IBM

01 — Executive summaryRevenue of $17.2 billion. Free cash flow of $15.7 billion. Only one of those is growing.

IBM reported revenue of $17.2 billion for the quarter ended 30 June 2026, an increase of 1% year over year. Software grew 5%. Consulting was flat. Infrastructure fell 7%, and within it the IBM Z mainframe fell 42%. GAAP diluted earnings per share were $2.27, down 2%; operating earnings per share were $2.93, up 5%. The company left its full-year free cash flow guidance intact at an increase of about $1 billion, which puts the figure near $15.7 billion, and it narrowed revenue growth guidance from "more than 5%" to "4% to 5%" at constant currency.

Those numbers are internally consistent and they describe two different companies. One is a $67.5 billion revenue base that has compounded at roughly 4% for five years, carries $62 billion of debt, employs 286,800 people, and converts about 22% of revenue into free cash flow. The other is the business that produced a 12.2% revenue growth quarter in the final three months of 2025 and a 1.1% growth quarter six months later.

The second company is the one the market was buying. IBM's forward multiple went from 12.8× at the end of FY2021 to 24.5× at the end of FY2025 — a re-rating of nearly twelve turns — as the market concluded that IBM was an artificial-intelligence beneficiary with a software mix that would keep improving. On 14 July 2026 that conclusion was tested. IBM published selected preliminary second-quarter results a week ahead of schedule, disclosed that they fell short, and the shares fell 25.2% in a single session on 67.4 million shares — the sharpest single-day decline in the company's history. The stock has recovered about half of that and sits at $237.49.

This report is about what the fall was actually telling investors, and about whether the recovery is justified. Our answer, compressed: IBM's problem in the quarter was not demand, it was allocation — and the distinction matters more than management's own framing admits. IBM attributed the miss to clients shifting capital spending "toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases". That is an accurate description of what happened and it is not entirely reassuring. A deferral thesis assumes the budget comes back. A crowding-out thesis assumes some of it does not. The evidence in IBM's own segment data supports both readings, and the difference between them is worth about $150 per share.

Six findings follow from the primary record. Each is developed in full later; each is stated here with the number that supports it.

1. Software is the company, and it decelerated three quarters in a row

Software produced $29.96 billion of revenue in FY2025 — 44.4% of the total — and $2.50 billion of segment profit in Q2 2026 at a 32.2% margin, which is 61% of total segment profit. It grew 14% in Q4 2025, 11% in Q1 2026 and 5% in Q2 2026. The deceleration is not broad: Data software grew 19% and Hybrid Cloud, which is mostly Red Hat, accelerated sequentially to 11%. The drag is Transaction Processing, an $8.60 billion annual line that fell 8% because it is sold against mainframe capacity that clients deferred. Three of four sub-segments compounded. The one that did not is the largest.

2. Consulting is 31% of revenue and has not grown in two years

Consulting revenue was $20.69 billion in FY2024 and $21.06 billion in FY2025 — 1.8% reported growth, flat in constant currency — and it was flat again in Q2 2026. IBM has repeatedly highlighted a cumulative generative AI book of business, which reached more than $12.5 billion at the Q4 2025 report. It has not published an updated figure since. The segment that books the AI work is the segment that is not growing, and the metric that was supposed to demonstrate the opposite has quietly left the disclosure set.

3. Free cash flow is the one line that has never broken — and it is increasingly bought

Free cash flow rose from $9.3 billion in FY2022 to $14.7 billion in FY2025, a 16.5% compound rate against 3.7% revenue growth, and the FY2026 guide is an increase of about $1 billion. On a $223.9 billion market capitalisation that is a 7.0% free cash flow yield, which is the strongest part of the case and is not in dispute. What is in dispute is how much of it is organic. IBM spent $10.93 billion on acquisitions in the twelve months to June 2026 and $8.29 billion in FY2025, and net debt rose $13.3 billion in six quarters while cash and investments halved to $8.1 billion. A meaningful share of reported growth is purchased, and the purchase was financed.

4. The AI capital cycle is a headwind for IBM, not a tailwind

This is the finding we consider most under-discussed. IBM's customers are the same enterprises buying accelerators, servers, storage and memory. When that market is supply-constrained and prices are rising, the marginal enterprise dollar goes to securing hardware, and it comes out of discretionary software and mainframe budgets. In Q2 2026 IBM's Distributed Infrastructure line — Power and Storage — had its best quarter in reported history, up 37%, with a $500 million backlog. The lines that require a discretionary commitment fell. IBM is not losing to a competitor; it is being crowded out by a budget line that is not a competitor. The same mechanism that makes the compute-stack series a coherent trade makes IBM its first casualty.

5. Nineteen times earnings is a fair blend, not an opportunity

At $237.49 IBM trades at 19.3× consensus FY2026 operating earnings of $12.32, 21.1× trailing GAAP earnings, and 14.3× guided free cash flow. IBM is 44% Software, 31% Consulting and 23% Infrastructure by revenue. Software peers trade at an average of 27.3×; IT services peers trade at 10.8× to 14.5×. A weighted blend lands within a turn of where IBM actually trades. The multiple is neither a value opportunity nor a bubble; it is an accurate summary of a conglomerate with one good business and two mature ones.

6. The options are real, expensive and far away

IBM has committed more than $10 billion to quantum computing over five years, is standing up Anderon — a pure-play 300mm quantum wafer foundry in Albany backed by $1 billion of IBM cash and a $1 billion CHIPS award finalised on 16 September 2026 — and says it remains on track to deliver the first large-scale fault-tolerant quantum computer by 2029. It has also committed $5 billion and 20,000 engineers to Project Lightwell, an open-source security clearinghouse built in response to the discovery by Anthropic's Claude Mythos model of nearly 3,900 high-severity vulnerabilities in open-source software. Both are genuine, both address markets that did not exist eighteen months ago, and both consume cash now for revenue that is, at the earliest, a 2029 story.

Our conclusion is Neutral, with a 12-month target of $248 — 4.4% above the 16 September close and within a few dollars of the $245.35 consensus compiled by S&P Global Market Intelligence. A $168 bear case, a $248 base case and a $322 bull case, weighted 25/50/25, produce $246.50. The bear case is that the crowding-out is structural and the multiple compresses to the services peer set. The bull case is that the deferred mainframe refresh returns in 2027 with the force of the programme-to-programme ratio that IBM keeps citing. We cannot yet tell which is right, and the Q3 print on 21 October is the first place the answer will show up.

What we can say is that a 7.0% free cash flow yield, a 2.85% dividend that has been paid since 1916, and a 19.3× forward multiple on low-single-digit organic growth describe an income instrument with a technology business attached. That is a defensible thing to own. It is not a re-rating story any more, and the report that follows explains why in detail.

02 — Company & business modelFour segments, one installed base, and a consulting arm that is not growing

International Business Machines was incorporated in 1911 as the Computing-Tabulating-Recording Company and renamed in 1924. It is the oldest large technology company in the world and, after four decades of reinvention, it is not obviously any one thing. The company reports four segments, employs 286,800 people, operates in more than 175 countries, and generated $67.54 billion of revenue in FY2025. Its market capitalisation of $223.9 billion is roughly the same as it was in 2013.

Software

Software is $29.96 billion of revenue, 44.4% of the total, and it is organised into four lines. Hybrid Cloud, at $7.33 billion in FY2025, is mostly Red Hat — the $34 billion acquisition of 2019 that remains the single most consequential capital allocation decision of the Krishna era. Automation at $7.73 billion includes the Apptio portfolio and the Ansible and Terraform estates, the latter arriving with HashiCorp. Data at $6.30 billion spans watsonx.data, Db2 and, since March 2026, Confluent's real-time streaming platform. Transaction Processing at $8.60 billion is CICS, IMS, Db2 for z/OS and the rest of the mainframe software stack — the highest-margin revenue IBM has, and the most tightly coupled to hardware shipments.

Consulting

Consulting is $21.06 billion, 31.2% of revenue, split between Strategy and Technology at $11.54 billion and Intelligent Operations at $9.52 billion. It is the segment that most resembles Accenture, and it carries a 12.1% segment margin against Accenture's low-teens operating margin. It grew 1.8% in FY2025 and was flat in the first two quarters of 2026.

Infrastructure

Infrastructure is $15.72 billion, 23.3% of revenue, and it is two businesses wearing one label. Hybrid Infrastructure at $10.62 billion is IBM Z mainframes, Power servers and storage — cyclical, hardware-led, and the reason IBM's revenue chart has steps rather than a slope. Infrastructure Support at $5.10 billion is the maintenance annuity that sits behind the installed base: flat to down for three years, and the cash engine that funds the rest.

Financing

Financing is $737 million of revenue and a 58.0% segment margin — the smallest segment and the most revealing, because it exists to make the hardware sellable. It carries its own debt, $13.0 billion at 30 June 2026, which IBM reports separately and which anyone modelling the balance sheet has to handle deliberately.

The shape of the company is therefore not complicated. Software carries 61% of segment profit on 44% of revenue, Consulting and Infrastructure are mature businesses that generate cash and employ people, and Financing is a captive lender. The investment case reduces to a single question: can Software keep growing fast enough to offset two flat businesses and to justify a multiple that sits well above the services peer set? Every other question in this report is downstream of that one.

03 — The financial recordFive years, and the year the top line stopped

IBM's revenue history is best read as a series of plateaus rather than a trend. FY2021 was $57.35 billion, the first full year after the Kyndryl separation removed roughly $19 billion of annual revenue. FY2022 was $60.53 billion, up 5.5%, assisted by currency. FY2023 was $61.86 billion, up 2.2%. FY2024 was $62.75 billion, up 1.4% — the weakest full year of the period. FY2025 was $67.54 billion, up 7.6%. Consensus for FY2026 is $70.38 billion, up 4.2%.

The FY2025 acceleration is the number that re-rated the stock, and it deserves to be examined rather than accepted. Two things were happening. The first was the z17 mainframe cycle, which launched in April 2025 and drove Hybrid Infrastructure from $8.91 billion to $10.62 billion. The second was the HashiCorp acquisition, which closed in February 2025 and contributed roughly $700 million of incremental revenue on an annualised basis, mostly inside the Automation line. Remove the mainframe cycle and the acquisition and the underlying business grew closer to 2%. None of this makes the FY2025 result false. It does mean that a 7.6% year is not a run rate.

The quarterly series is where the story turns. IBM grew 0.6% in Q1 2025, 7.7% in Q2, 9.1% in Q3 and 12.2% in Q4 — a genuinely accelerating curve, and the basis on which management guided to more than 5% constant-currency growth for 2026 in April. Then Q1 2026 came in at 9.5% and Q2 2026 at 1.1%. In two quarters the growth rate went from the best of the cycle to the worst.

Free cash flow is IBM's own definition, which excludes the change in IBM Financing receivables. Operating EPS for FY2021 and FY2022 is not presented because IBM's non-GAAP presentation changed with the Kyndryl separation. Net debt is total debt less cash, restricted cash and marketable securities on a standardised filing basis. Sources: IBM annual reports; S&P Global Market Intelligence.

Two lines in that table matter more than the others. The first is operating margin, which went from 9.7% to 18.7% in five years. IBM has run a productivity programme that management describes as delivering more than $4.5 billion of cumulative savings, and the effect is visible: in Q2 2026, with revenue growing 1%, operating pre-tax margin still expanded 30 basis points to 19.2% and operating EPS still grew 5%. A company that can grow earnings at 5% on 1% revenue growth is managing its cost base well, and the discipline is real.

The second is GAAP EPS, which has never behaved. It was $6.35 in FY2021, $1.79 in FY2022 — a collapse caused by $5.6 billion of pension-related and divestiture charges — $8.14 in FY2023, $6.43 in FY2024 on $3.1 billion of pension settlement charges, and $11.17 in FY2025, a figure that itself contains a tax benefit from the resolution of audit matters. There is no five-year GAAP earnings trend at IBM. There is a five-year series of operating earnings with non-operating items bolted on, and the non-operating items are large enough to change the sign of the year-over-year comparison in three of the last five years.

One further feature of the record is worth flagging before the segment work. IBM's share count is barely moving. Diluted weighted-average shares were 952.1 million in Q1 2026, against 942.1 million outstanding. IBM bought back $1.05 billion of stock in the last twelve months against $6.31 billion of dividends. A company that neither grows revenue quickly nor shrinks its share count depends entirely on margin expansion and on the multiple. Margin expansion has been delivered. The multiple has been taken away.

04 — Segment deep diveSoftware: three of four lines compound, and the largest one just stopped

Software is the report's central subject and it is worth stating the arithmetic plainly. The segment generated $29.96 billion of revenue in FY2025, up 11% reported and 9% at constant currency, and $2.50 billion of segment profit in Q2 2026 at a 32.2% margin. On a segment-profit basis that is 61% of the company. On a revenue basis it is 44%. Every incremental dollar of Software revenue is worth roughly two and a half incremental dollars of Consulting revenue in profit terms, which is why the mix shift matters more than the total.

Read over three years, the sub-segments divide cleanly. Hybrid Cloud — Red Hat OpenShift, Red Hat Enterprise Linux, and the associated subscriptions — went from $5.83 billion in FY2023 to $7.33 billion in FY2025, growth of 25.8%. It is the only part of IBM that behaves like a modern infrastructure software franchise: high gross margin, renewing subscription revenue, and a competitive position that is genuinely defensible because OpenShift is the default enterprise Kubernetes distribution.

Automation went from $6.01 billion to $7.73 billion, growth of 28.6%, the fastest of the four. The driver is a portfolio assembled largely by acquisition — Apptio for technology business management, HashiCorp for infrastructure-as-code and secrets management, Instana and Turbonomic for observability and resource optimisation. Automation is where IBM's "cost-takeout" pitch lands, and it is the line most exposed to the argument that agentic AI will eventually compress the market for IT operations tooling.

Data went from $5.46 billion to $6.30 billion, growth of 15.4%, and it is the line that changed most in 2026 with the Confluent acquisition. Confluent is the commercial company behind Apache Kafka and it is the standard for real-time event streaming in the enterprise. The strategic logic is that AI systems need a real-time data layer, that IBM already owned the transactional and analytical layers through Db2 and watsonx.data, and that Kafka filled the gap. Data grew 19% in both Q1 and Q2 2026, making it the most consistently fast line in the company.

Then there is Transaction Processing: $8.60 billion in FY2025, up 11.5% over two years, the largest sub-segment, and the one that broke.

The question that determines the valuation is whether the Q2 deceleration was cyclical or structural, and the evidence is genuinely mixed.

For the cyclical reading: IBM says the z17 programme is running at "nearly 130 percent program-to-program" against z16, which was previously the strongest mainframe programme in the company's history, and that clients representing 85% of installed MIPS are maintaining or growing capacity. A programme-to-programme ratio above 100% five quarters into a launch means the cumulative installed base is still larger than the prior cycle's. Deferred purchases in a supply-constrained hardware market are the classic pattern of a quarter that moves rather than disappears. Red Hat accelerated sequentially to 11%, which is not what a demand problem looks like.

For the structural reading: the deferral was not caused by anything about IBM, which means IBM has no control over when it ends. Clients moved budget to servers, storage and memory because those were supply-constrained and prices were rising. If that constraint persists — and the industry's own capacity plans suggest it will persist through 2027 — then the crowding-out persists with it. Transaction Processing is a mature, high-margin, licence-and-maintenance business tied to a hardware platform whose long-run trajectory is flat. A quarter that falls 8% in a growing software market is a data point about direction, not just about timing.

Our own reading sits between the two. We think the mainframe cycle is deferred rather than broken — the programme-to-programme ratio is a hard number and it is hard to fake — but we also think the market has been slow to price the possibility that the deferral is long enough to matter for a company whose largest software line is attached to it. A deferral of two quarters is a buying opportunity. A deferral of six quarters is a change in the earnings base. Nothing in the Q2 disclosure distinguishes between them, and that is precisely why the Q3 print on 21 October 2026 is the most important date in this report.

FY2023–FY2025 revenue in $bn. Growth rates for Q1 and Q2 2026 are as reported, i.e. including currency; constant-currency growth is 1–3 percentage points lower. Source: IBM segment disclosures.

05 — Segment deep diveConsulting: $21 billion of revenue that has not moved in two years

Consulting is the largest employer inside IBM, the second-largest segment by revenue, and the one that has failed most comprehensively to deliver what was promised for it. Revenue was $20.69 billion in FY2024 and $21.06 billion in FY2025. That is 1.8% reported growth and effectively nothing at constant currency. In Q1 2026 the segment grew 4% reported and 1% at constant currency. In Q2 2026 it grew 0%.

The sub-segments tell a slightly different story from the total, and it is worth being precise. Strategy and Technology — the higher-value work: cloud migrations, AI strategy, application modernisation — went from $11.49 billion to $11.54 billion, growth of 0.4%. Intelligent Operations — the run-and-maintain business, closer to outsourcing — went from $9.20 billion to $9.52 billion, growth of 3.5%. The faster-growing half is the lower-margin half. That is the opposite of the mix shift IBM needs.

IBM's answer to this is the generative AI book of business. The company began disclosing a cumulative figure in 2024 and grew it steadily: about $2 billion by mid-2024, around $5 billion at the FY2024 report, more than $6 billion in Q1 2025, more than $7.5 billion in Q2 2025, and more than $12.5 billion at the FY2025 report. Then it stopped.

There is a real question underneath all of this about what happens to systems-integration work in an agentic-AI world, and IBM is on both sides of it. On one side, IBM Consulting is a beneficiary if enterprises need help deploying AI systems, and IBM's own tools — Bob, watsonx Orchestrate — are designed to make that deployment faster, which raises the value per consultant. On the other side, the same tools are designed to reduce the number of people required per engagement, and IBM has said explicitly that it is "accelerating productivity by scaling software development leveraging AI, increasing the effectiveness of our sales and marketing organization, and optimizing our supply chain". A services business whose unit of sale is a person-hour cannot welcome a technology whose purpose is to reduce person-hours unless it can raise price faster than it reduces volume. IBM has not yet demonstrated that it can.

FY2023–FY2025 Consulting revenue is the sum of the Strategy and Technology and Intelligent Operations sub-segment disclosures. Segment profit for FY2023–FY2025 is not separately disclosed in the quarterly releases cited here. Source: IBM quarterly and annual segment disclosures.

The one genuinely encouraging number in the segment is the margin: 12.1% in Q2 2026 against 10.6% in Q1 2026 and roughly 10% two years ago. Consulting is not growing, but it is becoming more profitable, and the improvement is coming from mix and delivery productivity rather than from price. A flat $21 billion business at a rising margin is worth more than a growing one at a falling margin. It is just not worth a growth multiple.

06 — Segment deep diveInfrastructure: the strongest mainframe launch in IBM's history, and the quarter it wrapped

Infrastructure is where the quarter went wrong, and understanding why requires understanding what a mainframe cycle is. IBM launches a new generation of IBM Z roughly every three to four years. Because the hardware is expensive, because the software stack is priced against capacity, and because enterprises plan the migration years in advance, revenue arrives in a sharp spike in the launch quarter followed by a long tail of annuity. The pattern is so regular that IBM reports a "program-to-program" ratio — cumulative shipments of the current generation against the same point in the prior generation — as a forward indicator.

The z17 launched on 8 April 2025 and IBM Z revenue grew 51% in the launch quarter. Through its first full year, Hybrid Infrastructure — the line that contains IBM Z along with Power and Storage — grew from $8.91 billion to $10.62 billion. Then the comparisons got hard. IBM Z grew 6% in Q1 2026 as the launch was lapped, and in Q2 2026 it fell 42%. Infrastructure as a whole fell 7.4% to $3.84 billion, and Hybrid Infrastructure fell 10%.

Management's account, from the 14 July letter to investors, is specific and worth quoting at length. When IBM set expectations in April it said it would be "wrapping on the launch of z17 in the second quarter", that this had been "the strongest start to a mainframe program in our history", and that it therefore expected Infrastructure revenue to decline low single digits for the year beginning in that quarter. What actually happened was worse. In the last weeks of June, clients "shift[ed] their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases". IBM also cites clients being "distracted with rapidly-evolving, industry-wide cybersecurity concerns in the quarter" — a reference to the open-source vulnerability crisis that Project Lightwell was built to address, and which we cover in section 16.

Then the admission: "These conditions require our teams to execute perfectly, and this quarter we faltered. We did not adapt and move quickly enough, and numerous large deals failed to close on the timelines we expected, driving the majority of our shortfall."

The genuinely positive item in the segment is Distributed Infrastructure — Power servers and Storage — which grew 37%, the best performance in reported history, and exited the quarter with a backlog of approximately $500 million. That is the crowding-out mechanism operating in IBM's favour for once: when enterprises rush to secure compute and storage capacity ahead of price increases, Power and Storage are what they buy. The same dynamic that cost IBM its mainframe quarter gave it its best distributed-systems quarter. A reader looking for evidence that this is a reallocation rather than a demand collapse should start here.

Infrastructure Support — the maintenance annuity — declined 1% to $5.07 billion. It has been between $5.07 billion and $5.38 billion for four years. This is the line that pays for IBM's research and development, and it is slowly shrinking as the installed base ages. It is not a crisis. It is also not growth.

FY2023–FY2025 sub-segment revenue in $bn. IBM Z and Distributed Infrastructure growth are company-disclosed quarterly rates. Source: IBM quarterly and annual segment disclosures.

Note the margin: 21.8% in Q2 2026 against 15.8% in Q1 2026, on lower revenue. That is what happens when a high-margin hardware line collapses and a lower-margin but faster-growing one takes its place, and it is why the segment's 7% revenue decline did not produce a 7% profit decline. It also means the segment's profit quality deteriorated while its reported profit held up.

07 — The July warningHow $61 billion of market value disappeared in one session

The most consequential event in IBM's recent history as a public company is not a product launch or an acquisition. It is a press release issued before the market opened on Tuesday 14 July 2026.

IBM's second-quarter results were scheduled for 22 July. On 14 July the company published "Arvind Krishna's Letter to IBM Investors" — a document that disclosed selected preliminary second-quarter figures, a week ahead of schedule, with an explicit warning that "we are still working to close our financial reporting for the quarter and our final results could be slightly different". The shares closed at $217.07, down 25.2%, on 67.4 million shares. According to CNBC, it was the sharpest single-day decline in the company's history.

The final results, published on 22 July, were within rounding of the preliminary figures. Revenue of $17.16 billion against a consensus of $17.58 billion. Adjusted EPS of $2.93 against $2.97 expected. Full-year constant-currency revenue growth guidance cut from more than 5% to 4–5%. Free cash flow guidance unchanged. The stock rose 0.4% on the day.

Three things about this sequence deserve scrutiny, because each of them is a signal about how to read the company.

The pre-announcement itself

Pre-announcing a miss is a legitimate and, in most circumstances, a shareholder-friendly act. It gives the market accurate information earlier. But it is also a choice, and IBM made it six business days before the scheduled release and while the quarter was still open. The letter's language — "our teams to execute perfectly, and this quarter we faltered", "these are not excuses, but they are realities" — is unusually direct for a CEO letter. Management's willingness to take the hit in public rather than manage the narrative over a week is to its credit. It also establishes that management knew the numbers were bad enough that the normal process would not contain them.

The guidance cut was small, and that is the point

IBM cut full-year constant-currency growth guidance from "more than 5%" to "4% to 5%". That is a cut of perhaps one percentage point of revenue, or roughly $700 million. The market capitalisation fell by more than $60 billion. The market was not repricing one point of revenue; it was repricing the multiple it had assigned to IBM over the preceding four years. A stock that had been re-rated from 12.8× to 24.5× forward earnings on the strength of an accelerating growth story cannot survive the story decelerating. The 25% move is a statement about how much of the FY2024–FY2025 re-rating was narrative rather than arithmetic.

The litigation overhang

Within days of the warning, multiple securities law firms announced investigations and, by 19 July, a securities class action. The investigations focus on IBM's 2026 growth guidance and the subsequent reduction. The outcome of these matters is unknowable and this report assigns them no probability-weighted cost. What can be said is that the existence of a disclosed litigation overhang is one of the reasons the shares have not fully recovered: a stock trading below its 100-day moving average with an unresolved disclosure question attached carries a discount that a stock without one does not.

The recovery since has been real but incomplete. The shares bottomed at a close of $206.65 on 23 July and an intraday low of $199.19 — the 52-week low — and recovered to $248.37 by 15 September, which is still 14.4% below the pre-warning close of $290.23. They then gave back 4.4% on 16 September, the day Anderon's $1 billion CHIPS award was finalised and the Federal Reserve raised rates 25 basis points to 3.75–4.00%. The 52-week range is $199.19 to $332.46.

08 — The crowding-out mechanismWhy the AI capital cycle is a headwind for the company selling to it

The Farstar compute-stack series has covered NVIDIA as the supply side of the AI capital cycle, Amazon and Alphabet as the demand side, and Alphabet additionally as a participant that builds its own silicon. IBM is the fourth position in that sequence and it is a different kind of position: the incumbent whose customers are the buyers. IBM does not sell accelerators, does not build data centres at scale, and does not run a frontier model lab. It sells software, services and mainframes to the enterprises that are doing all three. And that is exactly why the boom is a problem.

Consider what a chief information officer faced in June 2026. Memory and server prices were rising because AI training and inference demand had absorbed supply. Storage was constrained. Any hardware bought in the second half of 2026 would cost more than hardware bought in the first half. Meanwhile a mainframe refresh could reasonably be deferred by a quarter, a licence renewal could be pushed into the next fiscal year, and a consulting engagement could be rescoped. When a budget is fixed and the price of one input is rising, the rational response is to pull that input forward and push everything else back.

That is precisely what IBM described, and precisely what the segment data shows. Distributed Infrastructure, which is Power and Storage — the products an enterprise buys when it is securing supply-constrained capacity — grew 37%. Data software, which includes Confluent and HashiCorp, grew 19%. Everything that required a discretionary decision fell or stalled: IBM Z down 42%, Hybrid Infrastructure down 10%, Transaction Processing down 8%, Infrastructure Support down 1%, Consulting flat.

There is a third consideration that cuts against the tidy version of the crowding-out story, and intellectual honesty requires stating it. If the mechanism were purely a reallocation of enterprise hardware budgets, IBM's Power and Storage lines would have benefited and nothing else would have changed. That is broadly what happened — except that IBM also said "numerous large deals failed to close on the timelines we expected, driving the majority of our shortfall". The word "majority" matters. It means that most of the shortfall came from execution, not from budgets. The crowding-out mechanism is real and it is visible in the data. It is not the whole explanation, and a reader who adopts it as the whole explanation is giving IBM more credit than IBM gave itself.

Where does that leave the mechanism as an investment thesis? We think it is the correct frame for understanding what happened, and a weak instrument for predicting what happens next, because it is silent on duration. What we can say with more confidence is that the mechanism is structural in one direction: as long as AI infrastructure demand absorbs enterprise budgets, IBM's discretionary lines face a headwind that has nothing to do with IBM's competitive position. A company with 44% of revenue in software and 31% in services, both of which are discretionary purchases, is more exposed to that headwind than its "boring legacy technology" reputation suggests. That is a genuine reason to apply a discount to IBM's multiple relative to the software peer set — and it is a reason that does not go away when the memory cycle does.

09 — Quality of earningsThe four adjustments between GAAP and operating, and what each one is worth

IBM reports two earnings figures and reconciles them in every release. In Q2 2026 the gap was $0.66 per share on a base of $2.27 GAAP, which is 29% of reported earnings. That is a large gap by the standards of large-cap technology and it has been growing.

The two adjustments are acquisition-related and retirement-related, and they behave very differently.

Acquisition-related adjustments were $0.58 per share in Q2 2026, up from $0.47 a year earlier. They consist principally of amortisation of acquired intangible assets — the customer relationships, developed technology and trademarks that came with Red Hat, Apptio, HashiCorp and Confluent — plus transaction costs and retention arrangements. The economic substance is straightforward. IBM paid cash for these assets, and the amortisation is the accounting recognition of that cash having been spent. A shareholder who excludes it entirely is pretending the purchase was free. A shareholder who treats it as a cash cost in the quarter is double-counting, because the cash already left the building on the closing date.

The right treatment is to capitalise the acquisition spend and treat amortisation as the return of that capital, which is what a discounted cash flow does naturally. The wrong treatment — and the one that flatters IBM most — is to value the company on operating EPS while ignoring that operating EPS is growing partly because IBM keeps buying earnings. The acquisition adjustment grew 23% year over year in Q2 2026. If acquisitions stop, the adjustment stops growing and reported growth slows. If acquisitions continue, the adjustment keeps growing and the balance sheet keeps absorbing the cost. There is no version in which the adjustment is free.

Retirement-related adjustments were $0.08 per share in Q2 2026, against $0.02 a year earlier. These are the amortisation of prior service costs, interest cost, the expected return on plan assets, and amortised actuarial gains and losses. This is the least defensible of the two exclusions, because IBM's pension plans are large, real, and — as section 12 shows — partly underfunded. IBM's FY2025 total retirement-related net periodic cost was $1.07 billion. That is a real annual economic cost of having promised benefits, and excluding it from "operating" earnings is a presentational choice rather than an economic one.

Source: IBM Q2 2026 press release, U.S. GAAP to operating (non-GAAP) results reconciliation.

Three further items sit outside the reconciliation and inside GAAP, and they matter for anyone comparing IBM's GAAP earnings across years. In FY2024 the company recognised $3.1 billion of pension settlement charges, which is why GAAP EPS fell from $8.14 to $6.43 while operating EPS rose from $9.60 to $10.32. In FY2025 GAAP EPS of $11.17 benefited from a tax benefit on the resolution of audit matters. And in FY2022 GAAP EPS of $1.79 reflected $5.6 billion of divestiture and pension-related charges. Across five years, GAAP EPS has moved in the opposite direction to operating EPS twice.

The practical conclusion is not that IBM is manipulating its earnings — it is not, and the reconciliation is disclosed in full. The conclusion is that the GAAP series is not usable as a valuation input and the operating series is not usable without an adjustment for capitalised acquisition spend. A reader who values IBM at 19.3× operating earnings is valuing it at a multiple that excludes the cost of the strategy that produces those earnings. A reader who values it at 21.1× trailing GAAP earnings is using a number that is dominated by items IBM does not control. The honest multiple is somewhere between, and it is higher than 19.3×.

10 — Free cash flow$14.7 billion, and the acquisition that is hiding organic growth

IBM guides free cash flow, not revenue, and it has never missed the guide. That is the strongest single fact in the company's favour and it deserves to be stated first. FY2022 was $9.3 billion, FY2023 $11.2 billion, FY2024 $12.7 billion, and FY2025 $14.7 billion — an increase of $2.0 billion year over year and a 16.5% compound rate against 3.7% revenue growth. The FY2026 guide is an increase of about $1 billion, which puts the figure near $15.7 billion, and the company reiterated it in July even as it cut revenue growth guidance.

On a $223.9 billion market capitalisation, $15.7 billion is a 7.0% free cash flow yield. For comparison, the dividend costs $6.76 per share, or about $6.4 billion — roughly 41% of guided free cash flow. The dividend is covered with room to spare, has been paid every quarter since 1916, and has been raised for 31 consecutive years. Whatever else is true about IBM, the cash generation and the distribution of it are not in question.

The definition, and why it matters

IBM's free cash flow is net cash from operating activities excluding the change in IBM Financing receivables, less capital expenditure net of proceeds. The exclusion is defensible — Financing is a captive lender whose receivables movements are financing activity dressed as operating activity — but it is a choice, and it is a favourable one. On the standardised presentation used by S&P Global Market Intelligence, which does not make the exclusion, FY2025 free cash flow was $12.1 billion rather than $14.7 billion, and the twelve-month figure to June 2026 was $13.79 billion against IBM's reported H1 figure of $4.76 billion for the half alone.

Both numbers are correct. They answer different questions. IBM's definition answers "how much cash did the parent company generate for shareholders", which is the right question for a dividend investor. The standardised definition answers "how much cash did the consolidated enterprise generate", which is the right question for an enterprise-value calculation. This report uses IBM's definition where discussing the dividend and the guide, and the standardised definition where discussing the cash flow statement, and labels each.

The cash flow waterfall for the twelve months to June 2026 is where the strategy becomes legible. On the standardised basis IBM generated $14.89 billion of operating cash flow, spent $1.10 billion on capital expenditure, $10.93 billion on acquisitions, $6.31 billion on dividends and $1.05 billion on buybacks, and raised $2.62 billion of net debt. Cash and investments fell $4.81 billion.

Read that list again and note the ordering. Acquisitions consumed 1.7 times the dividend. Capital expenditure — the investment in IBM's own future products — consumed $1.10 billion, or a tenth of what acquisitions consumed and a sixth of what the dividend consumed. IBM is not a company reinvesting in itself at scale. It is a company distributing cash to shareholders and buying other companies' products with the remainder, financing the difference with debt.

How much of the growth is organic?

This is the question the cash flow statement raises and the income statement obscures. In the twelve months to June 2026 IBM spent $10.93 billion on acquisitions. Confluent, which closed on 17 March 2026 at an $11 billion enterprise value, contributed to Data software for roughly three and a half months of the period and will contribute a full year in FY2027. HashiCorp, which closed in February 2025 at $6.4 billion, contributed a full year to FY2025 and to the twelve months to June 2026.

Precise organic growth is not disclosed — IBM does not publish a constant-portfolio revenue series — but the direction is not in doubt. If Confluent contributes roughly $1.0–1.1 billion of annualised revenue and HashiCorp roughly $0.6–0.7 billion, then acquisitions contribute approximately 1.2 percentage points of the 4.2% consensus growth for FY2026 and roughly 1.0 percentage point of the 7.6% reported growth in FY2025. Those are Farstar estimates, not company disclosures, and they are labelled as such. On that arithmetic, organic growth has been running at roughly 2.5–3.5% rather than the headline 4–8%.

That is not a scandal. Buying growth at a reasonable price is a legitimate strategy for a company whose own markets grow slowly, and Red Hat has been an excellent acquisition by any measure. It does mean that the "accelerating software company" narrative of 2025 was partly an artefact of the acquisition calendar, and that a reader who wants to know how IBM is actually performing has to look through it.

11 — Balance sheet & financing$62 billion of debt, $8.1 billion of cash, and a lender inside the company

IBM's balance sheet is the part of the story that gets the least attention and carries the most risk, because it is where the acquisition strategy is paid for.

On the standardised presentation, total debt was $65.27 billion at 30 June 2026 against cash, restricted cash and marketable securities of $8.13 billion — net debt of $57.14 billion. On IBM's own presentation, which reports IBM Financing debt separately, total debt at the same date was $62.0 billion including $13.0 billion of Financing debt, up $0.7 billion year to date.

The trajectory is what matters. Net debt was $47.9 billion at the end of FY2021, fell to $43.8 billion by FY2024, and has since risen to $50.2 billion at FY2025 and $57.1 billion at 30 June 2026 — an increase of $13.3 billion in six quarters. Over the same six quarters, cash and investments fell from $14.59 billion to $8.13 billion. IBM has funded the Confluent acquisition, the HashiCorp acquisition, and the dividend by spending down cash and issuing debt, in roughly that order.

FY2023 and FY2024 balance sheet totals are as reported by IBM; total debt and net debt are on the standardised filing basis. Net debt / adjusted EBITDA uses FY2025 adjusted EBITDA of $19.2 billion for the FY2025 column and an annualised H1 2026 adjusted EBITDA for the Q2 2026 column; the Q2 2026 ratio is a Farstar calculation. Source: IBM balance sheets; S&P Global Market Intelligence.

The leverage ratios look stable — debt to equity has actually fallen from 2.66× to 1.97× — and that stability is misleading in an instructive way. Equity is growing because IBM is retaining earnings, not because debt is falling. Net debt to adjusted EBITDA has gone from 2.5× to 2.9× in six quarters and would be higher if the FY2026 EBITDA were calculated on a 1% revenue quarter rather than on the full year. IBM is not in financial distress. It is steadily trading balance sheet capacity for revenue growth, and it has now done so for six consecutive quarters.

One further point on the asset side. Goodwill and acquired intangibles have grown with every acquisition and now represent a material share of total assets. Goodwill is not amortised under US GAAP and is tested for impairment annually. An impairment would not affect cash flow and would not affect operating earnings, but it would reduce book equity — which, at $33.06 billion against a $223.9 billion market capitalisation, is already the smallest part of the valuation. IBM trades at 6.8× book value. For a company with $14.7 billion of free cash flow and a $33 billion book, book value is not a useful valuation anchor in either direction.

12 — Pensions & retirement$9.8 billion of underfunding behind a 116%-funded headline

IBM is one of the largest private pension sponsors in the United States and has been since the middle of the twentieth century. Its retirement obligations are, in aggregate, larger than its equity. They are also the single most commonly misread part of its disclosure, because IBM reports a favourable headline and the detail behind it is less favourable.

At 31 December 2025, IBM's qualified defined-benefit pension plans worldwide were 116% funded. The US Qualified PPP — the main plan — was 137% funded, with plan assets of $18.07 billion against a benefit obligation of $13.21 billion, a surplus of $4.87 billion. That is a genuinely strong position and it reflects two decades of deliberate de-risking, including the 2024 settlement transactions that removed $6.2 billion of obligations from the US plan.

Now the rest of it. Offsetting the $7.54 billion of surplus across all overfunded plans is $9.83 billion of deficit across underfunded plans. That deficit is concentrated in two places: $4.54 billion of non-US non-qualified plans, which are unfunded by construction, and $1.27 billion of non-US qualified plans where assets fall short of obligations. There is also $1.25 billion of unfunded US non-qualified obligations. Separately from pensions, IBM carries $2.77 billion of non-pension postretirement obligations — primarily the US retiree medical plan, which is $2.22 billion unfunded with $5 million of assets against it.

Net of everything, the balance sheet at 31 December 2025 carried a $2.28 billion deficit on retirement plans: $3.61 billion of net prepaid US pension assets, less $3.14 billion of net non-US pension liabilities, less $2.22 billion of US non-pension liabilities, less $0.54 billion of non-US non-pension liabilities.

Three consequences follow for the equity.

First, the cash cost is modest but persistent. IBM expects no legally required contribution to its US defined-benefit plans in 2026 and approximately $100 million to non-US plans, mostly in Japan and Spain. Expected benefit payments from defined-benefit plans in 2026 are $3.71 billion, plus $0.31 billion of non-pension postretirement payments, but those are paid largely from trust assets rather than from corporate cash. Actual total retirement-related cash contributions and direct payments were $1.32 billion in 2025. That is the number to model.

Second, the earnings charge is real and is excluded from operating earnings. IBM's FY2025 total retirement-related net periodic cost was $1.07 billion, of which $477 million was defined-benefit pension cost and $461 million was defined-contribution cost. In FY2024 the equivalent figure was $4.48 billion because of $3.1 billion of settlement charges. The settlement charges are genuinely one-off. The $1.07 billion baseline is not, and it is added back in the operating earnings that IBM reports and that the market values it on.

Third, interest rates cut both ways, and the direction has changed. Higher discount rates reduce the present value of obligations and therefore improve funded status. The Federal Reserve raised rates 25 basis points to 3.75–4.00% on 16 September 2026 and signalled that further tightening may follow. That is, on the face of it, good for IBM's pension deficit. It is simultaneously bad for the multiple applied to a low-growth, dividend-paying equity, which is one of the reasons the shares fell 4.4% on the day. For a company in IBM's position the two effects partially cancel, and the net is a genuine question rather than a talking point.

13 — Capital returnsA 2.85% dividend, and a buyback that has never mattered

IBM's capital return policy is one of the most consistent in the market and one of the most old-fashioned. It pays a dividend every quarter, has paid one every quarter since 1916, and has raised it for 31 consecutive years. The current rate is $1.69 per share quarterly, or $6.76 annualised, which at $237.49 is a 2.85% yield. The increase in 2026 was $0.01 per share, or 0.6%, which is the same increment as in each of the four preceding years.

IBM also buys back stock, but so little that the buyback is better understood as a rounding item than as a capital allocation policy. Over the last twelve months the company repurchased $1.05 billion of shares against $6.31 billion of dividends. In FY2021 the buyback was $0.32 billion. It has grown every year — $0.41 billion, $0.40 billion, $0.65 billion, $1.02 billion — but from a base so small that even at the current rate it retires roughly 0.5% of the share count annually. Diluted weighted-average shares were 952.1 million in Q1 2026 and 942.1 million are outstanding.

The total shareholder return picture is therefore straightforward arithmetic. A 2.85% dividend plus low-single-digit earnings growth plus or minus a multiple change. If the multiple holds at 19.3× and operating EPS grows at the consensus 6.8% in FY2027, the total return is roughly 9–10% before any multiple change. If the multiple compresses one turn to 18.3×, the total return falls to roughly 4%. If it expands one turn, it rises to roughly 15%. The dividend provides a floor and the multiple provides the variance. That is the shape of the investment, and it is the shape of a bond proxy with an equity option attached rather than the shape of a growth stock.

Payout is dividends paid divided by IBM's own free cash flow definition. The TTM payout ratio uses the FY2026 guided free cash flow of approximately $15.7 billion as the denominator; it is a Farstar calculation. Source: IBM cash flow statements; S&P Global Market Intelligence.

14 — AcquisitionsHashiCorp, Confluent, and the price of buying growth

IBM has spent $33.2 billion of cash on acquisitions since the beginning of FY2021. That is the defining capital allocation decision of the Krishna era and it deserves to be assessed on its own terms rather than absorbed into the growth rate.

The pattern is a steady $2–5 billion per year through FY2024, then a step change: $8.29 billion in FY2025 and $10.93 billion in the twelve months to June 2026. The step change is two transactions. HashiCorp, announced in December 2024 and closed in February 2025 at approximately $6.4 billion, brought Terraform (infrastructure as code), Vault (secrets management) and Consul into the Automation line, and is the single largest reason Automation grew 28.6% over two years. Confluent, announced on 8 December 2025 at $31.00 per share in cash for an enterprise value of approximately $11 billion, closed on 17 March 2026 and brought the commercial Kafka platform into the Data line.

Both are strategically coherent. Red Hat gave IBM the enterprise Linux and Kubernetes franchise; HashiCorp gives it the provisioning and secrets layer that sits underneath; Confluent gives it the real-time event stream that AI systems consume. Taken together, IBM now owns a substantial part of the open-source enterprise data and infrastructure stack — Linux, Kubernetes, Ansible, Terraform, Vault, Kafka, Java, Cassandra — which is the same portfolio that made Project Lightwell possible in a matter of weeks. The strategy has an internal logic that is more persuasive than the sum of its press releases suggests.

There is a fourth consideration that is harder to quantify and worth naming. An organisation that integrates two large acquisitions in two years while simultaneously launching a mainframe generation, standing up a quantum foundry, committing $5 billion to an open-source security clearinghouse, and cutting costs through AI-driven productivity has a lot of change in flight at once. IBM's own account of the Q2 miss includes the sentence "we did not adapt and move quickly enough". That is a statement about organisational bandwidth as much as about markets, and bandwidth is finite. The acquisition strategy has been executed well on the closing side and it has a cost on the operating side that does not show up in any line of the income statement.

Consideration for Confluent is enterprise value at $31.00 per share in cash. The Anderon figure is IBM's cash contribution plus the CHIPS and Science Act award, finalised 16 September 2026. Sources: company announcements.

15 — Quantum$10 billion, a foundry in Albany, and a 2029 deadline IBM has put in writing

IBM has been working on quantum computing since 1999 and has been criticised for most of that period for over-promising. The current disclosure set is the most concrete the company has ever published, and it is worth taking seriously on that basis alone.

Three commitments define the position. First, IBM will invest more than $10 billion in quantum over the next five years, spanning research and development, capital expenditure, manufacturing scale-up, M&A and ecosystem expansion. Second, it remains on track to deliver the first large-scale fault-tolerant quantum computer — Starling — by 2029, having previously committed to demonstrating quantum advantage by the end of 2026. Third, it has stood up Anderon, a standalone IBM company operating a 300mm pure-play quantum wafer foundry in Albany, New York, backed by $1 billion of IBM cash and a $1 billion award under the CHIPS and Science Act that was finalised with the US Department of Commerce on 16 September 2026.

Anderon is the part of this that is new, and it is the part that changes the character of the commitment. Until 2026 IBM's quantum programme was a research programme with a roadmap. Anderon turns it into a manufacturing business with customers. The company describes itself as "the world's most advanced pure-play quantum foundry", producing specialised wafers supporting superconducting qubit arrays, quantum input/output signalling and readout signal chain components, with the stated intention to expand to other quantum modalities. The first wafers are already running through the facility — an operational milestone that matters more than the award, because it means the fab is producing rather than being planned.

The bear case on quantum at IBM is not that the technology will fail. It is that the timing is unknowable, that the revenue is years away, and that IBM is now spending at a rate that matters — $10 billion over five years is roughly 3% of IBM's cumulative guided free cash flow over the same period, before the $1 billion Anderon cash contribution. In a company already trading balance sheet capacity for growth, a $2 billion-a-year commitment to a pre-revenue technology is a real cost against a speculative return.

The bull case is that IBM is the only large technology company with a vertically integrated quantum position — process technology, hardware, software, an error-correction architecture and now a foundry — and that if fault tolerance arrives in 2029, IBM will own the manufacturing layer that the industry depends on. That is a genuinely asymmetric position. It is also a position whose value cannot be estimated with any confidence today, which is why this report assigns it a small explicit option value in the bull case and nothing at all in the base case.

What we can say with confidence is that the market has at various points in 2025 and 2026 given IBM credit for quantum optionality and has no way of pricing it. A reader who buys IBM for the quantum story is buying a call option with an unknown strike and an unknown expiry, financed by a business whose free cash flow yield is 7%. That is not an unreasonable thing to do. It is a different thing from buying a cash-generating compounder, and the two should not be confused.

16 — Lightwell & the Mythos shockA market that did not exist in April, and IBM was first into it

The most interesting thing that happened to IBM in 2026 was not in its own financial statements. It was a discovery by a competitor's artificial intelligence model.

In February 2026, Anthropic began using an early preview of a model called Claude Mythos to search for security vulnerabilities in software. By May, Anthropic was reporting that the model had identified nearly 3,900 high- or critical-severity vulnerabilities in open-source software. The company published a coordinated vulnerability disclosure programme and, through an initiative called Project Glasswing, began exploring how frontier AI systems change the economics of vulnerability discovery and exploitation. The UK's GCHQ and the IMF both issued warnings about the implications, with the IMF singling out financial services.

IBM's response, announced with Red Hat on 28 May 2026, was Project Lightwell: a $5 billion commitment, more than 20,000 engineers, and what the company describes as a "trusted enterprise clearinghouse" for open-source software security. The clearinghouse is a coordination layer. Enterprises with commercial subscriptions report vulnerabilities, receive validated production-ready patches, and coordinate responsible disclosure with upstream open-source communities. Lightwell published more than 7,500 patches in its first two weeks, reached general availability on 8 July 2026, and named early collaborators including Bank of America, BNY, Citi, Goldman Sachs, JPMorganChase, Mastercard, Morgan Stanley, Royal Bank of Canada, State Street, Visa and Wells Fargo.

There is a less comfortable reading of the same events, and it appears in IBM's own explanation of its second-quarter miss. Management cited clients being "distracted with rapidly-evolving, industry-wide cybersecurity concerns in the quarter" as one of the reasons large deals failed to close. In other words, the same crisis that created Lightwell also cost IBM revenue in the quarter it was launched. That is not a contradiction — it is what happens when a new market emerges from a shock. The opportunity and the disruption arrive together.

How to value Lightwell is genuinely difficult. It is not a segment, it has no disclosed revenue, and its economics are bundled into Software and Consulting. Our judgement is that it is more likely to show up as improved renewal rates and pricing in Red Hat and Automation than as a new revenue line, which makes it a quiet positive rather than a growth story. A reader who wants to track it should watch Red Hat growth and Automation growth rather than looking for a Lightwell line in the segment disclosure, because there will not be one.

17 — watsonx and the agentic pivotWhat IBM actually sells when it says it sells AI

IBM's AI positioning has gone through three phases since 2023 and understanding which one the company is in matters for how its AI claims should be read.

The first phase was watsonx as a platform: a model-development and governance stack, launched in 2023, aimed at enterprises that wanted to build on foundation models without sending data to a public cloud. The Granite family of models — IBM's own, smaller and cheaper than frontier models — sat underneath it. The commercial pitch was governance, data residency and cost, not capability. It was a coherent strategy and it produced slow revenue.

The second phase was AI embedded in the existing portfolio: automation products that use machine learning to optimise IT spend, observability products that detect anomalies, mainframe software that uses AI inference on the z17's on-chip accelerator, and security products that use AI for triage. This is where most of IBM's AI revenue actually sits, and it is why the generative AI book of business was never a clean metric — it was spread across products that existed before AI was in the name.

The third phase, which began in 2026, is agentic development. On 28 April 2026 IBM announced general availability of Bob, an AI development partner that the company describes as taking enterprises "from AI-assisted coding to production-ready software". Bob is not a code completion tool. It is positioned as an agentic development environment that coordinates multiple models and multiple agents across the software delivery lifecycle, with governance and auditability built in, and IBM says adoption has reached more than 80,000 of its own employees. The commercial logic is that enterprises will not adopt agentic coding without governance, and IBM sells governance.

One number is worth holding on to through all of this. IBM's Research and development expense was $2.17 billion in Q1 2026, against $1.95 billion a year earlier — an 11% increase on 9% revenue growth. IBM spends roughly 11% of revenue on research and development, which is a respectable figure for a company of its maturity and is the source of both the quantum position and the mainframe franchise. The question is not whether IBM innovates. It is whether the innovation reaches revenue fast enough to matter against a 4% revenue growth rate.

18 — Competition IConsulting against Accenture, and the AI-native attackers

IBM Consulting's competitive set is broader than it looks and the comparison that matters most is not the one IBM makes.

Against Accenture, IBM is the smaller, slower-growing, lower-margin competitor. Accenture has roughly $70 billion of revenue against IBM Consulting's $21 billion, has grown in the mid-single digits where IBM has grown 1.8%, and trades at 13.7× forward earnings. Against the Indian-scale providers — Infosys at 14.5×, Cognizant at 10.8× — IBM competes on proximity and on the ability to bundle software, and loses on cost. Against Deloitte, PwC and the other large private partnerships, IBM competes on technology depth and loses on partner economics.

The structural problem is not any one of these competitors. It is that the systems-integration business model is being attacked from below by AI-native tooling. The value of a systems integrator has always rested on three things: knowing how enterprise systems fit together, having people who can do the work, and being able to absorb risk on a fixed-price engagement. AI tooling compresses the second of those three directly. If an agentic development environment can produce the same application integration with a third of the people, the integrator's revenue per engagement falls unless the price per person rises.

IBM's response is to be the vendor of the tooling as well as the provider of the services, which is a hedge rather than a solution. If Bob reduces the labour required per engagement, IBM Consulting loses revenue and IBM Software gains licence revenue — but the licence revenue is smaller in absolute terms and IBM would have to sell the tool to its competitors to make up the difference. There is no scenario in which IBM Consulting grows at more than low single digits while agentic tooling is deflating the unit economics of integration work.

Peer revenue figures are approximate trailing figures from S&P Global Market Intelligence; forward multiples and market capitalisation as of 16 September 2026. IBM Consulting revenue is FY2025 as disclosed.

The competitive picture therefore supports the valuation argument in section 20 rather than contradicting it. IBM Consulting is a solid, profitable, mature services business that deserves a services multiple. It is not a business that deserves a premium to Accenture, and the fact that IBM as a whole trades at 19.3× while Accenture trades at 13.7× is a statement about Software, not about Consulting.

19 — Competition IISoftware and infrastructure against the hyperscalers and the specialists

IBM's Software segment competes in three different markets with three different competitive dynamics, and it is worth separating them because the aggregate growth rate conceals the differences.

Hybrid cloud infrastructure: a strong position in a market that is being redefined

Red Hat OpenShift is the default enterprise Kubernetes distribution and Red Hat Enterprise Linux is the reference enterprise Linux. That is a genuinely defensible position — the switching costs are high, the installed base is enormous, and the products sit underneath almost everything else in the enterprise stack. Hybrid Cloud grew 11% in Q2 2026, accelerating sequentially.

The risk is not competition from another Linux vendor; it is that the definition of the enterprise platform is moving up the stack. VMware's disruption following the Broadcom acquisition created a large migration opportunity that Red Hat has partly captured. But the hyperscalers are increasingly selling a bundled platform — compute, storage, database, Kubernetes, AI services — where the Kubernetes layer is free and the margin is captured elsewhere. A customer that runs OpenShift on AWS or Azure still pays Red Hat, but the customer that adopts EKS or AKS does not. Red Hat's growth depends on the hybrid and on-premises segment of the market continuing to be large, which is a defensible bet for regulated industries and a less defensible one everywhere else.

Automation: a good portfolio in a market that AI is about to reshape

Automation at $7.73 billion is the fastest-growing sub-segment over two years, at 28.6%, and it is largely an acquisition story — Apptio, HashiCorp, Instana, Turbonomic. The products address technology business management, infrastructure provisioning, secrets management, observability and resource optimisation. These are real enterprise needs and IBM has assembled a leading position in several of them.

The competitive threat is that this is exactly the category that agentic AI is designed to absorb. An agent that can provision infrastructure, manage secrets and optimise resource allocation does not need a separate tool for each function. IBM's answer — that agents need governance and that IBM sells the governance layer — is plausible, but it is the same answer every incumbent in the category is giving, and the category has more entrants than exits. Note also that Automation's growth decelerated from 10% in Q1 2026 to 4% in Q2 2026, the sharpest deceleration in the Software segment.

Transaction Processing: a fortress with a slow leak

CICS, IMS and Db2 for z/OS have no meaningful competitor. The mainframe transactional stack is decades old, mission-critical, and effectively impossible to replace at acceptable cost for a large bank or insurer. The margin reflects that. The growth rate reflects something else: the installed base of mainframe MIPS grows slowly, and Transaction Processing revenue is a function of it. Over two years the line grew 11.5%, which is roughly the pace of the z17 cycle, and in Q2 2026 it fell 8%. This is a business that will still exist in twenty years and will be approximately the size it is now.

The aggregate competitive conclusion is this. IBM Software's strongest asset is Red Hat, which is genuinely defensible and growing at low double digits. Its second asset, Automation, is well positioned today and structurally exposed. Its third, Transaction Processing, is a high-margin annuity with no growth. Its fourth, Data, is the one IBM has most recently and most expensively reinforced, and it is growing at 19%. A software business with one 19%-growing line, one 11%-growing line, one 4%-growing line and one declining line, selling into enterprises whose marginal dollar is going to hardware, is a business that should grow at 5–8%. It is currently growing at 5%.

20 — ValuationNineteen times forward earnings for two per cent organic growth, and why that is roughly right

IBM's share price on 16 September 2026 was $237.49, against a market capitalisation of $223.9 billion and 942.1 million shares outstanding. The 52-week range is $199.19 to $332.46. The shares are 18.8% below the high, 19.2% above the low, and 30% below where they were before the 14 July warning on a year-to-date basis through the July trough.

Consensus expectations, compiled by S&P Global Market Intelligence, are for FY2026 revenue of $70.38 billion — growth of 4.22% — and operating earnings per share of $12.32, growth of 6.31%. For FY2027 the consensus is revenue of $73.20 billion and operating EPS of $13.16. Consensus free cash flow for FY2026 is $15.48 billion, consistent with the company's own guidance of an increase of about $1 billion. Twenty-five analysts polled by S&P Global carry an average price target of $245.35; a separate compilation of 51 analysts puts the consensus at $256.69, and a third of 28 analysts at $265.16. The range across all compilations is $175 to $350.

What the multiple is made of

IBM trades at 19.3× consensus FY2026 operating earnings, 21.1× trailing GAAP earnings, and 14.3× guided free cash flow of approximately $15.7 billion. The three figures are not interchangeable and the differences between them are informative. The GAAP multiple is higher than the operating multiple because GAAP earnings are lower, and they are lower because of pension settlements, acquisition amortisation and tax items that do not recur predictably. The free cash flow multiple is the lowest of the three because IBM converts an unusually high share of its earnings into cash — 22% of revenue against the software peer group's mid-teens.

The right way to test whether 19.3× is reasonable is to ask what a blended multiple should be for this mix. IBM is 44% Software, 31% Consulting and 23% Infrastructure by revenue, and 61% Software, 16% Consulting and 20% Infrastructure by segment profit. Software peers — ServiceNow, SAP, Microsoft, Oracle — trade at an average of roughly 27× forward earnings, though the range is 17.7× to 34.3×. IT services peers trade at 10.8× to 14.5×. Hardware and infrastructure peers trade in the mid-teens.

A profit-weighted blend — 61% at 26×, 16% at 13×, 20% at 16× — produces approximately 21.5×. A revenue-weighted blend produces approximately 19.7×. IBM trades at 19.3×. The multiple is therefore almost exactly what a sum-of-the-parts calculation would produce, and the conclusion is uncomfortable for both bulls and bears: there is no obvious mispricing here.

Scenario valuation

We value IBM on FY2027 operating earnings, which is the first year in which the mainframe cycle normalises one way or the other, and we apply a multiple that reflects the mix. The three cases differ in their growth assumption and in their multiple, not in their arithmetic.

Weighted value: 0.25 × $168 + 0.50 × $248 + 0.25 × $322 = $246.50. Target rounded to $248. Scenarios are Farstar constructions based on the assumptions stated; they are not forecasts and they are not the company's guidance. FY2027 EPS for the base case is the S&P Global consensus of $13.16. Bear and bull EPS are Farstar estimates.

The weighted value of $246.50 against a $237.49 price is 3.8% of upside. Our target of $248 is 4.4% above the market and within a few dollars of the $245.35 S&P Global consensus. That is a deliberately unexciting conclusion, and it is the conclusion the evidence supports. The dispersion of analyst targets — $175 to $350, a 68% spread, with 22 of 51 analysts at Hold — is itself the tell. When a stock's fair value is genuinely contested across that range, it usually means the range is driven by a single binary question that nobody can answer yet. Here, that question is whether the second quarter of 2026 was a deferral or a repricing.

What we would need to see to change the rating

To move to Constructive we would need two consecutive quarters of Software growth at or above 8%, evidence that Transaction Processing has recovered, and a Consulting growth rate above 3%. The first is available on 21 October. To move to Cautious we would need Software growth below 4%, a further deterioration in Infrastructure Support, and any indication that the deferred mainframe deals were lost rather than delayed — which IBM would most likely signal through a further reduction in the programme-to-programme ratio, a metric the company has historically been willing to update.

21 — The bull caseSix arguments we take seriously

The bull case for IBM at $237.49 is not a story about artificial intelligence. It is a story about a cash-generating franchise that has been repriced for a quarter that will not repeat, and about a set of options that cost nothing in the current multiple.

1. The valuation has already reset, and the dividend sets a floor

The forward multiple has come down from 24.5× at the end of FY2025 to 19.3×, which is below the FY2024 year-end figure of 21.1× and close to the FY2023 year-end 16.6×. On a free cash flow basis the shares yield 7.0%. The dividend yields 2.85%, has been paid since 1916, and consumes 41% of guided free cash flow. An investor buying today is buying a 7% free cash flow yield with a covered dividend and an investment-grade balance sheet. That is a defensible entry point for an income-oriented portfolio regardless of what happens to growth.

2. The mainframe cycle is deferred, not cancelled

The strongest single number in IBM's favour is the programme-to-programme ratio of nearly 130% against z16. z16 was the strongest mainframe programme in the company's history; running 30% ahead of it five quarters into the cycle is not what a declining franchise looks like. IBM also states that clients representing 85% of installed MIPS are maintaining or growing capacity. A hardware purchase deferred because memory prices were rising is a purchase that happens when memory prices stop rising. If it happens in the first half of 2027, the FY2027 comparison becomes easy rather than hard.

3. Software's deceleration was one sub-segment, not the segment

Software grew 5% in Q2 2026, but Data grew 19% and Hybrid Cloud accelerated sequentially to 11%. Transaction Processing fell 8% for a specific, identified reason — it is sold against mainframe capacity — and Transaction Processing is the sub-segment least representative of IBM's strategic direction. The three lines that matter for the future grew 19%, 11% and 4%. A reader who strips out the mainframe-linked line finds a software business growing at roughly 11%, which is a perfectly respectable number for a $21 billion revenue base.

4. Margin discipline has been real, and it has not stopped

Operating pre-tax margin expanded 30 basis points to 19.2% in a quarter when revenue grew 1%. Operating margin has gone from 9.7% in FY2021 to 18.7% in FY2025. IBM describes more than $4.5 billion of cumulative productivity savings and is now applying AI to its own software development, sales and supply chain. Consulting margin improved from 10.6% to 12.1% sequentially. Cost discipline is the part of the IBM story that has never broken, and it means earnings grow faster than revenue in almost every scenario.

5. Project Lightwell created a market and IBM is the incumbent in it

The discovery of nearly 3,900 high-severity open-source vulnerabilities by an AI model in 2026 made software supply-chain security a board-level issue in a single quarter. IBM, through Red Hat, owns or maintains a disproportionate share of the enterprise open-source commons and had 20,000 engineers it could reassign. Lightwell shipped 7,500 patches in two weeks and signed Bank of America, Citi, Goldman Sachs, JPMorganChase, Morgan Stanley, Visa and Wells Fargo as early collaborators. Whether or not Lightwell ever produces a standalone revenue line, it makes the IBM and Red Hat stack the safest place to run enterprise software at exactly the moment when that property became valuable.

6. Quantum is an option with a stated date

IBM has committed more than $10 billion over five years, has a fault-tolerant system called Starling promised for 2029, and has now created Anderon, a pure-play 300mm quantum wafer foundry in Albany with $2 billion of committed capital and wafers already running through it. If fault tolerance arrives, IBM owns the manufacturing layer of a new industry. The option is not priced into a 19.3× multiple, and unlike most optionality in large-cap technology it comes with a specific date and a specific deliverable that can be checked.

22 — The bear caseSix arguments we cannot dismiss

The bear case for IBM at $237.49 is not that the company is failing. It is that the company has been priced for a growth rate it does not have, and that the mechanism suppressing its growth is not under its control.

1. Organic growth is roughly half the reported figure

IBM spent $10.93 billion on acquisitions in the twelve months to June 2026 and $8.29 billion in FY2025. Confluent and HashiCorp together contribute roughly 1.2 percentage points of the 4.2% consensus growth for FY2026 — a Farstar estimate, not a company disclosure, but the direction is not seriously contestable. Strip acquisitions out and IBM is a company growing revenue at 2–3%, carrying $57 billion of net debt, and trading at 19.3× forward earnings. At 2.5% organic growth, 19.3× is not cheap. It is expensive.

2. The crowding-out mechanism has no end date

The AI infrastructure build is not a two-quarter phenomenon. Memory, storage and accelerator capacity are being absorbed by hyperscale and enterprise demand on a multi-year horizon, and the industry's own capital plans imply the constraint persists through 2027 and possibly beyond. Every year of constraint is a year in which the marginal enterprise IT dollar goes to hardware rather than to IBM's discretionary software and services. This is the argument we consider most under-appreciated, and it is structural: it does not depend on IBM executing badly, and IBM cannot fix it.

3. Consulting has not grown in two years and the AI book stopped being disclosed

Consulting is 31% of revenue, 16% of segment profit, and the largest employer inside IBM. It grew 1.8% in FY2025 and was flat in both quarters of 2026. The generative AI book of business — the statistic IBM used to demonstrate that its AI strategy was working — reached more than $12.5 billion at the Q4 2025 report and has not been updated since. A company that stops disclosing its most-cited AI metric in the same year that its AI-adjacent segment stops growing has told investors something, and the market has been slow to hear it.

4. The balance sheet has been spent, and the spending continues

Net debt rose $13.3 billion in six quarters while cash and investments halved to $8.1 billion. The dividend consumes 41% of guided free cash flow. Quantum is committed at more than $10 billion over five years, Lightwell at $5 billion, Anderon at $1 billion of IBM cash on top of the CHIPS award. IBM is running a capital programme on a decelerating revenue base, and the response to a revenue miss has been to increase spending rather than to reduce it. That is a defensible strategy. It is also one with no slack in it.

5. Management's own explanation is worse than the market's

IBM said that "numerous large deals failed to close on the timelines we expected, driving the majority of our shortfall" and that "we did not adapt and move quickly enough". That is an execution diagnosis, not a demand diagnosis. Deferred deals come back; deals that a sales organisation failed to close on time may not. If a majority of the shortfall was execution, then the deferral thesis — which is the entire basis of the bull case — explains less than half of what happened.

6. The multiple is a fair blend, which means there is no upside

Software peers trade at 27×, services peers at 13×, and IBM at 19.3×. That is what the mix deserves. For the shares to work from here, either the mix has to change — Software growing fast enough to become a larger share of profit — or the whole blended multiple has to rise, which requires the market to change its mind about low-growth conglomerates. Neither is impossible. Neither is likely on a twelve-month horizon, and there is a securities class action attached to the guidance cut that could produce headlines at any time.

23 — Risk matrix & catalystsRanked by probability times impact

Catalysts

24 — ConclusionNeutral at $248, on a cash machine priced for a growth rate it does not have

IBM is a well-managed company that had a bad quarter for reasons that are partly its own fault and partly not, and the market has repriced it accordingly. The repricing has been large — 25% in a single session, 30% year-to-date at the July trough — and it has restored the shares to a valuation that is defensible on almost any reading of the business.

The case for owning IBM at $237.49 rests on three things that are not in dispute. It generates $14.7 billion of free cash flow and guides to about $15.7 billion, a 7.0% yield. It pays a 2.85% dividend that has been paid every quarter since 1916 and raised for 31 consecutive years, covered 1.4 times by free cash flow. And its largest and most profitable segment, Software, contains two sub-segments growing at 19% and 11%, with a hybrid-cloud franchise that has a genuine competitive moat. An income investor who buys IBM for those three things and accepts low single-digit growth is not making a mistake.

The case against owning it rests on three things that are also not in dispute. Organic growth is roughly half the reported figure once acquisitions are stripped out. Thirty-one per cent of revenue sits in a consulting business that has not grown in two years, at the same time as the metric IBM used to demonstrate AI traction has stopped being disclosed. And the mechanism that caused the July miss — enterprise capital being absorbed by AI hardware — is not under IBM's control and shows no sign of ending.

Our target of $248 is 4.4% above the market and effectively at consensus. We are not recommending the shares, and we are not warning against them. A 25/50/25 weighting of a $168 bear case, a $248 base case and a $322 bull case produces $246.50, which is what a genuinely unresolved binary looks like when it is priced honestly.

What would change our mind is not a narrative. It is two consecutive quarters of Software growth at or above 8% with Transaction Processing recovering, which would tell us the deferral was a deferral. The first of those two quarters is reported on 21 October 2026. Until then, the honest description of IBM is the one in the title: an incumbent with an excellent cash machine, a fortress in Red Hat, two mature businesses, and a growth rate that its customers' capital budgets are currently working against.

25 — AppendixFinancial summary tables

Annual income statement summary

Cash flow and balance sheet summary

Quarterly record, 2025–2026

Q3 2025 figures are omitted where not confirmed from the primary releases cited in this report. Q2 2025 revenue is derived from the reported 1.1% year-over-year growth rate applied to the Q2 2026 figure of $17.16 billion. Quarterly free cash flow for Q2 2025 and Q2 2026 is derived from the half-year and first-quarter disclosures. Source: IBM quarterly press releases.

Key metrics at a glance

26 — Sources & disclosurePrimary references, limitations, and revision policy

Primary sources

The financial figures in this report are drawn from the following primary sources, except where explicitly labelled as an estimate: IBM's quarterly press releases for Q3 2024 through Q2 2026, including the segment detail, the U.S. GAAP to operating reconciliation and the free cash flow reconciliation; IBM's fourth-quarter and full-year 2025 press release dated 28 January 2026; Arvind Krishna's letter to IBM investors dated 14 July 2026; IBM's FY2025 Form 10-K, including the retirement-related benefits note; IBM's Form 8-K filings relating to the preliminary second- quarter results; IBM's cash flow statements and balance sheets as standardised by S&P Global Market Intelligence; IBM's announcement of the Confluent acquisition dated 8 December 2025 and the completion announcement dated 17 March 2026; IBM's and Red Hat's announcements of Project Lightwell dated 28 May 2026 and the general availability announcement dated 8 July 2026; IBM's announcement of IBM Bob dated 28 April 2026; the Anderon announcement of the finalised $1 billion CHIPS award dated 16 September 2026; Anthropic's coordinated vulnerability disclosure reporting from 2026; daily closing prices from S&P Global Market Intelligence; and analyst consensus data from S&P Global Market Intelligence, TipRanks and MarketBeat as of 16 September 2026.

Estimates and limitations

Several figures in this report are Farstar estimates rather than company disclosures, and they are labelled as such where they appear. Specifically: the allocation of IBM's $10 billion quantum commitment across categories is an estimate; the contribution of Confluent and HashiCorp to reported growth is an estimate derived from disclosed enterprise values and publicly reported revenue run-rates; the split of Q2 2025 revenue is derived from the reported growth rate; the FY2026 GAAP earnings figure is a consensus estimate; the "IBM Consulting-only, implied" multiple in section 19 is a Farstar construction and not a traded multiple; and the scenario values in section 20 are Farstar constructions based on stated assumptions, not forecasts.

IBM does not disclose organic revenue growth on a constant-portfolio basis, so the organic growth figures in this report are inferred rather than reported. IBM does not disclose revenue by customer, so the crowding-out mechanism in section 8 is inferred from the segment data and from management's own explanation rather than measured directly. IBM does not disclose the number of mainframe customers, so the question of whether the installed base is stable or shrinking cannot be answered from public disclosure. Where a figure could not be sourced, it is marked "—" or omitted rather than estimated.