IBM Q3 2026 — The Deals June Lost Closed in September; the Mainframe Comparison Is Still Ahead

The setup

IBM’s third-quarter release opens with a quote from Arvind Krishna: “Disciplined execution throughout the quarter drove accelerated revenue growth and strong free cash flow.” The headline says IBM “exceeded expectations with accelerated revenue growth led by Software and Infrastructure.” Revenue was $17.10B, up 5% (6% at constant currency), Software rose 8%, Infrastructure rose 4%, and GAAP income from continuing operations jumped 26.1% to $2.20B.

Fourteen weeks earlier the same chief executive wrote a different letter. On July 14 IBM pre-announced a second quarter it called disappointing: Software up 5%, Infrastructure down 7%, and, in Krishna’s words, “numerous large deals failed to close on the timelines we expected, driving the majority of our shortfall.” The third quarter is where those deals landed. The 10-Q says so in plain terms: Software growth was led by “strong new business performance, including Confluent, and pipeline closure rates as we completed a significant portion of large deals that failed to close within the expected timelines from the second quarter.”

That sentence is the frame for this dive. A deal that slips one quarter shows up twice in the year-over-year rates, once as the miss and once as the beat, and neither quarter on its own tells you the growth rate. Read the second and third quarters of 2026 as one half-year against the same half of 2025 and the picture changes. Transaction Processing, the mainframe software that IBM blamed for the June miss, grew 0.1%. Infrastructure fell 2.1%. Total revenue grew 2.9%. The acceleration is one quarter’s timing plus an $11.27B acquisition.

Our call follows from the calendar IBM is now running into. IBM reports Q4 2026 Infrastructure revenue below Q4 2025’s $5.13B, the peak quarter of the z17 mainframe launch. We are wrong if the January 8-K shows Infrastructure revenue of $5,132M or more. The full-year guide does not need Infrastructure to grow, which is why the guide can hold and the call can still be right; the section on the Q4 bar computes what the rest of IBM has to deliver if it does not.

IBM closed at $251.46 on October 21, before the release, and traded at $249.70 at 11:22 a.m. PT on October 22, down 0.70%. The market read the print as roughly in line, which is the right read of the half-year and the wrong read of the headline.

The numbers

Read the Sankey from the right. Income from continuing operations was $2.20B, up 26.1%, on pre-tax income that rose 2.8%. The gap is the tax line: $298M this quarter against $686M a year earlier, when IBM booked a one-time, non-cash charge on the enactment of H.R. 1 that its own reconciliation sizes at $309M. Add that charge back to Q3 2025 and income from continuing operations grew 7.1%, not 26.1%. The operating (non-GAAP) line, which already strips it out, grew 11.7% on pre-tax income up 8.6%. Both are respectable numbers. Neither is the number in the headline table.

What’s actually driving it

A slipped deal counts twice, so read the half-year. The release calls Q3 “accelerated revenue growth.” Decode it against the quarter before. Transaction Processing, the software that runs on the mainframe, was $2,030M in Q2 2026, down 8.1% on $2,208M, and $2,111M in Q3, up 9.3% on $1,930M. Put the two quarters together and it is $4,141M against $4,138M, growth of 0.1%. The nine-month figure in the 10-Q says the same thing more gently: Transaction Processing up 1.8% for the year to date. Infrastructure follows the same pattern from the hardware side: $3,835M in Q2 (down 7.4%) and $3,707M in Q3 (up 4.2%) make $7,542M against $7,701M, a 2.1% decline over the half. Total revenue for the half rose 2.9%. That is the growth rate of IBM in the middle of 2026. The 4.7% of the third quarter is that rate plus the deals June lost.

This matters because the release asks you to read Q3 as a turn. Krishna’s July letter said the shortfall came from deals that “failed to close on the timelines we expected.” If those deals closed in Q3, as the 10-Q says a “significant portion” did, Q3’s growth includes revenue that belonged to Q2’s plan. One quarter on its own is a phase, and you cannot call it a decision until you have the series; here the series is two quarters long. It says IBM grew about 3%, and that the mainframe franchise, hardware and software together, is flat to down.

Confluent is most of the Software acceleration. The 10-Q credits Software’s growth to “strong new business performance, including Confluent.” Decode the lines. Software revenue rose $595M on the year, to $7,804M. Data, the line that holds Confluent, rose $279M to $1,738M, up 19.1%, and the 10-Q says that growth was “driven by the strong performance from Confluent.” IBM closed the Confluent acquisition on March 17, 2026, so Q3 2026 carries a full quarter of it and Q3 2025 carries none. The Data line alone supplied 46.9% of Software’s growth. Take it out and Software grew from $5,750M to $6,066M, 5.5%, and that 5.5% still contains the Q2 deals that closed late in Transaction Processing. The composite hides the component: an 8.2% Software line is a Red Hat business growing 10.6%, an Automation business shrinking 3.3% against a strong prior year, a Data line bought rather than built, and a mainframe software line catching up on a missed quarter.

None of that makes the Software segment weak. Red Hat’s Hybrid Cloud line was $2,086M, up 10.6% (11.1% at constant currency), with what the 10-Q calls “accelerated bookings growth,” and OpenShift annual recurring revenue is about $2.4B. That is the organic engine, and it is growing faster than it did in the second quarter. The point is narrower: the acceleration in the headline is not Red Hat’s, and it will not be in the fourth quarter’s comparison, because Confluent laps in March 2027 and the slipped deals have already been booked.

The software margin is real, and it is where the profit came from. Software segment profit was $2,734M, up 15.2% on $2,374M, a 35.0% margin against 32.9%. The $360M increase in profit came on $595M more revenue, an incremental margin of 60.5%. Find the segment whose margin actually moved before crediting the acquisition: here it is Software, and Consulting went the other way, with segment profit down 4.8% to $653M and margin down to 12.2% from 12.9%. Infrastructure profit rose 11.0% to $715M on a 19.3% margin. Software was 64.9% of the quarter’s segment profit across the four segments. IBM’s earnings are now a software company’s earnings with a services business and a hardware cycle attached.

The tax line wrote the 26%. CFO James Kavanaugh’s quote in the release points at “solid operating pre-tax margin expansion in the quarter.” That claim is accurate on the operating basis: operating pre-tax margin was 19.3%, up 0.7 points. Decode the GAAP line underneath it. GAAP pre-tax margin fell 0.3 points to 14.6%. GAAP pre-tax income rose 2.8%. The effective tax rate fell to 11.9% from 28.2% because Q3 2025 carried the H.R. 1 charge, and the 10-Q says the year-to-year change in the provision “is primarily driven by the one-time, non-cash income tax charge recorded in the third quarter of 2025.” The GAAP pre-tax line is where the cost of buying growth lands: acquisition-related adjustments rose to $707M from $590M, up 19.8%, and amortization of capitalized software and acquired intangibles rose 18.5% to $828M. IBM’s operating results exclude both. The difference between a 2.8% pre-tax line and an 8.6% operating pre-tax line is, almost entirely, the price of the companies IBM has bought.

Inside the operating line, SG&A paid for R&D. Operating (non-GAAP) SG&A fell 4.6% to $4,193M from $4,394M. R&D rose 14.2% to $2,378M, 13.9% of revenue against 12.7% a year ago. Stock-based compensation rose 20.5% to $535M. This is the productivity program Kavanaugh describes, and it is a reasonable trade: IBM is cutting selling and overhead cost to fund engineering. It also has a limit. SG&A cannot fall 4.6% a year indefinitely while R&D rises 14.2%, and the quarter’s operating margin expansion came from that gap rather than from gross margin, which fell 0.2 points on the operating basis to 58.5%.

The mainframe is in the second year of its cycle. The 10-Q says IBM Z revenue “decreased 19.6 percent as reported (19.3 percent adjusted for currency) reflecting product cycle dynamics.” IBM does not disclose IBM Z revenue in dollars, only its growth rate, so the cycle has to be read from the rates. IBM Z fell 15% in Q1 2025, then grew 70%, 61%, 67% and 51% in the four quarters after the z17 launched in June 2025. It has now fallen 42% and 20% in the two quarters that compare against that launch. The 10-Q also says that the z17 “remains our strongest program on record,” and both statements are true at once: the strongest launch on record makes the hardest year of comparisons on record.

Distributed Infrastructure is growing on the shortage the July letter described. The July 14 letter explained the June miss this way: “In the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases.” Decode what that sentence did to IBM’s own lines. Distributed Infrastructure, which is IBM’s Power servers and Storage, grew 37% in Q2 and 31% in Q3, the two fastest quarters in the seven-quarter series, while IBM Z and its Transaction Processing software lost the budget. The 10-Q credits the growth to “continued strong performance in both Storage and Power.” When memory is scarce, the server maker pays the rent; in this quarter’s version, the buyer pays it too, by pulling server and storage spend forward and deferring a mainframe upgrade it can defer. IBM sells both, so it captured some of the pull-forward. It did not capture all of it.

IBM is stocking up on the same parts. The 10-Q says working capital was “partially offset by an increase in inventory to support anticipated customer demand.” Decode the inventory note. Total inventory was $1,945M at September 30, up 59.4% from $1,220M at year-end and 39.2% from $1,397M a year earlier. The increase is not finished machines: finished goods were $320M, the same as at June 30. It is work in process and raw materials, which rose from $990M at December 31 to $1,208M, $1,425M and $1,625M at the three quarter-ends, about $200M or more every quarter, up 64.1% for the year to date. A hardware maker building raw-material inventory at that pace, in the year its own chief executive described clients buying memory ahead of price increases, is buying components ahead of the same increases. The filing does not say which components; the read that this is memory and storage media is ours, and it is consistent with Distributed Infrastructure’s growth. The warehouse is the position: at $1.63B of work in process and raw materials, IBM is carrying a component bet that pays if Power and Storage demand holds into 2027 and costs margin if prices turn first.

Consulting is flat, and the signings say it stays flat. Consulting revenue was $5,335M, up 0.2% (1.8% at constant currency), within the $5.07B to $5.35B band it has held for seven quarters. Signings fell 3.7% in the quarter to $5,030M from $5,223M, though they were up 3.1% for the nine months at $15,418M. The trailing twelve-month book-to-bill ratio was about 1.04, and backlog was $30.2B. A book-to-bill just above 1.0 is a business replacing what it bills, not one growing. Consulting gross margin fell to 27.4% from 29.3%, the largest gross-margin decline of the three main segments. This is the segment in which generative AI is supposed to show up as demand, and for the moment it shows up as price pressure.

The recurring base grew, mostly by purchase. The 10-Q says Software’s annual recurring revenue “was solid at $24.9 billion, and increased approximately $2 billion as reported year to year.” Decode it against the earlier filings. The Q3 2025 10-Q put ARR at $23.2B, so the change on the two filings’ own figures is $1.7B. ARR was $24.6B at both March 31 and June 30, 2026, after Confluent closed, and rose $0.3B in the third quarter. The 10-Q names Confluent first among the reasons for the year-over-year increase. ARR is a commitment; deferred revenue is a bill, and the bill moved the expected way for the season: current deferred income fell to $15.16B from $16.10B at year-end. Remaining performance obligations were about $69B, against $64B a year earlier and $68B at June 30. About 70% of it is expected to be recognized within two years.

The cash flow was strong, and the financing book helped the GAAP line. Q3 free cash flow, IBM’s non-GAAP measure, was $3,027M, up 27.6% from $2,373M. That number excludes the change in IBM Financing receivables, so it is the cleaner one. GAAP operating cash flow was $4,025M, and the 10-Q is direct about the nine-month figure: the $2.6B increase “includes an increase in cash provided by financing receivables of $1.9 billion.” Financing receivables are IBM lending to its own customers, mostly to buy mainframes; as the z17 lease book collects, it releases cash. Free cash flow for nine months was $7,787M, up $606M, a smaller increase than the operating cash flow line suggests, and the one the guide is written against.

Confluent was paid for with the cash pile and new debt. IBM paid $11,268M for Confluent’s common stock and $269M for its equity awards, received $1,165M of cash and $917M of short-term marketable securities in the deal, and settled $1,100M of Confluent convertible notes on April 15. Across four acquisitions in the nine months, the aggregate purchase price was $12,600M. Cash, restricted cash and marketable securities fell $3.4B to $11.1B; total debt rose $2.6B to $63.8B, of which $12.7B is IBM Financing debt. Goodwill rose $7.20B to $74.92B and net intangibles rose $2.21B to $13.60B; together they are $88.52B, 2.49 times IBM stockholders’ equity of $35.54B. In nine months, free cash flow of $7,787M covered dividends of $4,762M; it did not also cover $11,364M of acquisitions, and the difference came from the balance sheet. One line the release does not explain: noncontrolling interest transactions brought in $575M in the third quarter, and noncontrolling interests on the balance sheet rose to $669M from $93M at year-end.

Here is the bull case at full strength. Free cash flow rose 27.6% in the quarter and IBM raised its full-year free cash flow outlook to growth of more than $1 billion. Software margins expanded 2.1 points to 35.0% and Software produced nearly two-thirds of segment profit. Red Hat accelerated to 10.6% with bookings growth accelerating. Distributed Infrastructure grew 31% and the company is building inventory for demand it says it can see. RPO of about $69B is $5B above a year ago. On this reading the Q2 miss was a timing problem that has already been fixed, the mainframe decline is a known comparison that ends in mid-2027, and an IBM that converts 138% of net income into free cash flow in a quarter, as Q3’s $3,027M did against $2,190M of net income, deserves a higher multiple than an IBM that grows 3%. Our answer is that every number in that paragraph is true and the half-year arithmetic still governs: the bull case needs the third quarter’s growth rate to be the run-rate, and the second quarter says it is not.

The guide and the Q4 bar

IBM held its full-year revenue outlook and raised its cash outlook. The release says: “For the full year, we continue to expect constant currency revenue growth in the range of four-to-five percent, and we now expect free cash flow to increase by more than $1 billion year-over-year.” It also says currency “is expected to be neutral to growth for the year,” so we treat reported and constant-currency growth as the same for the full year; that is an assumption, and a currency move in the fourth quarter would shift every figure below by its size.

Compute what the fourth quarter has to deliver. 2025 revenue was $67.54B. Growth of 4% puts 2026 at $70.24B; nine months have produced $50.18B, so Q4 needs $20.05B, up 1.9% on Q4 2025’s $19.69B. Growth of 5% puts 2026 at $70.91B and Q4 at $20.73B, up 5.3%. The low end is easy on paper. The question is the mix. Q4 2025 Infrastructure revenue was $5,132M, its peak in the series, and the release for that quarter shows IBM Z up 67% and Distributed Infrastructure up only 3%. Infrastructure revenue in Q4 2025 was $1.57B higher than in Q3 2025, and on those growth rates most of that increase was mainframe. In Q3 2026, Distributed Infrastructure’s 31% growth was enough to outrun a 20% IBM Z decline. In Q4, the mainframe is a larger share of the base and the Distributed comparison is easier but smaller. That is why we expect Infrastructure to fall.

If Infrastructure falls 10% in Q4, to $4.62B, the rest of IBM must produce $15.44B for the low end of the guide, growth of 6.1% on the $14.55B the other segments and lines produced in Q4 2025. If Infrastructure falls 20%, to $4.11B, the rest must grow 9.6%. Software’s 8.2% in Q3 included Confluent, which is still in the base until March 2027, and the slipped Q2 deals, which are not. Consulting has not grown faster than 4% in any quarter of the series. The low end is reachable with Infrastructure down 10%; it gets hard below that.

The cash guide is a sharper test. Free cash flow in 2025 was $14,734M, so growth of more than $1 billion means more than $15,734M. Nine months have produced $7,787M, so Q4 2026 must produce more than $7,947M, against $7,553M in Q4 2025: an increase of more than $394M in a single quarter, after nine months that added $606M in total. IBM’s free cash flow is concentrated in the fourth quarter every year, when its software renewals bill. The guide says the fourth quarter of 2026 will be the largest single quarter of free cash flow in the seven-quarter series. A falling Infrastructure line makes that harder, because the hardware quarter is also a collections quarter.

How peers read this

IBM’s July letter is the most specific statement any large vendor has made about how the memory shortage reallocates a customer’s budget. Clients, it said, moved “quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases.” That is the demand side of the story we wrote up in Micron’s fiscal Q4: when memory is short and its price is rising, buyers pull purchases forward, and the budget they use is the one for the purchase that can wait. In IBM’s case the purchase that waited was a mainframe upgrade and its software. For memory suppliers, IBM’s Q3 is a confirmation that the pull-forward ran at least through September, since Distributed Infrastructure grew 31% and IBM’s own raw-material inventory kept rising.

For the hyperscalers, the read is the opposite one. In Meta’s Q3, written up this morning, the question was who funds a commitment book that outruns cash flow. IBM is the enterprise buyer’s side of the same market: it does not build data centers at that scale, it sells into the corporate budget the hyperscalers are competing with. A corporate client that reprioritizes toward servers and memory is a client spending its own capex rather than renting capacity, which is good for Power and Storage and neutral at best for the cloud contracts that sit behind IBM’s Consulting signings, down 3.7% in the quarter.

For enterprise software peers, Confluent is the comparison. Data revenue grew 19.1% almost entirely by acquisition, and Automation fell 3.3% despite HashiCorp. IBM is buying the data-movement and infrastructure-automation layers that agentic workloads need, and paying for them with goodwill and debt. Its organic growth is Red Hat, at 10.6%. That is the number to hold against any infrastructure software peer, not the 8.2% segment total.

Valuation reality check

At $249.70 on 945,120,805 shares outstanding at September 30, the cover-page count, IBM’s market value is about $236.00B. Add total debt of $63.82B (short-term debt of $7,013M and long-term debt of $56,804M) and subtract cash, restricted cash and marketable securities of $11.11B, and enterprise value is about $288.71B. Excluding the $12.7B of IBM Financing debt, which is matched by financing receivables, it is about $276.01B.

On trailing figures: adjusted EBITDA for the four quarters to September was $20.2B ($6.5B, $4.0B, $4.8B and $4.9B), so enterprise value is 14.3 times EBITDA, or 13.7 times excluding the financing debt. Operating (non-GAAP) diluted EPS for the four quarters was $12.31 ($4.52, $1.91, $2.93 and $2.95), a multiple of 20.3 times; GAAP diluted EPS from continuing operations was $11.72, a multiple of 21.3 times. Trailing free cash flow was $15.34B, a 6.5% yield on market value. The quarterly dividend of $1.69 is $6.76 a year, a 2.71% yield.

On an annualized basis, Q3 operating EPS of $2.95 times four is $11.80, a multiple of 21.2 times, and Q3 adjusted EBITDA of $4.9B times four is $19.6B. Annualizing any IBM quarter other than the fourth understates the year, because the fourth quarter carries the software renewals and, in a launch year, the mainframe; the trailing basis is the honest one here.

Split by segment, IBM’s trailing segment profit was $17.41B: Software $10.74B (61.7%), Infrastructure $3.68B, Consulting $2.52B and Financing $0.48B. A market value of $236.00B is 22.0 times trailing Software segment profit alone, before Consulting, Infrastructure and Financing and before corporate costs, interest and tax. That is a software multiple for a company whose software grew 5.5% excluding the line it bought. The market is paying for Red Hat and the mainframe annuity, and the 20.3 times operating multiple has room only if the 3% half-year becomes 5% again without a slipped quarter to help.

What we’d watch from here

• Q4 2026 Infrastructure revenue against $5,132M. That is the call: below it. At or above it, the mainframe trough is shallower than the series implies, and we are wrong.
• Q4 free cash flow against $7,947M. The raised guide needs more than that. A miss would mean the cash raise was a nine-month number dressed as a full-year one.
• Transaction Processing across Q3 and Q4 together. The half-year read put it at 0.1% growth. A Q4 decline confirms that the mainframe software stack follows the hardware down.
• Work in process and raw materials against $1,625M. Another rise of $200M says IBM still expects Power and Storage demand at higher component prices; a fall with flat Distributed growth says the pull-forward has run its course.
• Data line growth after March 2027. When Confluent laps, the Data line’s 19.1% becomes organic or it does not. Software growth excluding Data, 5.5% in Q3, is the run-rate to track until then.
• Consulting book-to-bill against 1.04 and signings against $5,030M. A ratio below 1.0 would turn a flat segment into a shrinking one.
• The $575M noncontrolling interest inflow. IBM sold a minority stake in something in Q3, and the release does not say what. The 10-K should.

The trade

Evidence-led, not a chase. A 0.70% move on the reaction day says the market priced the quarter close to in line, which is consistent with reading the half-year rather than the quarter. The cleanest expression of the call is not directional on the stock. It is a view on mix: Infrastructure and the mainframe software line weaken into the January print while Red Hat and the cash guide hold. Investors who want the Red Hat engine are paying a 20.3 times operating multiple for a company that is about 60% software by profit and about 3% growth by the half-year. Those who already hold IBM for the dividend have a 2.71% yield covered 1.90 times by Q3 free cash flow ($3,027M against $1,596M of dividends paid). Neither group gets paid for the acceleration in the headline, because it was a quarter late.

The risks to the call sit in IBM’s own lines. Distributed Infrastructure grew 31% against a 10% comparison in Q3 and faces a 3% comparison in Q4; if it grows 40% or more, it can carry Infrastructure through. A large mainframe deal that slipped from Q3 into Q4, if there was one, would do the same. And IBM has shown, twice this year, that its quarter-end close rates can move a segment by several points in either direction. Those are the conditions under which we are wrong.

Bottom line

The Q3 2026 release says IBM accelerated. The two quarters together say IBM grew about 3%, its mainframe franchise shrank, its Software growth was half acquisition, and its 26% earnings growth was a lapped tax charge. The good news is genuine: Red Hat at 10.6%, a 35.0% Software margin, free cash flow up 27.6%, and a raised cash guide. The fourth quarter now has to produce the largest free cash flow quarter in the series against the peak quarter of the z17 cycle. We expect IBM to make the cash guide and report Infrastructure revenue below Q4 2025’s $5.13B. The January 8-K settles it.