IBM Software Grew 8.2% in Q3 2026. Without Confluent, the Agent Trade Shrinks

IBM software revenue rose 8.2% to $7.8 billion in the third quarter of 2026, on total revenue up 4.7% to $17.1 billion. Roughly half of the software gain came from Confluent, which IBM bought in March for about $11.6 billion. Take that out and IBM's software business grew about 4%. The agent tailwind is real, but it shows up in one product line, not across the segment.

The short answer

IBM's Q3 2026 software growth was mostly acquired, not agent-driven. Confluent added an estimated $300 million to $355 million, about half the $595 million increase. The organic part, near 4%, came from Red Hat and mainframe software. OpenShift, growing more than 30%, is the one line where agent demand plainly shows up in IBM's numbers.

What IBM reported in its Q3 2026 earnings

IBM reported Q3 2026 revenue of $17,103 million on 21 October, against $16,331 million a year earlier. That is 4.7% reported growth, which the company rounds to 5%, or 6% at constant currency. Full-year guidance held at 4% to 5% constant currency growth.

Software was the engine. The segment grew from $7,209 million to $7,804 million, an increase of $595 million. Consulting was flat at $5.3 billion. Infrastructure rose 4% to $3.7 billion, even though IBM Z revenue fell 20% as the mainframe cycle aged.

Free cash flow was $3.0 billion for the quarter, up $0.7 billion, and IBM raised its full-year cash flow outlook to growth of more than $1 billion. Those are the numbers most coverage led with, and they are accurate.

Product-line figures from Note 3 of IBM's Q3 2026 Form 10-Q. Growth rates match the percentages in IBM's earnings release. Rows may not sum exactly because of rounding.

Margins moved more than revenue

The more striking line is profit. Software segment profit rose from $2,374 million to $2,734 million, a 15.2% increase on 8.2% revenue growth. The segment margin went from 32.9% to 35.0%.

That gap matters for anyone valuing infrastructure software. IBM is converting growth into profit at nearly twice the revenue rate. Consulting, by contrast, earned a 12.2% segment margin and Infrastructure 19.3%. Software now produces most of IBM's segment profit from less than half its revenue.

Where the $595 million of IBM software growth came from

The headline 8.2% is the sum of four very different stories. Two lines grew organically. One line shrank. One line grew mostly because IBM bought a company and added its revenue to a base that did not include it a year earlier.

The bridge below lays the increase out in sequence. It assumes Confluent contributed exactly what it earned as a standalone company in Q3 2025, $298.5 million, which is the most conservative reasonable estimate.

Read the bridge left to right and the shape is clear. Red Hat added $200 million. Mainframe software added $181 million. Automation lost $64 million, and the pre-Confluent data business was roughly flat to slightly down. Confluent then supplied the largest single step.

None of this is hidden. IBM's CFO said on the call that data growth was "driven by another strong performance from Confluent". The release simply does not split it out, and most coverage repeated the 8.2% without doing the subtraction.

IBM software revenue without Confluent

IBM closed the Confluent acquisition on 17 March 2026, so the deal sits inside Q3 2026 revenue and entirely outside Q3 2025. That is the definition of inorganic growth. To estimate the organic rate you need one number IBM does not publish: what Confluent earned inside IBM this quarter.

The nearest primary anchor is Confluent's own final filings. Confluent reported Q3 2025 revenue of $298.5 million, up 19% year on year. The Motley Fool cited a later quarter at $314.8 million. IBM said Confluent bookings grew about 20% in Q3 2026.

Three Confluent scenarios

Each scenario below holds everything else constant and asks one question. If Confluent contributed this much, what did the rest of IBM's software business grow?

Organic growth equals the gain after removing Confluent, divided by Q3 2025 software revenue of $7,209 million. Directional. IBM does not disclose Confluent revenue, and purchase accounting may reduce what it recognises.

All three land in the same place. IBM's organic software growth was between 3.3% and 4.1%. CFO James Kavanaugh described the quarter as "including over four points of organic growth", which sits at the top of that range. My estimate and IBM's own framing agree. The 8.2% headline does not.

I would not call this a scandal. Buying growth is a legitimate strategy, and IBM disclosed the deal openly. But if you are using this quarter as evidence that infrastructure software is winning the agent era on its own merits, you are using a number that is half acquisition.

Red Hat and OpenShift: the part of the agent thesis that holds

Red Hat is where the case for infrastructure software in the agent era looks strongest. IBM's Hybrid Cloud line, which is Red Hat, grew 11% to $2,086 million. Arvind Krishna said bookings growth was "accelerating back to the high teens". Bookings lead revenue, so that points to faster growth ahead rather than behind.

The product doing the work is OpenShift, Red Hat's container platform for running applications across clouds and data centres. Kavanaugh said "OpenShift is now a $2.4 billion ARR business, with growth accelerating to more than 30%."

OpenShift is the cleanest signal in the release

Krishna tied that acceleration directly to agents. He said OpenShift ARR growth accelerated "as organizations scale super intelligence applications and agents into production". That is a vendor describing its own demand, so treat it as a claim. It is still the most specific link between agent deployments and infrastructure revenue that any large vendor has offered this earnings season.

The logic is plausible on its face. An agent that runs inside a regulated company needs somewhere to run, and many of those companies will not hand every workload to a public model API. A platform that runs the same workload on premises, in a private cloud or on a hyperscaler is the kind of layer agent builders buy regardless of which model or application wins.

Why the bookings number matters more than the revenue number

Red Hat revenue grew 11% for a second straight quarter. On its own that is solid rather than spectacular, roughly the pace of Workday. The interesting part is the gap between 11% revenue growth and high-teens bookings growth. If that gap persists, Hybrid Cloud revenue should accelerate through 2027.

My view is that this is the single number worth tracking from IBM. If OpenShift growth stays above 30% for two more quarters, the infrastructure counter-trade has real evidence behind it. If it drops back to the low 20s, Q3 was a good quarter and nothing more.

Automation and data: the IBM software lines agents should help, and did not organically

If agents were lifting IBM's whole software stack, Automation and Data are where it should show. Automation covers workflow, integration and infrastructure management tools, including HashiCorp. Data covers databases, data integration and watsonx.data. Both are things agent builders say they need.

Automation fell against a hard comparison

Automation revenue fell 3% to $1,870 million. Kavanaugh's explanation was brief: "automation was impacted by a tougher compare." Q3 2025 Automation revenue of $1,934 million was itself a strong quarter, so some of this is arithmetic rather than demand.

IBM still guides to low double-digit Automation growth for the full year, which implies a strong fourth quarter. That guide carries execution risk, and it means this line has to do more work in Q4 than it did in Q3.

Data growth is mostly Confluent

Data revenue rose $279 million, or 19%. Krishna said it grew "supported by continued strength in Confluent". Under any of the three scenarios above, Confluent accounts for more than the entire increase.

That means IBM's pre-existing data business shrank, by somewhere between 1% and 5% depending on what Confluent contributed. This is the most uncomfortable finding in the quarter. The part of IBM's portfolio most directly pitched at "AI-ready data" did not grow on its own.

Krishna made the case for why Confluent fits anyway. "Agents need the current state of the business," he said, describing streaming data as a feed for agents and inference. I find that argument persuasive, and it is consistent with the problem described in the analysis of stale data producing wrong agent answers. It is also an argument IBM had to buy its way into.

IBM software growth against application SaaS peers

The SaaSpocalypse thesis holds that agents hollow out application software, the seat-priced tools that sit on top of the stack. The counter-trade says infrastructure underneath wins instead. If that were happening, IBM's organic software growth should beat the application layer. It does not.

ServiceNow reported Q3 2026 subscription revenue of $3,988 million, up 21%, on 28 October. SAP reported cloud revenue of €6,668 million, up 26%, on 21 October. Workday's subscription revenue rose 13.9% to $2.471 billion in the quarter to 31 July.

Salesforce is the closest match to IBM's situation because it also bought growth. Its subscription and support revenue rose 12% to $10,820 million, including $440 million from Informatica. Remove that and organic growth was about 7.1%, a calculation laid out in the breakdown of Salesforce's Informatica-adjusted growth. Even after the adjustment, Salesforce grew faster than IBM.

The last column is where this comparison is weakest. Only IBM and Salesforce are adjusted here. SAP's figure includes what it called "a multi-quarter revenue catch-up from customers returning from financial distress". The SAP picture is examined separately in the analysis of SAP's cloud backlog and organic growth.

This does not mean the application layer is safe. Several of these vendors are growing more slowly than they did two years ago, a trend tracked in the revenue growth tracker for listed SaaS. It means the reported numbers do not yet show value moving from applications to infrastructure. The market priced that shift in early 2026, as covered in the explainer on the SaaSpocalypse selloff. Revenue has not followed.

What IBM's Q3 2026 numbers say about where AI value accrues

The quarter gives a more precise answer than "infrastructure wins". Value is accruing in the specific products agent builders cannot avoid, and those products are narrow. OpenShift is one. Distributed infrastructure, IBM's storage and server line, grew 31%, and Krishna said "AI continues to increase demand for storage".

The broader categories are not moving together. Automation fell. Legacy data shrank. Consulting was flat. The pattern looks less like a rising tide across the stack and more like a handful of chokepoints pulling ahead while everything around them grows at historical rates.

That matches the argument in the analysis of point SaaS being absorbed into agent ecosystems: value concentrates in layers that every agent has to pass through. A platform that runs the workload is one of those layers. A middleware catalogue is not automatically one.

So my position is narrower than the bullish headlines. Infrastructure software sold to agent builders is a real trade, but it has to be bought product by product. A company's segment total tells you little. Its fastest-growing product with a disclosed ARR tells you most of what you need.

Where this argument is weakest

Every estimate above rests on assumptions IBM has not confirmed. These are the places where I could be wrong, in order of how much they could move the conclusion.

Purchase accounting could flatter the organic estimate in either direction

When a company is acquired, its deferred revenue can be revalued, which can reduce the revenue the buyer recognises for several quarters. If Confluent contributed less than $298.5 million inside IBM, organic growth was higher than 4.1%. Confluent may also have grown faster than 19%, which would push organic growth below 3.3%. IBM has not disclosed the figure, so the range is a bracket, not a measurement.

The peer comparison mixes quarters, currencies and acquisitions

Salesforce and Workday report on quarters ending 31 July, not 30 September. SAP reports in euros and flagged a one-off catch-up. I did not adjust ServiceNow, Workday or SAP for their own acquisitions. A fully organic peer set would likely show slower application growth than the chart suggests, which would narrow the gap with IBM.

One quarter, with slipped deals inside it

IBM said Q3 included "around two-thirds of the large deals that slipped from the second quarter". That flatters Q3 and depressed Q2, when software grew 5.1%. IBM now expects full-year software growth of about 7%, which implies a slower fourth quarter. Judging the agent thesis on any single IBM quarter is a mistake, including this one.

The strongest counter-case is simple. Buying Confluent was itself the infrastructure bet. A buyer who pays $11.6 billion for streaming data because agents need it is acting on the thesis, and acquired growth is still growth for shareholders. If you only care about where enterprise money lands, not whether IBM built the product, the 8.2% is the right number.

Frequently asked questions

Where to start

Run the Confluent test on the next vendor growth claim that lands in your inbox. Ask what the figure is excluding acquisitions made in the last 12 months. If the vendor cannot answer, assume the organic number is materially lower and plan around that. The same discipline applies to the metrics compared in the ranking of AI revenue disclosure metrics.

If you are deciding where your own product sits in the stack, find the one line in your business that an agent cannot route around. Price and invest around that line. IBM's quarter shows that a segment total can look healthy while only one product is doing the compounding.

Related reading