re-verify on 21 October.
Revenue flat, +1% cc. Signings +6%, with generative AI accounting for about half of quarterly signings and more than 30% of the consulting backlog. Annual recurring software revenue $24.6B, +8%. Consulting segment margin expanded 150bp to 12.1%. This segment is not the problem; it is quietly the most improved line in the company.
IBM's most-repeated commitment is free cash flow up ~$1B year over year. FY2025 free cash flow was $14.7B, so the target is roughly $15.7B. Management reaffirmed it on 22 July even while cutting revenue guidance from “5%+” to “4–5%” constant currency. The market treated the reaffirmation as the reassuring part of the release. Run the halves:
| Period | Free cash flow | Y/Y change |
|---|---|---|
| H1 2025 | $4,808M | — |
| H1 2026 | $4,760M | −$48M (flat/down) |
| H2 2025 (derived) | ~$9,890M | — |
| H2 2026 required | ~$10,940M | +$1,050M |
H1 figures from IBM's GAAP-to-FCF reconciliation; H2 derived from FY2025 FCF of $14.7B.
Macro linkage worth naming: an A-rated issuer paying ~4.75% for eight-year paper, with the 10Y at 4.734% and the 30Y at 5.273%, is the same story as hyperscaler IG issuance running >2x last year's pace. Corporate long-duration demand is competing directly with the Treasury for capital. IBM is not a bystander in that; it is a participant, and every 25bp of term-premium widening from here costs it real money on refinancing.
From the cash flow statement: common stock repurchases for the first half were $465M — and the line item is explicitly “repurchases for tax withholdings.” Not a discretionary buyback. Meanwhile diluted share count went up, from 948.0M to 953.3M year over year, and proceeds from share issuance were $418M. IBM diluted holders ~0.6% over the year.
The stock fell from above $306 on 7 July to a 52-week low of $199.19 on 23 July. If management believed that price was wrong, the response was available and free. They spent the money on Confluent instead and let the count drift higher. That is a revealed preference, and it argues against anyone claiming insiders think the stock is mispriced to the upside.
The Form 4 record for the last 90 days contains director stock awards at $281.21 on 30 June and tax-withholding dispositions by an officer at $286.725 on 1 July — both non-discretionary. There is not one open-market purchase after the crash. Insiders buying the low is the classic confirming signal for a capitulation bottom. It is absent here.
IBM declared a quarterly dividend of $1.69. That is $6.76 annualised. At $235.68 the yield is 2.87%. Some data feeds (including the one I pulled first) are printing 4.28%. Anyone screening IBM as a 4%-plus yielder is buying a 2.87% yielder. Correct the input.
The coverage picture is genuinely fine but worth stating precisely: $6.76 × ~940M shares ≈ $6.35B of dividends against $15.7B of guided FCF — a 40% FCF payout, comfortable. Against GAAP earnings it is a different picture: H1 dividends paid $3,166M versus H1 GAAP net income $3,381M, a 94% GAAP payout ratio. The dividend is covered by cash flow, not by accounting earnings. That is fine for now and it is a fragility if the FCF guide slips.
H1 GAAP EPS $3.55; operating EPS $4.84. The bridge is $1.11 of acquisition-related adjustments and $0.18 of retirement-related, versus $0.92 and $0.05 a year ago. 27% of “operating” EPS is add-backs, up from 22%. The more IBM acquires, the wider the gap between the number it reports and the number it earns. R&D also rose 10.8% in H1 ($4,485M vs $4,047M) — faster than revenue — which is partly the ~$2B/year quantum commitment. Genuine long-duration optionality; a genuine near-term margin drag.
IBM committed >$10B to quantum over five years and remains on schedule for the first large-scale fault-tolerant machine, Starling, in 2029. The delivery cadence has been genuine, not promotional:
My view: this is the most credible quantum roadmap in the industry and it is worth approximately nothing to a 2026 or 2027 earnings model. Treat it as a free call option attached to a cash-generative business, not as a valuation input. The correct way to express a quantum view is not IBM — a $222B market cap dilutes it to irrelevance. If quantum is your thesis, own the pure-plays and accept the risk. If cash flow is your thesis, own IBM and treat 2029 as upside you did not pay for. Separately, IBM selling a chunk of its global blockchain patent portfolio to Circle on 21 August is a small but telling data point: management is actively monetising IP that does not serve the AI/quantum/hybrid-cloud core.
| Metric | Value | Note |
|---|---|---|
| Price (21 Aug close) | $235.68 | +0.85%; range $233.12–$238.72 |
| 52-week range | $199.19 – $332.46 | Low 23 Jul '26; high 2 Jun '26 |
| Off the high / off the low | −29.1% / +18.3% | — |
| Market cap | ~$222B | ~940M shares basic |
| P/E TTM | ~20.9x | EPS TTM $11.27 |
| Forward P/E | ~18.4x | — |
| FCF yield (guided $15.7B) | 7.07% | vs 10Y at 4.734% → +234bp |
| EV / FCF | ~17.6x | EV ~$276B |
| Dividend yield | 2.87% | $1.69/qtr; 40% of FCF |
| Beta | 0.71 | Genuinely defensive |
| Next catalyst | Q3 print 21 Oct, AMC | Cons. EPS $2.91, rev ~$16.98B |
The sell-side, per Investor's Business Daily's post-warning survey: Oppenheimer downgraded to Perform and withdrew its price target entirely; BofA cut to $280 from $330 but stayed Buy; UBS reiterated Neutral at $236 and cut its 2026 revenue growth estimate from 5.5% to 3.6%; BNP is Underperform; Barclays stayed Overweight on Red Hat.
The dollar is an underpriced tailwind. IBM guided 4–5% constant-currency growth with currency “neutral to growth” at the FX rates prevailing on 22 July. DXY closed Friday at 98.82, a three-month low, down 0.9% on the week. With roughly half of IBM's revenue non-US, every 1% of sustained dollar weakness is worth on the order of 50bp to reported revenue — roughly $340M annualised. If the dollar stays here or lower into year-end, the July FX assumption becomes conservative and reported growth prints above constant-currency growth. Note this contradicts some secondary coverage claiming FX would drag reported growth “closer to zero”; IBM's own release says neutral. Trust the release.
The 18.3% bounce has already paid you for the Q3 deferred-deal catch-up. Above $240 you are buying the Q4 guide, and the Q4 guide has to clear a +21% Infrastructure comparison with enterprise hardware budgets still being consumed by DRAM at $17.60/GB. Two independent signals support this: the comp arithmetic, and the H2-loaded FCF requirement ($10.94B needed versus $9.89B delivered a year ago, off an H1 that was down $48M). Confirming bear trigger: a Q4 Infrastructure guide implying a y/y decline worse than 10%.
Beta 0.71, a 2.87% dividend at a 40% FCF payout, an unbroken payout record since 1916, and a 7% FCF yield are not things you sell into a rate-volatile, geopolitically noisy tape. Selling here also crystallises the July damage at the worst point of the information cycle. The correct action is not to sell but to own the 21 October event asymmetry deliberately: October puts around the $215 strike cost far less than the 25% single-day move IBM has already demonstrated it is capable of producing. IBM has now gapped down on three consecutive reporting events this year (Q1 beat → −6%; the 14 July pre-announcement → −25%; 22 July → further weakness). The market is not pricing a fourth. It should.
I move [IBM-1] to HIGH CONVICTION on either: (a) a weekly close above $256 on the Q3 print, or (b) IBM explicitly disclosing ex-acquisition software growth above 3% on the 21 October call. Not on price alone. The second condition is the one that actually matters — and if management declines to give the organic number when asked, treat the refusal as the answer.
Nothing here is investment advice. Positions can and do change. Every level above is derived from Friday 21 August 2026's close and the primary filings linked above.