IBM Bets Its Record Cash Engine on a Software‑Heavy, AI‑Fueled Future

A year when the flywheel finally shows

For much of the past decade, IBM has asked investors to take it on trust that a lumbering hardware-and-services giant could remake itself as a software-led, hybrid cloud and AI platform player. In 2025, the numbers finally lined up with the narrative.

Arvind Krishna opened the fourth-quarter presentation with something IBM has not been able to say in years: the company exceeded every major metric in the model it set out at its 2025 Investor Day. Revenue grew 6% at constant currency, above the “5% plus” pledge. Free cash flow reached $14.7bn, up 16% year on year and the highest level – and margin – in the company’s recorded history. Operating pretax margin expanded by a full percentage point; adjusted EBITDA jumped 17%.

Crucially for a company trying to convince the market it is no longer tethered to the product cycles of its mainframes, software has become both the centre of gravity and the growth engine. It now represents roughly 45% of IBM’s revenue, up from about 25% in 2018. In 2025 that software segment grew 9%, the fastest rate in IBM’s history, with three of its four sub-segments in double digits.

Beneath those headline figures lies a balance sheet that looks markedly different from the IBM of old: a larger, high-margin recurring software base; a mainframe franchise that has morphed into an AI and security play; and an acquisition machine that is being used as much to knit together a data and automation platform as to bulk up top-line growth.

The macro winds and a changing technology map

Krishna was at pains to situate IBM’s outperformance in a macro environment that is hardly benign. He characterised demand as resilient rather than exuberant, but disciplined in where it lands. Enterprises are prioritising technology that delivers productivity, resilience and flexibility – the properties that hybrid cloud architectures, AI and mission-critical infrastructure are supposed to offer.

In his telling, these categories are no longer optional “tools” but platform decisions that determine how businesses scale and compete. As clients modernise core systems and attempt to squeeze value out of swelling data volumes, they are demanding tighter integration, stricter security and better performance. IBM’s pitch is that its mix of software, consulting and infrastructure – stitched together with partnerships across hyperscalers and chip designers – is well aligned with these structural shifts.

That macro scaffolding matters because IBM is positioning itself not just as a purveyor of discrete products, but as an orchestrator of complex systems. Its partnerships in 2025 spanned AMD and NVIDIA at the silicon layer, Anthropic, OpenAI and Mistral in frontier models, and AWS, Microsoft and Oracle in cloud. Overlay that with Red Hat and IBM’s own Granite models, and the company is effectively arguing it can be the neutral broker in a multi-cloud, multi-model world.

Software: from single-engine to diversified growth

The most striking change in IBM’s profile is in software. Jim Kavanaugh, the CFO, reminded investors that three years ago IBM effectively had one software “growth factor” – Red Hat. Today, it has three sub-segments growing at double digits and a software franchise capable of 10% growth in 2026, with more than seven points of that expected to be organic.

In the fourth quarter, software revenue accelerated to 11% growth, on top of a 11.5% comparison a year ago. The underpinnings are instructive for investors tracking where IBM’s future earnings power will come from:

• Data grew 19% in the quarter, fuelled by demand for generative AI products under the Watsonx banner and reinforced by partnerships with data providers. IBM is explicit that its planned acquisition of Confluent is meant to sit here, as a “smart data platform” that unifies streaming data across applications, clouds and APIs – in effect, a real-time data fabric for AI agents.

• Automation rose 14%, with another record bookings quarter at HashiCorp, which IBM acquired in 2024. That business is being used to anchor an automation suite that ranges from infrastructure provisioning to cost management (via Apptio) and integration software. IBM’s thesis is that as enterprises sprawl across more clouds and more AI workloads, they will need automation to manage complexity – a secular, not cyclical, driver.

• Red Hat decelerated to 8% growth, a wobble that drew close investor attention given its strategic importance. IBM blamed two transitory factors: a tough comparison against elevated consumption-based services in 2024, and delays in U.S. federal deals tied to the government shutdown. Underneath that, it insisted the growth logic remains intact. OpenShift, its Kubernetes platform, is now a $1.9bn ARR business growing over 30%, and virtualization demand has generated more than $500m in contracts over two years. Red Hat Enterprise Linux is growing in the mid-single digits, broadly in line with server markets but with share gains.

• Transaction processing – effectively the mainframe software stack – has quietly strengthened. After a flat year in the first Z16 cycle, 2025 delivered an inflection back to 4% growth in the fourth quarter, off a much higher base. That is a vital profit and cash-flow engine: IBM describes Z software as carrying a “three to four times stack multiplier” on hardware and a “tremendous source of profitability and free cash flow”.

The software business now boasts an annual recurring revenue base of $23.6bn, up more than $2bn year on year and growing in the high single digits. Kavanaugh underlined that subscription revenue under contract is itself rising in the mid-teens, building visibility into future growth.

GenAI as both revenue and cost lever

AI runs as a connective thread through almost every part of IBM’s presentation. The company has been reporting a cumulative “GenAI book of business” since 2023, when it was measured in the low hundreds of millions. By the end of 2025, that figure had swollen to more than $12.5bn – over $10.5bn in consulting and more than $2bn in software – with both arms logging their largest quarterly increases to date.

From 2026, IBM will stop breaking out this metric, arguing that AI is now embedded across its portfolio, from how services are delivered to the features in software and infrastructure. For investors, that decision is a double-edged signal: evidence that AI is truly pervasive, but also a reminder that disentangling its specific economic impact will become harder.

On the revenue side, IBM’s AI narrative rests on breadth. It offers its own Granite models; third-party models from Anthropic, Meta, Mistral and others; and an orchestration layer that can route workloads to the most appropriate model. Watsonx, the AgenTeq platform and Red Hat AI are meant to be the connective tissue that allows enterprises to build domain-specific models while maintaining control over data, security and compliance.

On the productivity side, AI is becoming an internal engine. IBM set itself a target in 2023 to achieve $2bn in annual run-rate productivity savings by the end of 2024. It exited 2025 with $4.5bn, and now expects that figure to rise to $5.5bn by 2026. A central tool here is “Project Bob”, an AI-based software development system that orchestrates multiple models – including Anthropic’s Claude, Mistral, IBM Granite and custom models – to assist developers. Some 20,000 IBMers are using it, reporting average productivity gains of 45% in software development tasks.

These gains are not just a nice internal story: they are central to how IBM plans to absorb acquisition dilution, fund R&D and still expand margins. Kavanaugh framed productivity alongside portfolio mix and revenue scale as one of three levers for operating margin expansion.

Mainframes reimagined: Z17 as AI and security backbone

If software is the new centre of gravity, the mainframe remains the gravitational well. IBM’s Infrastructure segment grew 17% in the fourth quarter; hybrid infrastructure was up 24%, while support services slipped 2%. Within that, IBM Z – the mainframe line – posted a 61% surge in quarterly revenue, delivering the highest fourth-quarter haul in more than twenty years.

The Z17 launch in 2025 has outpaced even the strong Z16 cycle and delivered the highest annual Z revenue in about two decades. Krishna and Kavanaugh attribute that not to nostalgia for legacy systems, but to a convergence of pressures that play to the platform’s strengths.

First, there is a renewed focus on sovereignty and control. In an era of data-localisation rules and board-level cybersecurity concerns, many large enterprises are re-evaluating which workloads they are willing to leave in public clouds. Krishna argued that for certain mission-critical, transaction-heavy applications, the mainframe remains the lowest unit-cost compute platform, with unmatched resilience.

Second, IBM has chipped away at the perception that Z is a difficult and aging environment. Tools such as Watson Code Assistant for Z can refactor COBOL into Java, help developers understand legacy code and modernise applications in place. That addresses a chronic skills and complexity problem that has hung over the platform for years.

Third, IBM is building AI in-line with transactions via Z17’s Telum-based AI accelerators. These “Spire” GenAI cards allow inferencing to occur on the same platform and within milliseconds of a transaction, rather than shipping data off-platform and accepting multi-second latencies. Many Z17 customers, IBM says, have left capacity headroom in their systems to add these cards as their AI use cases mature.

Despite this momentum, IBM is guiding to a low-single-digit decline in Infrastructure revenue in 2026, with about a half-point drag on IBM’s total growth. That reflects the familiar pattern of mainframe product cycles: heavy placement in the first 18 months, then a taper. Privately, though, the company clearly sees room for AI-related tailwinds to sustain Z workloads beyond traditional refresh rhythms.

Consulting: grafting AI onto a maturing services engine

By contrast, IBM’s consulting arm had a more modest year on the top line, but made meaningful strides in profitability. Consulting revenue grew 1% in the fourth quarter, with intelligent operations up 3% and strategy and technology flat. Over 2025, however, operating pretax margins in consulting expanded by 180 basis points, reaching their highest level in three years.

The business is being steered away from traditional, lower-margin outsourcing and towards higher-growth domains: business application transformation, application migration and modernization, cybersecurity, and AI-enabled operations. Discretionary spending is still being scrutinised by clients, but Kavanaugh detected a change in tone from 18 months ago: AI is now about unlocking growth and new business models, not just efficiency and cost cuts.

AI is central here, too. IBM’s generative AI consulting book of business surpassed $2bn in the fourth quarter alone, and generative AI now accounts for more than a third of bookings, over 25% of the consulting backlog and more than 15% of revenue on an exit run-rate basis. The company pegs its GenAI consulting ARR at $3.6bn.

Backlog stands at $32bn, up 2%, with what IBM calls “record low erosion” and improving realisation. It added more than 400 new consulting clients in 2025, shifting its mix towards net new business. The guidance for 2026 – low-to-mid single-digit revenue growth and roughly 150 basis points of further margin expansion – depends on that backlog converting, AI work scaling, and IBM extracting more leverage from its asset-based “services as software” models.

Capital discipline meets M&A ambition

If there is a single sentence that captures IBM’s financial stance, it is Kavanaugh’s line that investors should assume “all in” numbers: dilution from acquisitions is a given, not an exception, and must be absorbed within the model.

In 2025, IBM spent $8.3bn on acquisitions and returned $6.3bn in dividends. It finished the year with $14.5bn of cash and $61.3bn of debt, of which $15.1bn relates to its financing arm. The financing receivables book is almost 80% investment grade, buttressing IBM’s investment-grade rating.

HashiCorp provides the template IBM wants to showcase. Despite being a high-growth, high-investment asset, it was adjusted-EBITDA accretive within the first full year after closing, and delivered another record bookings quarter in Q4. IBM credits its global distribution, product integration – such as fusing Hashi’s InfraGraph with IBM’s automation product Concert – and G&A discipline for those synergies.

Confluent is next. IBM expects the deal to close by mid-2026 and is upfront about the near-term hit: roughly $600m of dilution in 2026, driven largely by stock-based compensation and interest expense. It aims to make Confluent adjusted-EBITDA accretive within its first full year in IBM and free cash flow accretive in year two, underpinned by $500m in operational spend run-rate synergies by 2027 and revenue synergies from cross-selling and product integration.

To make those sums work, IBM is leaning hard on its productivity programme. It accelerated cost actions in the fourth quarter of 2025, including workforce rebalancing, specifically to pre-fund Confluent dilution. Excluding those charges, operating pretax margin expansion in 2025 would have been 140 basis points rather than 100.

Quantum: still long-dated, but edging towards advantage

Almost lost amid the AI and cash-flow headlines was an update on IBM’s quantum computing roadmap. The company deployed its first 120-qubit Nighthawk-based system in December for client use, improving error correction and expanding ecosystem partnerships with Cisco, the U.S. Department of Energy’s Genesis Mission and DARPA’s quantum benchmarking initiative.

IBM reiterated its 2024 prediction that “Quantum Advantage” – a point at which quantum systems deliver clear, credible speed or capability gains over classical for practical problems – would emerge by 2026. Krishna said partners in the scientific computing community are beginning to make the first credible advantage claims, buttressed by rapid iterations in hardware and software. The longer-term target remains a large-scale, fault-tolerant quantum computer by 2029.

For now, quantum is more about signalling IBM’s long-term innovation credentials than moving the P&L. But in a market where investors are increasingly trying to distinguish between AI-driven hype and real technological moat, it serves as a reminder that the company’s R&D budget – up roughly $2.5bn since 2019 – is not being spent solely on the latest buzzword.

2026: guidance with room to outperform

Looking ahead, IBM’s 2026 outlook is deliberately conservative in tone but quietly ambitious in content. The company expects to sustain 5%+ constant-currency revenue growth, with:

• Software growing about 10%, driven by organic momentum in data and automation, recurring revenue strength, genAI traction, M&A synergies and ongoing monetisation of the Z17 cycle.

• Consulting accelerating to low-to-mid single-digit growth, supported by backlog composition and AI-driven transformation work.

• Infrastructure declining low single digits as Z17 moves deeper into its cycle, subtracting roughly 0.5 percentage points from IBM’s overall revenue growth.

Operating pretax margin is guided to expand by about a point, with half a point coming from revenue scale, a half-point drag from mix as the mainframe boom moderates, and a full point from productivity. The operating tax rate should sit in the mid-teens, with quarter-to-quarter volatility from discrete items.

Free cash flow is expected to rise by about $1bn to $15.7bn, a high single-digit increase in line with IBM’s Investor Day model. Adjusted EBITDA growth will again be the primary driver, partially offset by higher cash taxes, higher capex – reflecting ongoing investment in growth areas – and higher net interest expense.

Underneath that guidance sits an implicit challenge: to continue shifting the company’s earnings base towards software and AI while using a disciplined productivity machine to swallow acquisition dilution and cyclicality in hardware. For investors who have long wondered whether IBM’s vaunted “flywheel” is anything more than a metaphor, 2025 delivered the first convincing spin. The task now is to keep it turning at speed, without the tailwind of an exceptionally strong mainframe launch.