International Business Machines Corporation IBM

Key data

The business

IBM sells four things to the largest institutions on earth. Software, now its biggest and best profit engine, is Red Hat (the plumbing that lets a bank run its apps across its own data centers and any cloud), automation, data and the watsonx AI tools. Consulting is the army of people who install and run all of it. Infrastructure is the Z mainframe, the fridge-sized machine that still clears a huge share of the world's credit-card swipes and bank transactions, plus storage. Financing lends customers the money to buy the rest. The moat is switching cost in its purest form: a core banking or airline reservation system that has run on IBM for thirty years does not move without years of risk and cost, so the customer renews and pays the price increase. IBM's acquisition of Confluent, the data-streaming company, folded another recurring-revenue product into the software stack. The tension in the whole enterprise is simple: software and the mainframe refresh grow, the older distributed hardware shrinks, and the net of the two has to stay positive to justify the price.

The numbers

Revenue is grinding higher while profit has stopped rising, and that is the story of the last year in two lines.

Q2 2026 revenue rose 1.1% against Q2 2025, but net income slipped to $2.17B from $2.19B and diluted EPS fell to $2.27 from $2.31. The margin softness flagged a quarter ago held rather than healed: gross margin came in at 57.7%, down a full point year to year, which the company pins on a revenue shortfall and mix, softened by productivity actions. Consulting signings did their job, up 5.0% in the quarter and 6.8% over six months, while infrastructure kept declining. The offset is working, barely.

Revenue has compounded about 4% a year since 2021, no faster. Net income is lumpy, from $1.6B in 2022 to $10.6B in 2025, so the earnings line is the wrong anchor. The durable number is free cash, roughly $13B a year at about a 20% margin on sales, which puts the stock near 16 times free cash and a free-cash yield close to 6%. Two cautions sit under that. The 2025 profit jump to $10.6B is large enough that trailing EPS flatters the multiple, and cash on hand fell from $13.6B at year-end to $7.2B by June, drained by the Confluent deal, buybacks and the dividend against net debt already at about 3.6 times EBITDA. The variant view here is narrow: the market prices IBM as a 4%-grower worth 20 times, and it is right unless software growth visibly widens its lead over the hardware drag. The print that settles it is total revenue growth staying positive with gross margin holding above 57%.

Management

The insider record this year is one modest sale against two small buys. An officer sold $5.76M of stock on August 26 (plan status not disclosed in the Form 4 footnote), while two directors, Michael Miebach and Michelle Howard, bought a combined $113k back in February. None of that is a signal of size. Capital allocation is where management shows its hand, and the priorities are clear from the cash statement: the dividend comes first and is treated as untouchable, buybacks and a sizeable acquisition took the rest, and the balance sheet absorbed the difference as cash fell $6.4B in the half. Operating cash generation is steady and improving, up to $5.2B in the March quarter from $4.4B a year earlier, so the drawdown is a use-of-cash choice, not a generation problem. The record on hitting its own guidance has been reliable, which is most of why the multiple sits where it does.

How it fails or surprises you

Software cannot outrun the hardware drag. Infrastructure declined last quarter and total revenue grew just 1.1%. If the mainframe cycle rolls over before the next Z refresh and software slows at the same time, total revenue turns negative and gross margin breaks below 57%. At that point a 20x multiple on a shrinking business compresses fast. Watch the two lines together over the next two quarters.

The AI backlog converts faster than priced (right tail). IBM's generative-AI book has been building across software and consulting. If watsonx and Red Hat automation convert that backlog into recognized software revenue and reaccelerate the segment to high single digits, the whole company re-rates off "4%-grower." The tell is software revenue growth outpacing total revenue and RPO growth widening ahead of sales.

The 2025 earnings and the cash drain. Net income leapt 76% to $10.6B while cash fell $6.4B in six months against 3.6x leverage. If 2025 profit carried one-time help, normalized earnings sit above 20x and the balance sheet has less give than the headline suggests. The proof is H2 free-cash conversion and whether cash rebuilds toward year-end.

Closing thoughts

At 16 times free cash for a sticky, slow-growing franchise, the market has this one about right. The left tail is real but cushioned: if revenue shrinks and margins keep sliding, the multiple compresses, but the dividend and buyback give time to fix it. The right tail is narrower: if software reaccelerates to high single digits and pulls the top line to 6% or 7%, you re-rate a few multiple points, not a double. The fatter tail is probably the benign one, because the switching costs are genuine and the free cash is dependable, but the payoff for being right is measured, not transformational.

The bet is still that the world's biggest banks, insurers and governments keep running their most important computer systems on IBM, and keep paying more each year for the software that ties those systems together and bolts AI onto them. It breaks the day total revenue growth turns negative while gross margin slips under 57% in the same quarter, and those are the two numbers to read side by side every ninety days.