In 2025 IBM turned over $67.5B (up 8%) and generated $14.7B of free cash flow[6][15] — its fastest growth in years, on the back of Red Hat, the AI-focused z17 mainframe and a generative-AI book of business it says now tops $12.5B.

The genuinely open question is not whether IBM is a cash-rich franchise — its $14.7B of free cash flow settles that — but whether 2025 marks a durable turnaround or a cyclical high. The same company posting record results spent the 2010s shrinking while rivals grew, and built its modern AI pitch on the ashes of Watson. Weighed across the four questions below, the evidence leans toward a real but modest turnaround — durable single-digit growth and compounding cash, not a re-rating-grade re-acceleration — with the AI book best read as pipeline and the mainframe surge as cyclical. The full weighing, with the tripwires that would change each call, closes the Risks section.

The decisive questions

Each links to the section that lays out the evidence on both sides.

Five years of revenue, continuing operations

Total revenue, US$B. The 2021 base is post-Kyndryl-spinoff (continuing operations), so the line understates how much smaller IBM became before it re-accelerated. FY2025 is the sourced focus; prior years are shown for trend.

How to read this

Nine sections, each built the same way: a neutral synthesis, a two-sided case-for / case-against ledger, sourced data and charts, and dated facts. Start with the question that interests you, or read in order from the Overview.

Today's IBM is a $67.5B-revenue[6], ~286,800-employee[5] enterprise-technology company organized around four segments — Software, Consulting, Infrastructure and Financing. The defining strategic moves of the modern era were buying Red Hat[3] and spinning out Kyndryl[4]: together they reshaped IBM from a sprawling services giant into a software-and-hybrid-cloud company with a still-large consulting arm.

A company that has reinvented itself before

IBM's history is a sequence of bet-the-company pivots. It dominated tabulating machines, then mainframes (the System/360 defined the category[2]), survived a near-death experience in the early 1990s via Lou Gerstner's shift to services, and under Ginni Rometty and then Arvind Krishna re-tooled again around cloud and AI. The through-line is an enormous, sticky base of the world's largest banks, insurers, retailers and governments — a base whose switching costs have accumulated over a century.

The modern reset: Red Hat in, Kyndryl out

The two transactions that define the current IBM both landed within three years. In 2019 it paid roughly $34B for Red Hat[3] — the open-source company behind Enterprise Linux and OpenShift — making hybrid cloud the centre of the strategy. In 2021 it spun off its lower-margin managed-infrastructure-services business as Kyndryl[4], shedding ~$19B of revenue to concentrate on software and higher-value consulting. The result is a smaller, more profitable IBM — and a revenue line that looks like a reset rather than a continuous trend.

Timeline

IBM competes in large, slow-growing-but-durable enterprise markets where trust, compliance and switching costs matter more than raw scale. It chose not to fight AWS, Azure and Google Cloud head-on for hyperscale workloads[8], instead positioning as the hybrid-cloud and "picks-and-shovels" enabler of enterprise AI — a smaller addressable pie than the hyperscalers', but one where IBM's century of relationships is a real advantage.

Four overlapping markets

• Enterprise software & hybrid cloud — Linux, containers (OpenShift), automation, data and security, anchored by Red Hat. Competes with Microsoft, Oracle, VMware/Broadcom and ServiceNow.
• IT consulting & systems integration — large-scale transformation work, where Accenture, Deloitte, TCS and Infosys are the rivals[18].
• Mainframe & infrastructure — the IBM Z franchise, a market IBM effectively owns and where it has no direct competitor[22].
• Enterprise AI — watsonx and Granite models, sold into the same regulated enterprises, against hyperscalers and AI labs.

The deliberate non-bet: hyperscale cloud

The single most important market fact about IBM is what it is not. After failing to build a top-tier public cloud in the 2010s, IBM ceded hyperscale to AWS, Microsoft and Google and re-cast itself as the company that helps enterprises run AI and applications across clouds and on-premises[8]. The bet is that most large enterprises will stay hybrid — keeping sensitive and regulated workloads on systems like the mainframe while bursting to public cloud — and will pay IBM to stitch it together. The z17's on-chip AI inference (the Spyre accelerator) is the clearest expression of this: AI that runs next to the data, on premises, for banks and insurers that will not move it[7].

Software is now IBM's largest and most strategic segment at $29.96B (+11%)[10], followed by Consulting at $21.06B (+2%, flat at constant currency)[12] and Infrastructure at $15.72B (+12%), the latter supercharged by a +67% Q4 jump in mainframe revenue[13]. High-margin, recurring software plus a cash-cow mainframe carry the model; consulting is the drag.

FY2025 revenue by segment

Source: IBM FY2025 results.[10][12][13][14]

The two speeds, in one chart

Segment revenue growth, FY2025 reported (% YoY).

How each segment makes money

The economics: high margins, strong cash

The mix is increasingly favorable. FY2025 GAAP gross margin reached 58.2% (operating non-GAAP 59.5%), and free cash flow hit $14.7B, up $2.0B year over year[15] — the highest in years. Software's recurring revenue (Red Hat alone, with OpenShift ARR of $1.9B growing 30%+[11]) and the mainframe's rich margins fund a fortress balance sheet and a 30-year dividend record. The weak link is Consulting: a third of revenue that grew essentially zero in real terms.

IBM's competitive position is bimodal. In its mainframe core it is a de facto monopoly[22]; almost everywhere else it is the slower-growing incumbent. Accenture out-grew IBM Consulting in 2025 (7% vs ~flat)[18], and Oracle and SAP posted faster cloud growth than IBM Software[19][20]. The strategic question is whether IBM's lock-in and hybrid pitch defend its base faster than rivals erode it.

The competitive set, by layer

Porter's Five Forces

Click each force for the rated pressure and the evidence behind it.

Positioning: breadth vs. growth

Hover a point for the basis. IBM owns one of the broadest enterprise stacks — software, services, hardware and financing — but sits below cloud-native peers on growth and AI momentum.

IBM's revealed strategy is "own the enterprise plumbing, not the frontier model." Its sources of defensibility are infrastructural rather than glamorous: Red Hat/OpenShift as the hybrid-cloud fabric, a mainframe monopoly, and deep switching costs. Its differentiated AI bet is that small, open, cheap models (Granite) running on a customer's own data beat giant general models for enterprise work[23] — and CEO Krishna is unusually willing to say the hyperscalers' spending will not pay off[25].

The moats

• Red Hat / OpenShift. The open hybrid-cloud layer that runs across every cloud and on-prem — OpenShift ARR reached $1.9B, up 30%+[11]. It is the strategic center of gravity post-2019.
• The mainframe. A protected franchise no rival contests, throwing off cash and now positioned for on-prem AI inference via the z17's Spyre accelerator[7].
• Switching costs & trust. The world's largest banks, insurers and governments run mission-critical workloads on IBM; migrating them is risky and expensive.
• M&A muscle. The $6.4B HashiCorp acquisition (closed Feb 2025) added Terraform and Vault to the automation portfolio[24].

The AI thesis: small is the new smart

Rather than spend tens of billions chasing a frontier model, IBM open-sources its Granite family and pitches them as far cheaper to run — it cites figures like 98% cheaper than GPT-4 Turbo and 75% cheaper than Llama 3 70B for some enterprise uses[23]. The logic: enterprises want models they can govern, tune on private data and run on their own infrastructure, where cost-per-inference and control matter more than benchmark supremacy. IBM frames this as a coherent, differentiated position — and also, critics note, a comfortable story for a company that could not win the frontier race anyway.

The contrarian: Krishna on the AI build-out

IBM's strategy is wrapped in an explicit bet against the hyperscalers' capital intensity. In December 2025, on The Verge's Decoder podcast, Krishna argued the industry's ~$8 trillion of planned AI-data-center capex simply cannot earn its return[25]:

For IBM this is strategy as much as commentary: if the GPU build-out is overdone, IBM's capital-light, small- model, on-prem approach looks prudent rather than behind. If it is not — if scale keeps winning — IBM risks being a follower that talked itself out of the biggest platform shift in a generation.

IBM's generative-AI "book of business" topped $12.5B inception-to-date by year-end 2025[27] — but it is a company-defined metric blending software revenue, new SaaS contract value and consulting signings, which are ~80% of the total[27]. Signings are contracts, not recognized revenue, and IBM itself says they "take time to convert"[28]. On that composition, the better-evidenced reading is pipeline, not revenue: the durable, recurring core today is the ~$1.9B software slice[27], and the rest is a forward indicator to be marked to conversion, not counted as sales.

What the number actually contains

Per IBM's own definition, the book combines inception-to-date software transactional revenue, new SaaS annual contract value, and consulting signings tied to specific AI offerings. At Q3 2025, of a $9.5B book, consulting was "a bit over 80%" — roughly $7.6B — with software the remainder[27]. That composition is the crux: consulting signings convert to revenue only as projects are staffed and delivered, over quarters or years, and can be delayed or scaled back. As CEO Krishna put it, new AI consulting business "will take time to convert to revenue"[28].

watsonx and Granite: the product behind the number

The software side runs on watsonx (IBM's enterprise AI platform) and the open-source Granite models, which IBM embeds into partners' stacks — Oracle, AWS, Salesforce and SAP — and ships under open licenses on Hugging Face[30]. The strategy is distribution-by-embedding and openness rather than a single destination model. It is early, but OpenShift-style recurring AI software revenue is what would make the book durable rather than project-based.

The ghost of Watson

IBM's AI credibility carries scar tissue. Its previous flagship, Watson, became a textbook case of overpromising: MD Anderson Cancer Center abandoned a Watson oncology project after spending roughly $62M, and IBM was widely criticized for marketing capabilities the technology could not deliver[29]. IBM eventually sold the Watson Health data assets in 2022. None of that means watsonx will repeat the mistake — the technology and the go-to-market are different — but it is why skeptics discount IBM's AI claims more than they would a newcomer's.

IBM earned $10.6B from continuing operations on $67.5B of revenue in 2025, with diluted EPS of $11.14 (GAAP) and free cash flow of $14.7B, up $2.0B[31][15]. It guided to >5% constant-currency revenue growth and another ~$1B of free cash flow in 2026[32]. The financial profile is high-margin and cash-rich — the debate is about growth rate and valuation, not solvency.

Free cash flow trajectory

Free cash flow, US$B. The 2024 figure ($12.7B) is implied by the disclosed "$14.7B, up $2.0B"; earlier years are shown for trend context.

Profitability and capital returns

Margins moved the right way: FY2025 GAAP gross margin reached 58.2% (operating non-GAAP 59.5%), up over a point[15]. The cash funds a long-standing dividend — raised for 30+ consecutive years — IBM declared a $1.68 quarterly payout with its results, has raised the dividend for 30+ consecutive years, and has paid one continuously since 1916[33]. That makes IBM a Dividend Aristocrat and a core holding for income investors, though the flip side is a high payout that constrains reinvestment.

What the market thinks

Investors rewarded the turnaround: IBM's stock traded near record highs in mid-2026, with a market capitalization around $287B (closing price $305.63, 3 June 2026)[34]. What that price assumes, derived from the cited figures (illustrative): against $11.14 of GAAP diluted EPS[31], IBM trades at roughly 27x trailing earnings and ~19.5x its $14.7B of free cash flow[15] — a multiple that prices durable mid-single-digit growth with compounding cash, roughly the >5% IBM itself guides to[32], against 8% trailing reported growth[6]. By IBM's own history the bar is high: the lost decade ended with the market value down ~44%, implying roughly $120B[37] — less than half today's. Versus peers, the market pays IBM ~4.3x sales, between Accenture's ~1.6x ($109B on $69.7B)[36][18] and Oracle's ~11.5x ($662B on $57.4B)[35][19] — priced as more than a services firm but nothing like a cloud grower. That is the bar both bull and bear are measured against: the bull needs growth to hold the guide; the bear needs only a slide back to low single digits for 27x to look expensive.

Stress-testing the guide: what the mainframe cycle does to 2026

A simple bridge, derived here from the disclosed FY2025 segment figures (all derived numbers illustrative). Hold each segment at its FY2025 constant-currency rate — Software +9% on $29.96B adds ~$2.7B[10], Consulting flat adds nothing[12], Financing is negligible[14] — and the >5% guide (~$3.4B of new revenue on a $67.5B base[6][32]) leaves a ~$0.7B gap that Infrastructure must fill: roughly +4% in the year after a launch spike, when mainframe revenue historically recedes[13]. If Infrastructure instead gives back 10% (−$1.6B on its $15.72B base), total growth falls to roughly +1.7% — and hitting the guide would require Software to accelerate to ~16% or Consulting to break decisively out of flat.

Method: FY2025 segment bases and constant-currency rates from IBM's Q4 release[10][12][13][14]; growth target from the 2026 guidance[32]. The bridge shows the guide quietly assumes Infrastructure keeps growing through a post-launch year — the single most checkable assumption in the 2026 story.

On revenue IBM ($67.5B) sits just behind Accenture ($69.7B) and ahead of Oracle ($57.4B)[6][18][19]. But on market value the gap is enormous: Oracle is worth ~$662B to IBM's ~$287B[35], because Oracle's cloud is growing fast and IBM's top line grows in single digits. The peer set shows IBM's problem isn't size — it's growth and the multiple it earns.

Revenue, latest fiscal year (US$B)

SAP converted from €36.8B at ~1.08 USD/EUR (≈$40B), labeled approximate. Sources: IBM, Accenture, Oracle, SAP filings.[6][18][19][20]

Market capitalization (~3 June 2026, US$B)

Sources: StockAnalysis (IBM, Oracle, Accenture); SAP approximate.[34][35][36]

The benchmarking table

The biggest risks are history, conversion and cannibalization. IBM spent a "lost decade" shrinking while peers grew and leaned on buybacks[37][38]; its $12.5B AI book is mostly unconverted signings[28]; and generative AI could compress the billable consulting hours that fund ~31% of revenue[40]. None is fatal, but together they explain the market's discount.

The "lost decade" shadow

Critics anchor on the 2010s. Under CEO Ginni Rometty, IBM's revenue fell roughly 28% and its market value about 44% (~$95B) while cloud rivals compounded[37]. Detractors argue leadership pursued an "EPS roadmap instead of a product roadmap" — spending around $55B on buybacks and adding ~$34B of debt for Red Hat rather than out-innovating[38]. Bulls counter that the Krishna-era IBM is a different company with real software growth; bears see the 2025 surge as cyclical and ask whether the old habits are truly gone.

The AI it sells could hollow out the AI work it bills

IBM Consulting is roughly a third of revenue, and much of consulting's value is human hours. Generative AI — IBM's own product — threatens exactly that model: if AI automates application modernization and integration, the billable hours shrink. Accenture is already converting AI demand into bookings faster ($5.9B of GenAI bookings, AI revenue tripling)[40], raising the risk that IBM is disrupted in consulting even as it sells the disruptor.

SWOT

None of the bear case says 2025 was fake; it says 2025 was the easy part. Here is where the evidence in this study actually lands on the four questions it opened with — and what would change each call.

The weighing

On whether the $12.5B AI book is real revenue or a slide number: the evidence leans toward pipeline, not revenue (high confidence). The controlling evidence is that ~80% of the book is consulting signings[27] and IBM's own caution that new AI business "will take time to convert to revenue"[28], which outweighs the headline's sheer size because signings recognize only as projects are staffed and delivered, and can be descoped before they ever become sales. The strongest surviving counter-argument: the ~$1.9B software slice is recurring, and embedding Granite and watsonx into Oracle, AWS, Salesforce and SAP gives it distribution most rivals lack[30]. What would flip this reading: the software share of the book rising decisively above ~20% at the Q4 2026 results (late January 2027); a consulting book-to-bill falling below 1.0 (from 1.03[12]) would harden the bear case instead. Pre-mortem: if this looks wrong in two years, the most likely reason is that signings converted into recurring software pull-through faster than the composition suggested — or, on the other side, that clients quietly descoped AI pilots and the book stopped growing altogether.

On whether the two-speed portfolio keeps accelerating: the evidence leans toward deceleration from 8% toward the >5% guide (medium confidence). The controlling evidence is that the +67% Q4 mainframe surge is a z17 launch-cycle spike[13] and that Consulting — about 31% of revenue — was flat at constant currency[12], which outweighs Software's +11%[10] because the Infrastructure comparison turns hostile in exactly the year the guide must be met. The strongest surviving counter-argument: OpenShift ARR of $1.9B growing 30%+[11] says the recurring engine is compounding underneath the cyclical noise. What would flip this reading: Infrastructure posting growth again in FY2026 despite post-launch comparisons, at the Q4 2026 report; or Consulting reaching 3% or better constant-currency growth in the same release. Pre-mortem: if this looks wrong in two years, the most likely reason is that on-prem AI inference turned the mainframe's launch spike into a new baseline — or, on the other side, that consulting tipped from flat into outright decline.

On whether hybrid-cloud-plus-AI is a moat: the evidence leans real but defensive (medium confidence). The controlling evidence is the mainframe's de facto monopoly — its best launch in company history[22] — and OpenShift's $1.9B ARR growing 30%+[11], which outweighs the "follower's rationalization" critique because both are recurring, uncontested positions a hyperscaler cannot replicate by outspending. The strongest surviving counter-argument: ceding hyperscale handed rivals the fastest-growing layer of the market[8], and "open and cheap"[23] defends an installed base rather than expanding one. What would flip this reading: OpenShift ARR growth slipping below ~20% in the FY2026 disclosures; or Red Hat growth falling out of the high single digits in successive quarterly releases[11]. Pre-mortem: if this looks wrong in two years, the most likely reason is that small, governed models won enterprise AI outright and the moat proved expansionary, vindicating Krishna's capex skepticism[25] — or, on the other side, that scale kept winning and IBM's restraint read as absence.

On durable turnaround versus false dawn: contested — leaning durable but modest. What deadlocks it is genuinely symmetric evidence: record $14.7B free cash flow and 11% software growth on one side[15][10]; on the other, a company that shrank ~28% through a lost decade after earlier convincing-looking recoveries[37], while its closest services rival converts AI demand into bookings far faster[18]. The controlling evidence, such as it is, is cash: $14.7B of free cash flow, guided ~$1B higher[32], buys IBM the time and optionality no false dawn of the 2010s had. The strongest surviving counter-argument: generative AI could compress the billable hours behind roughly a third of revenue faster than software grows[40]. What would flip this reading: FY2026 free cash flow landing below the ~$15.7B the guidance implies[32] at the January 2027 report; or Consulting revenue declining outright for two consecutive quarters. Pre-mortem: if this looks wrong in two years, the most likely reason is having underestimated how fast recurring software compounds — or, on the other side, having mistaken a mainframe cycle plus an unconverted pipeline for a structural re-acceleration.

Method

Research proceeded by fan-out web search across nine question areas (overview, market, business model, competition, strategy, the AI book, financials, peers and risks), followed by direct fetching of primary and reputable secondary sources — the URLs cited here were opened and read during the research run, not merely linked. Each claim was then transcribed into a structured manifest tagging it with a source tier (Tier 1 primary, Tier 2 reputable secondary, Tier 3 tertiary), a confidence level, and a stance (supporting, critical, or neutral). For IBM the load-bearing figures are the disclosed FY2025 results — $67.5B revenue, $10.6B income from continuing operations, $14.7B free cash flow, and the four-segment split — taken from IBM's fourth-quarter earnings release; the rest of the study is built around verifying and contextualizing those.

Frameworks used

The analysis applies Porter's Five Forces to read IBM's industry structure, a breadth-versus-growth positioning map to place it against software and services peers, peer benchmarking on revenue and market value, a SWOT, and a case-for / case-against ledger in every section so strengths and weaknesses get equal scrutiny. These frameworks organize the evidence; the weighed judgments themselves sit in the closing weighing of Risks & Skeptics. A formal valuation was deliberately skipped: IBM's value hinges on contested judgments — how much of the AI book converts, whether the mainframe lift persists — that a single intrinsic-value number would obscure rather than clarify.

Disclosed vs. estimated

Disclosed figures — the FY2025 revenue, segment splits, margins, free cash flow, EPS and 2026 guidance — come straight from IBM's fourth-quarter release and SEC filings. Two categories are flagged as softer. First, the generative-AI "book of business" is a company-defined metric with no third-party standard: it mixes recognized revenue with SaaS contract value and consulting signings (~80% of the total), and is a forward pipeline indicator, not booked revenue. Second, the multi-year revenue and free-cash-flow trajectory charts show prior-year figures for trend context; only the FY2025 endpoints (and the disclosed 2024 free cash flow, implied by "$14.7B, up $2.0B") are directly sourced here. Peer revenue and market caps are from each company's filings and market-data aggregators on the dates noted. Source mix: supporting 17 · critical 13 · neutral 10; 12 Tier 1 primary, 16 Tier 2 reputable secondary, 12 Tier 3 tertiary. Because IBM is an anglophone US company, all sources are English-language.

Neutrality & independence

Each section deliberately pairs the case for and the case against — the bull view (re-accelerating software, mainframe strength, record cash, a coherent AI strategy) and the bear view (flat consulting, an unconverted AI book, a history of overpromising and financial engineering) both get full argument — and the closing weighing in Risks & Skeptics then states where the evidence leans on each of the four decisive questions, at what confidence, and which checkable metrics would flip the call. It is not investment advice and is not affiliated with, sponsored by, or endorsed by IBM. It is a point-in-time artifact as of 4 June 2026; IBM's results, market value and AI metrics move fast, and figures will age.

Full bibliography with tiers, stance, and links on the Sources page.

Cross-checked at build time by an automated link checker. A few primary sources (SEC EDGAR filings) bot-wall automated fetchers and were verified manually against the filing contents. See Methodology & Limits.