Reading Time: 48 minutes

---

On July 14, IBM released preliminary second-quarter results a full week ahead of schedule. Revenue up 1 per cent. Software up 5 per cent. Infrastructure down 7 per cent as the strongest mainframe launch in the company’s history wrapped, and a candid admission that numerous large deals had slipped past the quarter boundary. For this, the market handed IBM the sharpest single-day fall in its modern record, erasing roughly a quarter of the company’s value, and within hours the commentary industry had declared that software was dying and IBM with it.

We have seen this film before, projected in reverse. In September 2025, Oracle announced a reported $300 billion compute contract with OpenAI, its shares had their sharpest day since 1992, and the market read a backlog as a bank balance. We wrote that week, in Backlog Isn’t Bank Balance, that theatre is not delivery, and that the applause was premature. Roughly half a trillion dollars of erased market value later, that caution needs no defending.

The market has now made the same mistake twice in ten months, in opposite directions: it priced Oracle’s applause as fact, and it is pricing IBM’s panic as fact. Euphoria, it has learned, has a shelf life. This note argues that panic has one too.

So before joining either the festival or the funeral, look at what actually moved, because there are quarters that reveal a company, and there are quarters that reveal its customers. This one revealed the customers. In the final weeks of June, technology buyers did something enterprise budgets are not supposed to do: they moved mid-quarter, at pace, redirecting capital into scarce hardware ahead of expected price increases. The behaviour was rational, defensive, and large enough to reorder vendor pipelines across the industry. IBM, standing directly in the path of that reflex, absorbed the visible damage and disclosed it early.

This note is not about the share price; that is not our trade, and it was not our trade in September either. It is about what the buyers did, why they did it, what IBM’s own disclosure actually says, and why the sum of that evidence leaves Greyhound Research more settled in its position on IBM’s franchise, not less. It draws on three bodies of evidence: IBM’s preliminary disclosure, published as Arvind Krishna’s letter to investors; the supply-side and buyer record around the event, including Greyhound CIO Pulse 2026; and our own published coverage, of IBM across more than a decade, and of this market’s moods across both of its recent extremes.

First, A Word On The Word Bullish

Greyhound Research are industry analysts, not equity analysts. We do not make share price calls, we do not publish targets, and we do not advise on positions. When we say we remain bullish on IBM, we mean bullish on the franchise: on the construction of its portfolio, on the durability of its demand, on the structural adaptability it has demonstrated across four decades of platform transitions, and on its relevance to the technology buyers we advise. Readers seeking an investment view should look elsewhere. Readers seeking a demand-side view, grounded in what enterprises actually do with their budgets, should read on.

What The Letter Says, And What The Tape Did

Strip away the noise and the preliminary print divides into three clean parts.

The first part is growth where the market claims there is none. Total revenue rose 1 per cent. Software rose 5 per cent, with Red Hat accelerating sequentially to 11 per cent growth and both HashiCorp and Confluent described as strong performers. Consulting revenue was flat, up 1 per cent at constant currency, while consulting signings continued to grow, led by GenAI engagements. Apply the same rule here that this note applies to backlog, because signings are not revenue either. Flat consulting revenue against growing signings is a conversion problem, and it is the one weakness in this print Greyhound Research will not attribute to June: we documented the friction in August 2025, in procurement cycles, in pricing and go-to-market alignment, and in delivery timelines running longer than buyers expected. Operating pre-tax margin expanded 30 basis points despite the shortfall, and operating earnings per share rose 5 per cent to $2.93 even as GAAP diluted earnings slipped 2 per cent to $2.27. These are not the vital signs of a broken model.

The second part is the miss, and it is unusually concentrated. Infrastructure revenue fell 7 per cent against IBM’s own expectation of a low-single-digit decline as the z17 launch cycle wrapped. The shortfall sat in the Z programme and the software stack attached to it, primarily Transaction Processing. Krishna’s letter locates the cause precisely: in the last weeks of June, clients shifted quarterly capex toward servers, storage, and memory to secure supply-constrained infrastructure ahead of expected price increases, at a magnitude IBM did not anticipate; clients were simultaneously distracted by fast-moving, industry-wide cybersecurity concerns; and numerous large deals failed to close on expected timelines, which the letter says drove the majority of the shortfall.

Even the shape of the margins corroborates the concentration; a miss landing on the richest shelf of the portfolio compresses gross margin even as revenue grows, which is exactly what the print shows: operating gross margin of 59.4 per cent, down 70 basis points, and GAAP gross margin of 57.7 per cent, down 100 basis points. The letter’s most quoted admission is two words long: “we faltered”. Note what the admission covers. Timing and adaptation. Not demand. And note that this is not the letter’s word taken on trust: coordination, delivery tempo, and go-to-market friction are precisely the weaknesses Greyhound Research has been documenting in this company since August 2025, across four continents. July 14 is that failure mode arriving at scale, in public. And note the posture that follows it, stated in the letter’s own words: “These are not excuses, but they are realities,” with the company’s job recast as helping clients through uncertainty rather than explaining itself out of a quarter.

Table 1 lays out the print in one view, exactly as disclosed and nothing more.

The third part is what the tape did with the first two. IBM closed down 25.2 per cent, its worst single day on record, deeper than Black Monday 1987, while the broader indices rose and memory, hardware, and cybersecurity names rallied. The market did not sell enterprise technology on July 14. It re-ranked it, punishing whatever sat adjacent to the scarce layer of the stack and rewarding whatever constituted it. The scale of the reaction, a quarter of the company against a revenue shortfall of well under a billion dollars, tells you what was being priced: fear, not arithmetic. And beneath the fear sits the event this note is really about. The enterprise technology budget is being rebuilt in real time, and July 14 is what first contact with that rebuild looks like when it lands on a single vendor’s tape.

Greyhound Standpoint: At Greyhound Research, we believe the market priced three fears on July 14, not one quarter: the fear that hardware scarcity permanently outranks software in enterprise budgets, the fear that IBM’s mainframe economics have entered structural rotation, and the fear that management no longer sees its own demand clearly. The first fear misreads how enterprises operate. The second is a real question with, so far, contrary evidence. The third was answered by the letter itself, which chose candour over choreography.

Ten Months Ago, We Made This Argument In Reverse

Anyone can be contrarian. The test of a method is whether it points the same way when the mood inverts, so the September episode deserves fuller detail than the opening gave it. Oracle had announced a reported $300 billion, five-year compute contract with OpenAI. The stock rose nearly 40 per cent in a single day. Larry Ellison briefly became the world’s richest man. Remaining Performance Obligations had soared to $455 billion, up 359 per cent. The choreography was flawless, and the market treated the backlog as though it were cash in the bank.

We at Greyhound Research declined the champagne. Our argument then was that theatre is not delivery; that RPO is an accounting device, conditional and renegotiable; that the $300 billion was not an iron-clad cheque but a ceiling of possibility, at best a promissory note. We wrote that Oracle had been paid in applause for a banquet that had yet to be cooked. We flagged four risks in a specific order: concentration on a single unprofitable client, contract fragility, execution across 5 gigawatts of capacity that has to be built and powered, and a valuation implying flawless delivery. And we noted, dryly, that when narrative races ahead of fundamentals, corrections are rarely gentle.

The correction has not been gentle. Since that September peak, Oracle’s market value has fallen by roughly half a trillion dollars. The stock sits near 52-week lows, down some 62 per cent from its high. S&P Global has cut the credit rating to a step above junk. Capital expenditure has climbed toward $90 to $95 billion, funded by tens of billions in new debt and equity, and the company is cutting around 21,000 roles. Every one of the four risks we named has arrived, in the order we named them.

And look closer at the OpenAI contract itself, because it remains the era’s cleanest specimen of commitment mistaken for assurance.

At Greyhound Research, we noted at the time that the reported $300 billion was a ceiling, not a cheque: staged usage ramps, undisclosed rate cards, renegotiation gates, and a counterparty spending far beyond its income. The record since has stress-tested every seam. OpenAI’s arrangement with Microsoft has itself been renegotiated, in a restructuring that paired an equity stake of roughly 27 per cent with a contracted purchase of some $250 billion in incremental Azure services. OpenAI’s projected cash burn runs to roughly $27 billion this year and $63 billion next, losing by one estimate around $1.22 for every dollar it earns, with cash-flow positivity not expected until about 2030.

More than half of Oracle’s backlog is estimated to hang on this single client, Oracle’s debt has passed $100 billion, and Bloomberg has reported delivery slippage at some sites toward 2028, which Oracle denies. The commitment has not broken. But a promise whose keeping depends on a loss-making counterparty, disputed timelines, and continuous access to capital markets is precisely what we meant by commitment, not assurance. It is the same test we apply to IBM’s own backlog later in this note, because intellectual consistency is the whole point.

Now hold July 14 next to September 2025 and look at what the two events share. Both were single-day market verdicts on a company’s relationship with the AI infrastructure wave. Both compressed a multi-year question into a single session’s emotion. Both were wrong in the same way, and only the direction differed: Oracle’s backlog was read as money it was not, and IBM’s deferral has been read as destruction it is not. The market mistook a moment for a trajectory, twice, in opposite directions, inside ten months.

This is why our position is not contrarianism. Contrarianism is a reflex against consensus. What we practise is duller and more useful: check the bill. In September 2025 the bill was hiding behind applause, so we counted the debt, the gigawatts, and the client’s ability to pay. In July 2026 the bill is hiding behind panic, so we count the arithmetic of hoarding, the sunset of ownership, the operationalisation clock, and the company’s record of structural adaptation. Same discipline. Opposite mood.

Our buyers have been ahead of both readings, and the distance is widening. When we made the Oracle call, Greyhound CIO Pulse 2025 found 62 per cent of enterprise technology leaders dismissing mega-announcements as more optics than delivery. Greyhound CIO Pulse 2026, across more than 1,000 leaders, finds 82 per cent. Between the euphoria and the panic, buyer scepticism did not soften; it deepened by twenty points. That scepticism was correct about Oracle’s mega-announcement. It is equally correct about IBM’s mega-selloff. A market that overreacts to good theatre will overreact to bad theatre, and CIOs, who have to live with the consequences of both, discount each accordingly.

Step back once more, because underneath the rotation sits the largest force in this note, and it deserves to be said without decoration. This is a budget story wearing a stock story’s clothes. The enterprise technology budget, the actual document boards approve and chief information officers defend, is being reorganised in real time.

A line item that did not exist as a first-class entry two years ago, AI capacity with its memory, power, and silicon attached, now sits at the top of the stack and pulls funding toward itself mid-quarter. Line items that were untouchable for a decade, application renewals, steady-state software, standing services, are being interrogated one by one. And the reallocation no longer waits for planning season. June proved the point at scale: demand large enough to bend a Fortune 10 vendor’s quarter moved in under three weeks. Budget season is becoming continuous reallocation, and every operating rhythm built on the annual cycle, vendor account plans, renewal calendars, board reviews, now runs against a clock it was not designed for.

The agenda changed with the arithmetic. Greyhound CIO Pulse 2026 has the sequence in the buyers’ own numbers: 84 per cent of enterprise technology leaders have shifted their focus from AI adoption to AI accountability, 55 per cent name energy as the binding constraint on scale, and fewer than one in three can yet point to measurable outcomes across a majority of funded AI initiatives. Put the three together and the new agenda writes itself. Concentrate spend on the layer in shortest supply. Interrogate everything adjacent to it. Fund what converts, defer what does not, and defend every deferral with a return date. That is not a sentiment cycle. That is the operating model of the technology budget being rebuilt, and it will outlast every quarter it touches.

At Greyhound Research, our read is blunt: every technology vendor’s revenue line is now downstream of a budget being rebuilt around a different question, and July 14 made IBM the first large vendor to disclose contact with that rebuild in public.

Ericsson warned on the same economics the same morning. The vendors that compound through this era will be the ones whose products answer the budget’s new question, feed, secure, or govern the layer in shortest supply, and whose commercial machinery can move at the budget’s new tempo. That is the lens for every result this earnings season, and it is the lens this note has applied to IBM.

Greyhound Standpoint: At Greyhound Research, we believe the AI era has made markets structurally poor at reading enterprise technology. They price applause and panic at the same volume, because both are legible in a single session, while the things that actually decide outcomes, conversion, integration, delivery, and time, are not. Our method does not change with the mood. When Oracle was toasted, we checked the bill. Now that IBM is being mourned, we are checking the same bill.

This Was Never Only About IBM

Zoom out from Armonk and the pattern becomes unmistakable. July 14 was not an event; it was the sharpest single frame of a film that has been running all year.

Wall Street has spent 2026 conducting a rolling repricing of the entire enterprise software complex, branded the SaaSpocalypse, which has erased by one count roughly $2 trillion of software market value since January on a single fear: that AI agents will dissolve the per-seat licensing model. In February, a set of frontier model releases triggered a selloff approaching a trillion dollars across enterprise software within days. Salesforce has fallen more than 40 per cent over twelve months. Workday is down roughly 45 per cent this year and more than 60 per cent from its early-2024 peak.

By mid-April, HubSpot, Atlassian, and Figma sat 70 to 80 per cent below their 52-week highs, the sector’s benchmark ETF had lost 27 per cent in six months, and forward multiples across enterprise software had compressed from roughly ten times revenue to six.

Two details in that record matter more than the totals. The first: the verdicts no longer respond to performance. In April, ServiceNow beat expectations on both revenue and earnings and fell 18 per cent the next day; IBM beat in the same week and fell alongside it. When beating the numbers produces the same outcome as missing them, the market is not pricing companies. It is pricing a category, on mood. The second: on July 14 itself, IBM’s preannouncement dragged Microsoft down 3 per cent in pre-market trading and pushed Oracle lower still, though neither had reported anything. Contagion without information is the definition of sentiment.

Set the tape against the buyers, and the gap this note has described all along reappears at sector scale. Greyhound CIO Pulse 2026 shows enterprises interrogating their software estates, not abandoning them; renewals are being examined, consolidated, and made to prove indispensability, which is a demand for outcomes, not an exit from software. And in the very latest sessions the tape itself has begun to blink, with capital rotating from the infrastructure layer into oversold application software, the earliest market acknowledgement of the second act this note describes, in which the money moves to whatever makes expensive capacity produce.

So read July 14 for what it was: not an IBM story, but the most violent chapter yet in a market-wide failure to distinguish between software that feeds, secures, and governs the scarce layer and software that merely sits beside it. That distinction will be drawn, vendor by vendor, over the coming quarters, and it will be drawn by buyers rather than by the tape.

Every chief executive of an enterprise software company, at Oracle, at Microsoft, at Workday, and far beyond, is sitting the same examination IBM sat on July 14. The playbook in this note, candour paired with conversion, commercial machinery rebuilt for the compressed clock, and outcomes proven rather than asserted, applies well beyond Armonk.

Greyhound Standpoint: At Greyhound Research, we believe 2026’s software repricing is a verdict on a category delivered before the evidence on its members is in. The dispersion will come, and it will be decided by the sorting question this note keeps returning to. Until then, single-day verdicts, whether on Oracle in September, on ServiceNow in April, or on IBM in July, tell you about the market’s nerves, not about enterprise demand.

Where We Stood Coming Into July 14

The word remains in our title is doing real work, because the position predates the panic, and so does the warning.

Our Q3 2025 note on IBM, published in November, carried a title that reads differently today than it did then: Delivery Strength Holds, But Integration Demands Intensify. The quarter itself was excellent. Revenue of $16.3 billion, up 9 per cent year on year. IBM Z up 59 per cent, the strongest two-quarter mainframe launch in two decades. Red Hat bookings up around 20 per cent. And yet Greyhound Research flagged one line of weakness inside all that strength: Transaction Processing fell 3 per cent, and it fell because clients were redirecting spend toward newer Z hardware.

Read that again, because it was published nine months before July 14. The precise mechanism that broke this quarter, clients moving money toward hardware and away from the software attached to it, was visible then, named then, and treated then as a footnote inside a strong print. It was not a footnote. It was the tell.

We drew the consequence in the same note. IBM’s ability to convert hardware momentum into recurring software monetisation, we wrote, remains the hinge between short-term strength and long-term valuation. That hinge is the whole story of July 14. The market has now discovered, violently, what our Greyhound Fieldnotes had already recorded: the Z cycle and the software attached to it do not move in lockstep, and the gap between them is where a quarter can be lost.

The rest of that note still stands. Greyhound Fieldnotes showed enterprise buyers treating IBM as an operational partner rather than a supplier of components, with Red Hat as the backbone of hybrid-cloud design and HashiCorp folded in as a control layer across clouds. Buyers had stopped assessing Red Hat, watsonx, and Consulting in isolation and started judging IBM’s ability to coordinate them into one experience. One CIO gave us the tension in a sentence: IBM’s architecture is integrated; its delivery still isn’t.

And we closed with a bar. IBM has proven it can execute; the question is whether it can make that execution feel effortless to the clients who depend on it.

July answered that honestly. Not yet. A quarter in which numerous large deals slip past their expected close is the opposite of effortless. But note what July did not overturn: not the system-buying pattern, not the operational-partner standing, not the architectural bet. Those remain live, and the preliminary disclosure, read closely, supports them.

The immediate run-in sharpens the point. In the first quarter of 2026, IBM grew revenue 9 per cent to $15.92 billion, with software up 11.3 per cent and infrastructure up 15.2 per cent on z17 momentum, and management called generative AI in mainframe modernisation accretive to the portfolio. Ninety days separated that print from this one. When a customer base reverses that fast, the prime mover is rarely the vendor. It is the market underneath the vendor.

The Anatomy Of The Miss

Greyhound Research’s assessment weighs the causes of the quarter in the following order.

Product cycle first. IBM itself guided to infrastructure decline as the z17 launch wrapped, because that is what mainframe programmes do; they surge and they settle. The surprise was magnitude, not direction. Critically, the programme’s health indicators remain exceptional, and both figures are IBM’s own disclosed numbers: z17 is tracking nearly 130 per cent of z16 programme-to-programme, and z16 was the strongest programme in IBM’s history; clients representing 85 per cent of installed MIPs are maintaining or growing capacity. A collapsing franchise does not produce those numbers. A franchise whose buyers postponed a quarter does.

Scarcity second. The late-June capex shift IBM describes is corroborated across the supply chain. Micron has disclosed that customers have pre-committed $22 billion under multi-year supply arrangements to lock memory capacity, with tightness expected through at least 2027. SK Hynix’s chief executive has warned of the worst memory shortage on record arriving in 2027, with demand outstripping supply into the next decade. Dell expects AI server revenue to more than double in fiscal 2027. Even Ericsson has flagged memory-driven cost pressure spreading into adjacent equipment, a warning it issued on July 14 itself, with its shares falling nearly 12 per cent the same day.

When the layer in shortest supply begins to reprice, procurement stops behaving like strategy and starts behaving like inventory management. That is what IBM’s clients did in the final weeks of June. It is rational, it is defensive, and it is, by construction, temporary; a pull-forward borrows from future quarters, it does not manufacture new ones.

Execution third. The letter concedes the point without decoration, and we take the concession at face value. IBM misjudged the speed of the reprioritisation and let deals drift past the quarter boundary. This is a real failure, and it belongs to management. It is also the most fixable item on the list, and the only one entirely within IBM’s control.

Distraction fourth. The letter’s reference to industry-wide cybersecurity concerns is the least examined line in the disclosure and, we suspect, one of the more consequential. Enterprises confronting fast-moving vulnerability exposure do not sign large platform deals that month. They triage. The same dynamic, as we discuss below, also handed IBM its most interesting new product of the year.

The Shortage Itself Has A Shelf Life

The June panic priced scarcity as a permanent condition. At Greyhound Research, we read it differently: two forces already visible in the record argue that even the scarcity is on a clock.

The first is that the announced supply was never going to arrive as announced. Bloomberg reported this spring that of the roughly 12 gigawatts of US data centre capacity slated to come online in 2026, only about 5 gigawatts is actually under construction, with between a third and a half of planned projects expected to slip or die. The forward pipeline is thinner still: 21.5 gigawatts announced for 2027 with only 6.3 broken ground, and 37 gigawatts planned for 2028 to 2032 with just 4.5 under construction. The constraint has migrated from chips and capital to the physical electrical layer; transformers and switchgear that took two years to procure before 2020 now take as long as five.

And the opposition has become an institution of its own. NBC News has reported at least 75 projects worth roughly $130 billion blocked or delayed in the first quarter of 2026 alone, as much as in the whole of 2025, with active local opposition groups more than doubling to 833 across 49 states, at least 69 local moratoriums in force, and Seattle imposing a one-year pause.

Read those numbers with the discipline this note has already used twice. Announced gigawatts are the infrastructure world’s promissory note. Texas makes the point at grid scale: by ERCOT’s own mid-2026 count, more than 438 gigawatts of proposed demand sit in its connection queue, nearly 90 per cent of it from data centres, against an all-time peak demand record of 85.5 gigawatts, and simple arithmetic makes the large majority of that queue phantom. The demand curve that justified the hoarding was drawn, in part, from announcements that will never be energised.

And the constraint is not an American peculiarity; the world’s other data centre capitals wrote this chapter first. In Dublin, where data centres already draw more than a fifth of the country’s electricity, an effective four-year moratorium on new grid connections ended in December only by becoming a mandate: new facilities must bring matching generation of their own, site where the grid can bear them, and tie their demand to Irish renewables. The Netherlands confines new hyperscale builds to a handful of designated sites, and Singapore, which froze new data centres outright in 2019, now admits only the most efficient through a competitive gate. Wherever AI capacity meets a real grid, the grid writes the terms.

The second force sits on the consumption side, and it is quieter but heavier. Greyhound Research believes AI’s claim on enterprise technology budgets is expanding faster than the evidence of measurable business impact. Greyhound CIO Pulse 2026 finds that just 32 per cent of enterprise technology leaders say a majority of their materially funded AI initiatives have delivered measurable business outcomes against their approved business cases, based on responses from more than 1,000 leaders. While this represents an improvement from 25 per cent in 2025, it still means fewer than one in three technology leaders can point to measurable outcomes across most of their funded AI portfolio. Most enterprises have yet to move the bulk of their AI experiments into production, and a meaningful share of the agentic projects being stood up today will not survive contact with their own economics. The US Federal Reserve’s own monitoring records the same picture from the outside: firm-level adoption growth decelerated through 2025, and a plurality of workers use AI for an hour a week or less.

Usage is real and rising; it is simply rising slower than the capacity being provisioned for it. When consumption lags provisioning, the arithmetic ends one way: capacity hoarded against scarcity becomes capacity sitting under-used against interrogation.

Hold the two forces against the supplier warnings quoted earlier in this note, and the honest conclusion is uncomfortable for everyone with a confident forecast. The supply side warns of the worst memory shortage on record into 2027. The demand side shows announced capacity evaporating and usage trailing budgets. The market cannot decide whether the decade’s defining problem is scarcity or glut, and it has priced both verdicts, with total conviction, inside twelve months. That is not a market reading the future. That is a market reading its own mood.

We at Greyhound Research do not pretend this cuts only one way for our thesis. If consumption disappoints for long enough, some of the operationalisation demand we describe softens with it. But note what under-used capacity does to the outcomes economy: it intensifies the interrogation. Idle racks are the most visible unproven dollars in the enterprise, and boards do not respond to idle capacity by buying more of it; they respond by funding whatever makes it produce. Scarcity pays the hardware layer. Glut pays the layer that makes hardware answer for itself.

Greyhound Standpoint: At Greyhound Research, we believe the scarcity that broke IBM’s quarter is real today and fragile tomorrow. Between capacity that will not be built and consumption that has not kept pace, the squeeze that triggered June’s hoarding has a shelf life of its own, and every quarter it shortens, the case for reading July 14 as a permanent verdict weakens further. Panic has a shelf life. So does the shortage that caused it.

The Compression Of Transitions

Step back from IBM for a moment, because July 14 disclosed something about the industry that matters to every chief executive who reads this note.

There have always been two ways a platform transition could arrive in enterprise technology: slowly, across years of eroding quarters, or suddenly, in a single trading day.

Until this month, the industry had only ever experienced the first kind. The cloud transition registered against IBM as twenty-two consecutive quarters of shrinking revenue, and the bleed was slow for a structural reason: rotating to cloud demanded migration. Applications had to be replatformed, teams retrained, operations rewritten. That friction metered the pace of budget rotation, and in doing so it gave the incumbent years to respond. IBM used those years to remake itself.

The AI rotation carries no such friction. Securing scarce infrastructure requires no migration programme, no replatforming, no retraining. It requires a purchase order. Which is why a few weeks of buyer behaviour in June could move a $17 billion quarter, where the equivalent rotation in the cloud era took half a decade to register. The friction that once protected every incumbent’s response time has been removed from the system.

The implication is uncomfortable and general. Response windows in enterprise technology have collapsed from years to weeks. Strategy cycles, deal desks, supply sensing, pricing authority, and disclosure practice were all built for the slower clock. July 14 is what happens when a company, even a well-run one, meets the new clock with the old machinery.

For IBM specifically, the question July posed is whether its celebrated structural reflexes now operate at compressed speed. The evidence is split, and instructively so. The product reflex demonstrably does: weeks separated the arrival of new frontier AI capabilities from Lightwell’s general availability, a sequence the letter itself sets out. The commercial reflex, by the letter’s own admission, does not yet; the deal desk lost races it should have won. The remediation programme, which we address directly below, must close precisely that gap.

Six Arguments For Staying Bullish

The bearish reading of July 14 asks you to believe that one quarter’s timing reveals a franchise’s trajectory. We hold the opposite view, for six reasons, none of which is sentiment. Each argument that follows rests on disclosed numbers, observed buyer behaviour, or two decades of watching this company restructure its way through platform shifts. Each is falsifiable, and each carries its breaking condition in the July 22 watchlist later in this note. Together they describe a company standing in the path of the market’s next spending phase, not at the end of its last one.

One. The arithmetic of hoarding. There is only so much hardware an end-user enterprise can buy, and the buyers who moved IBM’s June were end-user enterprises, not hyperscalers or neoclouds building capacity for resale. Budgets cap it. Data centre space and power cap it. Depreciation schedules and chief financial officers cap it. A capex pull-forward is, by definition, borrowed demand; the racks purchased in June are the racks not purchased in December. Hoarding is a one-time repositioning of inventory posture against a shortage, not a new demand curve. The market extrapolated a reflex into a regime.

Two. The sunset of own-and-operate. The June rush ran directly against the decade-long direction of enterprise operating models. The era in which technology decision makers wanted to own and operate large hardware estates is on aggressive sunset. In Greyhound Fieldnotes, CIOs describe the June purchases in the language of insurance, not conversion: capacity secured against price rises, for infrastructure they have no intention of running the old way. A purchase made against the secular current is a hedge. Hedges do not found eras.

Three. The board reset and the outcomes economy. AI has rewritten what boards expect of their organisations. Every enterprise now carries an instruction that borders on the existential: adapt the business model, almost overnight, or become the technology’s victim. That instruction is precisely what sent buyers racing for capacity. But the same boards are enforcing accountability for every dollar at levels this industry has never seen, and our own buyers told us so before the market noticed.

Greyhound CIO Pulse 2026, across more than 1,000 enterprise technology leaders, finds 84 per cent have shifted their focus from AI adoption to AI accountability, up from 78 per cent a year earlier, concentrating on how AI integrates with existing systems of record and whether it can operate under the same governance standards that regulate their core data.

That is the whole thesis of this note, stated by the buyers themselves, nine months early. Spending is running ahead of proof, and boards know it. Capacity bought under pressure will be interrogated under pressure.

Four. The operationalisation clock. A rack produces nothing on arrival. Integration, security, and governance come first; data engineering, skills, and amperage follow; the whole takes quarters, and quarters are exactly what CIOs do not have. The board clock is set to months. Greyhound CIO Pulse 2026 finds 55 per cent of CIOs citing energy cost and availability as their single biggest barrier to AI scale, up from 47 per cent a year ago; the constraint has not only moved from the purchase to the deployment, it is tightening. Buying the rack was never the hard part.

That gap closes only one way: with software, platforms, and partners. This is not our hope; it is the industry’s revealed behaviour. Demand is concentrating around targeted AI investments rather than broad programmes. Tata Consultancy Services is building a team of up to 8,900 forward-deployed engineers to implement AI inside client environments. Cognizant is acquiring Astreya to deepen AI infrastructure capability. The services industry is staffing for the operationalisation wave because the operationalisation wave is where the money goes next. IBM’s core, hybrid platform software plus consulting plus infrastructure attach, sits directly in its path.

Five. The print already shows rotation, not ruin. The most under-read paragraph of Krishna’s letter is the strength list. Distributed Infrastructure posted the best performance in its reported history, up 37 per cent with strong growth in Power and Storage, against 8 per cent growth in the same business in Q3 2025. That is not a drift; it is an inflection.

The late-June scramble did not merely bypass IBM. A meaningful share of it landed inside IBM’s own distributed portfolio, alongside Red Hat accelerating to 11 per cent, strong performance from HashiCorp and Confluent, consulting signings growing on GenAI, and operating margin expanding through the shortfall. The quarter’s money moved within IBM at least as much as it moved away from it.

The same paragraph reports a backlog of approximately $500 million exiting the quarter, and here we must apply our own rule to a company we are defending. Backlog isn’t bank balance. Greyhound Research wrote that about Oracle in September 2025, and the discipline does not soften because the conclusion this time is favourable. IBM’s half a billion is a promissory note, not money. It is unbanked until it converts.

What differs is the character of the promise, not its nature. Oracle’s backlog was concentrated on a single client who was not yet profitable, denominated in hundreds of billions, and contingent on building gigawatts of capacity that did not exist. IBM’s is a few hundred million, spread across existing clients on a diversified installed base, for Power and Storage systems the company already manufactures, requiring no speculative capex to fulfil. One promise needs a decade and a miracle. The other needs a supply chain and two quarters. Both, though, are promises, and neither is proof.

So we hold this evidence at exactly the weight it deserves: the backlog tells us the direction of demand, not the fact of revenue. If it ages rather than converts, our reading is wrong, and that condition sits in the July 22 watchlist below, with the others.

Six. The structural reflex. This company’s defining trait across our two decades of coverage is not any single product. It is the reflex to answer platform shifts with structure. PwC Consulting in, 2002. Personal computers out to Lenovo, 2005. SoftLayer and the public-cloud chase in, 2013; x86 servers out, 2014; the hyperscaler chase conceded and its brands retired by 2017. Red Hat in, 2019. Kyndryl out, 2021. HashiCorp and Confluent in since. Through the cloud transition, revenue shrank for twenty-two consecutive quarters, the harshest weather any large enterprise vendor of its era endured, and IBM emerged with software at its centre and a hybrid platform under it. The reflex is visibly intact in the present tense: on July 15, the day after the letter, IBM launched new Power systems and software for enterprise risk, productivity, and flexibility, fresh supply pointed straight at the demand that produced the record quarter and the backlog.

Within weeks of the introduction of Mythos, the frontier AI generation the letter itself names, resetting the terms of open source security, IBM and Red Hat stood up Lightwell: a $5 billion commitment, more than 20,000 engineers, a trusted enterprise clearinghouse for open source software vulnerabilities, generally available from July 8. The quarter’s cybersecurity headwind became the quarter’s most interesting product. That is the reflex working at modern speed. Quantum sits behind it as optionality, not load-bearing thesis: more than $10 billion committed over five years, a foundry letter of intent alongside the US Department of Commerce, spanning research and development, capital expenditure, manufacturing scale-up, acquisitions, and ecosystem, and a 2029 target for large-scale fault tolerance. The letter’s framing does not hedge: quantum computing “is no longer decades away, it is upon us”.

We do not need quantum for this case; we simply note that the company being priced for decline is the one making the decade’s most aggressive frontier commitments.

Greyhound Standpoint: At Greyhound Research, we believe the bullish case for IBM after July 14 is not a bet against the AI infrastructure wave. It is a bet on the wave’s second act. Hardware can be hoarded; outcomes cannot. In the AI era, outcomes are the only currency boards accept, and outcomes are manufactured in the layers IBM occupies: the platform, the integration, the governance, the operating discipline. June paid the scarce layer first. The quarters ahead must pay the layer that makes scarcity useful.

The Competitive Map After July 14

Start with the category error, because it explains the violence of the repricing, and because IBM has lived it before. On July 14 the market measured IBM with a yardstick built for someone else. IBM is not a hyperscaler. It is not a neocloud. It does not sell raw capacity by the gigawatt, does not win by owning the most accelerators, and the capital arithmetic quoted earlier in this note is the opposite of the capacity race. The infrastructure outcomes IBM optimises for are different in kind: transaction integrity at transaction latency, sovereign and regulated compute, estates that must be governed and audited rather than merely provisioned.

IBM knows the difference because it once paid the tuition. In 2013 it bought SoftLayer, then the largest privately held infrastructure cloud in the world, and took the hyperscaler fight to Amazon directly, full-page newspaper advertisements included, after losing the CIA’s cloud contract. By 2017 the SoftLayer and Bluemix brands were retired into IBM Cloud; by 2019 the answer was Red Hat and hybrid. The company tried the yardstick, conceded the race, and rebuilt around a different definition of infrastructure.

Measured on the hyperscaler yardstick, IBM will always look like it is losing. Measured on the outcomes its buyers actually purchase, the yardstick inverts. The buyers know the difference. On July 14, the tape did not.

Every rotation produces a scoreboard, and June’s is legible. The immediate winners were the constituents of the scarce layer: memory makers repricing multi-year supply, distributed infrastructure vendors shipping into the rush, and the foundry base running at records. The immediate casualties were vendors whose revenue depends on considered, quarter-end decisions: platform deals, large software renewals, transformation programmes. IBM had the misfortune of being the most visible vendor with exposure on both sides of that line, and the market punished the visible side while ignoring the other.

Because the other side matters. IBM is one of very few vendors that participated in the rush and owns the layer that follows it. Its Distributed Infrastructure business took a record share of June’s scramble and banked a backlog; its platform, integration, and consulting businesses stand where the operationalisation demand arrives next. Vendors that only shipped boxes in June met the demand once. Vendors that attach software, governance, and services to those boxes keep it.

And there is a structural fact the industry’s capex race obscures: IBM competes in this era on strikingly little capital of its own: by its own preliminary disclosure, the half produced $7.8 billion of net operating cash and $4.8 billion of free cash flow, against net capital expenditure of just $743 million. It is one of the few large vendors that does not need to buy its way into the AI build-out. It needs to attach to it.

Then there is the question the market has not asked: where does compute gravity actually pull? The honest answer, for much of the market, is toward hyperscale estates. IBM’s counterweight is deliberate and, in our view, underpriced by the narrative: neutrality and sovereignty. For regulated industries, for governments, and for enterprises whose boards have begun asking uncomfortable questions about concentration risk in the AI supply chain, the non-hyperscaler platform that respects architectural boundaries is not a legacy position. It is a governance position, and governance is precisely what the outcomes era audits.

Which brings us to the single most revealing detail in the entire disclosure, and it is not a number. Read Lightwell’s early-adopter list: Bank of America, BNY, Citi, Goldman Sachs, JPMorganChase, Mastercard, Morgan Stanley, Royal Bank of Canada, State Street, Visa, Wells Fargo. That is not a general-market logo wall. That is the systemically important core of global finance, the same constituency that anchors the Z estate and its installed MIPs base. The institutions the market believes are abandoning IBM’s centre spent early July signing into IBM’s newest governance layer. Constituencies do not deepen their relationship with a platform they are exiting.

The Machinery Beneath The Argument

Everything above is strategy, and strategy without machinery is a press release. A note that asks readers to stay bullish on a franchise owes them a tour of what that franchise actually ships. So here is the stack, layer by layer, in the terms a buyer would use.

The silicon layer. Start where the quarter broke, because the Z estate is not what the tape thinks it is.

z17 is not a faster mainframe; it is a wager about where inference belongs. Telum II executes AI inference on the processor itself, in the path of the transaction, so fraud scoring and anti-money-laundering models run at transaction latency without sensitive data ever leaving the platform. The Spyre accelerator extends that footprint from predictive scoring toward generative use cases on the same estate. The programme metrics quoted earlier in this note, the near-130 per cent programme-to-programme trajectory and the MIPs base holding or growing, are the commercial readout of that wager.

And the estate is bending toward deployment flexibility: IBM now offers z17 and LinuxONE 5 in rack-mount configurations for the first time, which matters because the argument between data-centre floor systems and rack economics is precisely the argument June’s buyers were having with their CFOs. Alongside the hardware sits watsonx Code Assistant for Z, which modernises decades-old estates in place rather than forcing a migration; State Street has been using it to renovate mainframe applications on-platform. The estate’s future, on this evidence, is renovated rather than vacated.

The open question our Greyhound Fieldnotes carry from North American buyers is composition rather than capability: financial institutions and federal agencies satisfied with z17’s confidential inferencing want a clearer account of how Spyre, OpenShift AI, and watsonx behave as one system across hybrid estates. Confidence in the compute has been earned. Confidence in the composition is the next test.

Beside Z sits the distributed portfolio that produced the quarter’s record. Power11 is positioned for exactly the workloads June’s buyers were securing: data-sovereign computing, SAP RISE estates, and controlled environments where energy per unit of work is a design constraint rather than a footnote. With energy the top barrier to AI scale for CIOs in Greyhound CIO Pulse 2026, a platform engineered around that constraint is not a legacy line. It is where a meaningful share of the late-June money landed, and the 37 per cent record quarter is the receipt.

The platform layer. Above the silicon sits the substrate that makes the portfolio one thing rather than five. Red Hat OpenShift is the hybrid foundation, with recurring revenue past $1.8 billion as of our Q3 2025 coverage, having grown more than 30 per cent over the prior year and Red Hat growth now accelerated to 11 per cent through a down quarter. HashiCorp gives that substrate its control layer: one place to manage infrastructure and application policy across clouds, which is the capability enterprises reach for the moment their AI estate spans more than one provider.

Confluent adds the data in motion, streaming operational events into AI systems in real time rather than in batches, and the acquisition closed into this posture during the first quarter, so the stream is now owned rather than partnered. Substrate, control plane, and stream: that is the anatomy of the integration buyers told our Greyhound Fieldnotes they now judge IBM on, and it is why the letter’s line that HashiCorp and Confluent performed strongly through the miss carries more information than its length suggests.

One more fact completes the layer, and it is strategic rather than technical: watsonx is deliberately embedded within AWS, Azure, Oracle, Salesforce, and ServiceNow rather than walled against them, positioning IBM’s governance and orchestration as the layer that keeps enterprise AI compliant, auditable, and connected across clouds IBM does not own. The moat is placement, not exclusivity.

The intelligence layer. watsonx is the family the market keeps mispricing as a chatbot brand. Orchestrate puts agents into business workflows. watsonx.governance is the audit and explainability layer that European banks and regulators have pulled toward rather than resisted. And underneath them, the Granite 4.0 models carry the quarter’s best-hidden product fact: they run on roughly 70 per cent less memory at about twice the inference speed of conventional architectures.

Read that against everything this note has said about the memory market. In a cycle where memory is the binding constraint, where its price broke IBM’s own quarter, model efficiency stops being benchmark vanity and becomes procurement relief. The vendor whose models sip memory is selling a hedge against the very scarcity that bruised it. Nor is IBM alone in engineering around the constraint; Qualcomm has signalled AI silicon designed for cheaper memory tiers, which tells you where the industry’s centre of effort is moving.

IBM’s inference stack extends through partnerships with Groq for high-throughput serving and Anthropic for conversational systems, and the commercial scale is no longer hypothetical: IBM’s generative AI book of business stood at more than $12.5 billion by the close of 2025.

Deutsche Telekom runs watsonx for predictive network reliability. S&P Global has embedded watsonx Assistant into its reporting workflows. IBM also runs this machinery on itself: the Client Zero programme has automated more than seventy internal workflows for around $4. And the layer the February fear actually points at, agents, is one where the company has already fielded both halves of an answer. Bob, IBM’s AI development partner, went generally available in April after eleven months as Client Zero’s sternest test: more than 80,000 IBM employees now use it, reporting an average 45 per cent productivity gain, with each task routed across frontier and open models, Anthropic’s Claude, Mistral, and IBM’s own Granite among them, by accuracy, cost, and speed.

The agent control plane in watsonx Orchestrate applies the placement logic one layer up: it observes, governs, and cost-controls agents wherever they were built and wherever they run, the moat-is-placement wager restated for the agent estate. The company the tape priced as the agent era’s prey ships both the tool that builds for that era and the control plane that governs it. And this month the same machinery reached the estate where the quarter broke: Bob’s premium package for Z went generally available, superseding watsonx Code Assistant for Z and pointing agentic development directly at the decades of COBOL and PL/I the Z franchise carries.5 billion in annualised savings, which means the reference case for the portfolio is IBM’s own operating statement.

One honest entry belongs on this layer’s ledger, because our Greyhound Fieldnotes have carried it from European buyers for three consecutive quarters: the complaint is packaging, not capability, with overlapping modules and licensing structures that clients describe as harder to buy than to run. The catalogue is strong. Its commercial wrapper is still catching up, and July 22 would be a fine moment for IBM to say how.

The trust layer. Lightwell deserves the product detail the headlines skipped, because its mechanics explain its adopter list. Lightwell Network, now generally available, is a catalogue of more than 6,500 remediated, digitally signed, and certified application-layer dependencies across the Java and Python ecosystems: not advisories about vulnerabilities, but fixed components an enterprise can consume.

The economics underneath explain the urgency. By the companies’ own count, open source now constitutes as much as 90 per cent of enterprise codebases, the average codebase carries 581 known vulnerabilities, and AI-generated exploits trade for as little as $50. Lightwell Clearinghouse Premier, in limited availability and financial services first, acts as the trusted intermediary for secured patch embargoes and coordinated response across an industry.

Underneath both runs a remediation engine that combines frontier and open AI models with human engineering to find, validate, and fix flaws at a pace no single institution can match. That is why eleven of the most systemically important names in global finance signed at launch: the quarter’s cybersecurity distraction is, for them, a permanent operating condition, and Lightwell converts it from each bank’s problem into a shared utility.

The ecosystem is already layering around it: Palo Alto Networks is pairing its virtual network patching with Lightwell’s fixes, shielding estates while permanent patches are still in testing, and the Clearinghouse roadmap points next at healthcare and government, a promised expansion worth holding to its stated timeline.

The frontier layer. Anderon completes the stack as optionality with real specifications: a standalone company headquartered in Albany, the world’s first pure-play quantum wafer foundry, backed by $1 billion in CHIPS incentives from the US Department of Commerce and matched by $1 billion of IBM’s own cash, operating at 300 millimetres and offering fabrication to multiple quantum vendors rather than IBM alone, beginning with superconducting qubit wafers and related electronics before expanding into other quantum technologies. And one discipline this note applies everywhere applies here too: the arrangement is a letter of intent, with definitive documents still to be negotiated and executed, which makes it, by our own standard, a promise on its way to proof rather than proof itself. The thesis in this note does not need quantum to work.

What should a CIO do with this tour? Notice that the stack’s economics are attach economics. Each layer makes the next cheaper to adopt: the silicon makes the platform sticky, the platform makes the intelligence governable, the intelligence makes the trust layer necessary, and every layer generates the consulting that operationalises the rest. That is what the July 22 attach question in this note is actually testing, product by product.

Greyhound Standpoint: At Greyhound Research, we believe the product answer to July 14 is already shipping. Inference beside the transaction. Models that economise on the scarcest input in the industry. A control plane that spans clouds. A trust layer built from the quarter’s own headwind. The market priced a franchise in decline; the catalogue describes a franchise in position. Machinery, not messaging.

The Architecture Question We Are Watching

A serious bullish case names its risk, and ours is architectural.

For decades, enterprise computing obeyed a single law: compute moves to data. That law is why the mainframe outlived every obituary written for it; the systems of record sat there, so the workloads stayed, so the budgets followed. Data gravity was the moat. The AI build-out introduces a countervailing force. Scarce accelerators and scarce memory exert their own pull, and data pipelines, workloads, modernisation roadmaps, and budgets have begun drifting toward wherever scarce compute can be secured. Call it compute gravity.

IBM’s product strategy is a deliberate wager against that drift: bring AI to the transaction, with Telum II and Spyre executing inference beside the workload, and code assistants modernising the estate in place. June’s buying behaviour is the first quarter-scale evidence of clients running the opposite play, moving spend, and potentially data, toward the scarce layer.

If June was a scarcity reflex, the wager holds and the deferred demand returns. If June was the first visible movement of a structural rotation away from centralised transaction processing, then Transaction Processing economics erode across years rather than quarters, and our thesis requires revision. Greyhound Research weights the first reading substantially higher today, on the strength of the z17 programme metrics and the MIPs base. But we hold the second reading open, in writing, on purpose.

Greyhound Standpoint: At Greyhound Research, we believe the defining architecture question of the next three years is whether AI comes to the transaction or the transaction’s data goes to the AI. IBM has bet its infrastructure franchise on the first answer. One quarter cannot settle the question, and honest analysis refuses to pretend otherwise.

The Recovery Runs On A Slower Clock

Honesty about the fall demands equal honesty about the repair, so let us set expectations plainly: the revival will be measured in quarters, not in months, and certainly not in weeks.

The compression we described earlier runs one way. Budgets can leave at the speed of a purchase order; they return at the speed of deployment. What flows back to IBM is not hardware grabbed off a truck but deals, platforms, and programmes, and those carry natural cycle times that no urgency can shorten: procurement reviews, security assessments, integration schedules, the hiring and training of people. The operationalisation clock that guarantees the demand also paces it.

Four further weights sit on the timetable. Slipped large deals re-close across one or two quarters at best, not in the weeks after a conference call. The pull-forward must be paid back; some of the racks that would have been bought in December were bought in June, so near-term infrastructure comparisons are owed a hole before they are owed a recovery. The memory shortage runs into 2027 on the supply chain’s own warnings, which keeps procurement noisy and the temptation to hoard alive.

And the same board-level interrogation of every dollar that guarantees the outcomes era will also lengthen approval paths, even for the software and services that ultimately win it.

None of this weakens the thesis. It disciplines it. A reader expecting July 22 to deliver the full repair is making the mirror image of the market’s error on July 14: misreading the clock. The fall was priced in a day, because falling requires only a session. The repair will be priced across quarters, because repairing requires work. We would rather be right on that timetable than loud on a shorter one.

Three Scenarios To 2027

Analysis that reaches an executive desk should say what happens next, with weights and signposts, and accept the accountability that follows. Greyhound Research’s are below.

Scenario one: reflex and recovery. The supply panic normalises as multi-year contracts replace spot hoarding. Slipped deals close through the second half of 2026; the z17 wrap settles to its guided trajectory; the Distributed Infrastructure backlog converts; software reaccelerates as operationalisation budgets arrive. By early 2027, this quarter reads as timing. The forces described in The Shortage Itself Has A Shelf Life, announced capacity evaporating and consumption trailing provisioning, are this scenario’s quiet accelerants. We assign this scenario the bulk of the probability. Signposts: deal closures disclosed on July 22, backlog growth rather than ageing, Red Hat holding double digits, Transaction Processing stabilising within two quarters.

Scenario two: the extended squeeze. The shortage deepens on the trajectory SK Hynix has warned of, pulling repeated hoarding waves through 2027. Deal timing stays noisy for several quarters across every platform vendor, margins absorb continued mix pressure, and one sting lands specifically on IBM: its own Power, Storage, and Z systems compete for the same scarce memory its clients are pre-buying, so the record backlog converts more slowly than it was signed. Direction intact; path uglier. We assign this a meaningful minority of the probability. Signposts: memory pricing through late 2026, backlog ageing beyond two quarters, a second consecutive infrastructure miss.

Scenario three: structural rotation. Compute gravity proves durable rather than cyclical. Transaction Processing enters a multi-year erosion, the MIPs base begins to shrink, and IBM is forced toward its next structural move, an acquisition into the scarce layer or a shedding of what the rotation has stranded. Its history says it would make that move rather than deny the need for it. We assign this a tail probability, and we refuse to round the tail to zero. Signposts: the second tell below breaking on both its faces, the programme metrics and Transaction Processing together.

What We Are Watching On July 22

IBM’s full results and conference call arrive on July 22 at 5 pm Eastern, and we will be listening for six tells. Each carries its threshold: the reading that confirms the thesis, and the reading that breaks it. Bullishness without checkpoints is cheerleading; these are Greyhound Research’s, in writing.

The first is deal motion since quarter end: how many of the slipped transactions have closed in the first three weeks of July, and at what aggregate value. That single disclosure separates timing from trajectory, and everything else on the call is commentary by comparison. If the slipped deals prove cancelled or materially resized rather than delayed, or no meaningful closures are disclosed, timing becomes trajectory.

The second is the Z estate’s full picture: how much of the miss sat in Z hardware, how much in Transaction Processing software, how much in consulting timing, and where the programme metrics now stand. Programme performance decaying toward z16 levels in successive quarters, or the share of installed MIPs maintaining or growing capacity falling from 85 per cent, breaks the cycle reading. And the trajectory we are watching hardest is Transaction Processing, because we named it first, in our Q3 2025 note. If that weakness outlives the z17 wrap, the hinge we identified nine months ago has not merely swung, it has broken: hardware momentum would be consuming the software attached to it rather than pulling it along. At that point this stops being a cycle and becomes a rotation.

The third is the character of the redirected client capex: whether IBM can evidence that June’s purchases were general price hedging across servers, storage, and memory rather than a targeted architectural exit. The letter’s language supports the former; the call should quantify it.

The fourth is the backlog, and its supply chain: the trajectory of the approximately $500 million in Distributed Infrastructure, whether it has grown since June 30, and whether IBM has secured the memory and components required to convert it. The shortage that created the backlog must not become the thing that gates it. If it ages rather than converts across two quarters, or shrinks as memory prices normalise, then the deferral was demand on paper, and the promissory note was never money.

The fifth is attach: whether Red Hat, watsonx, HashiCorp, and Confluent are landing on the capacity clients just secured, the second act made measurable. Red Hat decelerating from the 11 per cent it has just accelerated to would tell us capacity spending has reached platform renewals after all; consulting signings failing to convert into revenue across two or more quarters would tell us the operationalisation demand is being captured elsewhere, or insourced.

And the sixth is guidance posture: reaffirmed, rebased, or withdrawn, and with what treatment of the cybersecurity distraction and the Lightwell pipeline.

Table 4 condenses the six into the scorecard we will grade against.

To IBM’s Leadership: The Initiatives That Matter

The letter commits to new and accelerated initiatives without naming them. Since this note may reach the desks where those initiatives are being drafted, here is what we believe they must include. Little of it is new counsel; most of it is critique Greyhound Research has already published, in August and November 2025, and July 14 is what those findings look like when they compound.

Rebuild demand sensing. A company whose portfolio sells decision intelligence did not detect a capex reprioritisation moving through its own pipeline until the final weeks of a quarter. Instrument the pipeline the way IBM asks clients to instrument their operations: leading indicators, procurement-side signals, supply-chain telemetry feeding the forecast weekly, not quarterly.

Fix the accountability seams, because the deepest initiative is one the letter does not name. In August 2025 Greyhound Research documented post-pilot delivery faltering on fractured accountability between Red Hat, Consulting, and watsonx teams, with enterprise accounts running escalation loops simply to establish who owned the architecture; we wrote then that this is not a product flaw, it is a coordination gap. By November, buyers across India, Singapore, and Malaysia were describing timelines running longer than expected, with delays traced to coordination between global and regional delivery, while buyers in Australia and the UAE told us local solutioning depth still trailed their intent. The deal-desk races lost in June and the flat consulting conversion are the same gap, now priced in public. Whatever the accelerated initiatives contain, a single owner for the seam between hardware, software, and services belongs at the top of the list.

Re-engineer the deal desk for the compressed clock. If transitions now move at the speed of a purchase order, quarter-end commercial machinery built for a slower era will keep losing races it should win. Velocity, delegated authority, and pre-approved flexibility bands are now competitive weapons, not administrative details. We flagged the shape of this in August 2025, when consulting renewals in Spain and Italy slipped on pricing and go-to-market terms misaligned with procurement cycles; June was the same machinery losing faster races.

Take the friction out of staying. Accelerate consumption-based and as-a-service commercial models for the Z estate, so that a client hoarding capex elsewhere never has to choose between securing the scarce layer and refreshing the system of record. Deferral should be made commercially unnecessary. The strongest answer to a capex panic is a platform that does not require capex.

Deploy IBM Financing as the operationalisation bridge. Clients have exhausted capital envelopes buying hardware; the platform and services demand this note describes will arrive constrained by financing capacity, not by intent. IBM owns a financing arm. In this cycle it is strategy, not plumbing.

Secure the backlog’s supply chain, and say so on July 22. A record backlog is an asset only if it converts, and it converts only if IBM wins its own share of the scarce components.

And protect the candour, pairing it with proof. The letter’s tone earned more goodwill than choreography would have; candour is rewarded when conversion follows it. Then meet the bar we set in our third-quarter 2025 note: make execution feel effortless. July 14 is what missing that bar looks like in public.

To The Technology Buyer: Five Moves While The Window Is Open

For the technology buyers Greyhound Research advises, the counsel is fivefold, and the window for most of it is measured in days, not quarters. Because this note concerns IBM, each move carries its IBM-specific application: the ask a buyer should put on the table now, and every one of them an ask a vendor defending a franchise can say yes to.

Treat AI infrastructure as a full-stack cost object rather than a hardware line. Memory, storage, power, networking, security, model operations, and staffing belong in one total-cost view, reviewed at board cadence, because June just demonstrated what fragmented views miss: a capex reprioritisation large enough to move a seventeen-billion-dollar quarter travelled through enterprise pipelines without most governance processes registering it until vendors disclosed it. If the estate is priced in pieces, it will be repriced by events. The total-cost view is the instrument that lets a buyer distinguish a genuine scarcity premium from a panic premium, which is precisely the distinction the market itself failed on July 14. With IBM, the application is direct: ask for the estate on one commercial sheet, hardware, software, services, and financing quoted as a system against your total-cost view, with delivery dates in writing, because the disclosure has already told you supply is constrained, and a constrained vendor commits in writing or not at all.

Ring-fence the software that makes scarce capacity productive: security, data, integration, observability, governance. When budgets tighten around hardware, the reflex is to cut horizontally, and the horizontal cut lands hardest on exactly the layers that convert racks into outcomes. Cut duplication rather than muscle; most estates carry two or three overlapping tools per category, and consolidation funds the squeeze without touching capability. The hoarded capacity of June becomes the stranded capacity of December only if the software that operationalises it is starved in between. With IBM, trade proof for price: the company needs attach evidence, Red Hat, watsonx, and Lightwell landing on newly secured capacity, more than it needs list price this half, and a buyer willing to stand as a named reference or a measured case should extract pilot economics and success-based terms for exactly the layers this move ring-fences.

Put one question to every renewal and every programme: does this feed, secure, or govern the scarce layer of the stack, or does it merely sit beside it? Fund the first group ahead of the queue. This is the buyer-side version of the adjacency re-ranking the market performed in June, applied deliberately rather than in panic. A spend item that cannot answer the question is not automatically cut, but it queues behind everything that can, and the queue itself becomes the governance. With IBM, run the test on IBM’s own paper: ask the account team to map every renewal line to feed, secure, or govern, and let whatever merely sits beside it fund the consolidation. The in-path inference case for Telum and Spyre is theirs to prove; ask for benchmark commitments on your workloads, not the brochure’s.

Discipline the deferrals. Give every postponement a return date, because a refresh deferred without one is not a deferral; it is an abandonment happening slowly, and it will surface as risk in an audit rather than a plan. Assign each deferred item an owner and a trigger, the price point or supply signal at which it returns to the front of the queue, and review that register with the same seriousness as new spend. IBM’s own quarter is the cautionary case read from the vendor side: deferred demand is real, but only the buyers who tracked what they deferred will convert the recovery on their own terms rather than the market’s. With IBM, make the vendor underwrite its own thesis. The letter calls the lost demand deferred, not destroyed; a vendor that believes that should have no difficulty attaching price protection and a guaranteed build slot to your return date. If the account team hesitates to contract on IBM’s own reading of the quarter, the hesitation is information.

And use the moment: negotiate now. The weeks after a vendor’s public stumble are the strongest commercial window a buyer receives, and this one has a visible closing date, because July 22 either restores the vendor’s footing or resets the terms entirely. Seek multi-year price protection, capacity guarantees, and flexibility clauses while it is open, specifically and respectfully: specifically, because vague asks waste the moment; respectfully, because the counterparty across the table will remember the tone long after it remembers the discount. A buyer who arrives with the total-cost view from the first move and the deferral register from the fourth negotiates from evidence rather than opportunism. And with IBM, the service ask matters as much as the price ask: a single named executive accountable for the seam between Red Hat, Consulting, and watsonx on your account, with an escalation clock attached, because the coordination gap Greyhound Research has documented since August 2025 is contracted away far more reliably than it is reorganised away. Put the asks on the table before July 22, while the answers are still being drafted.

Greyhound Standpoint

At Greyhound Research, we believe July 14 priced a permanent verdict on a temporary reflex. The arithmetic of hoarding, the sunset of own-and-operate, the board-level reset toward outcomes, the operationalisation clock, the evidence of the preliminary print, and a two-decade record of structural adaptation all point the same way: the demand IBM lost in June was deferred, not destroyed, and it returns with urgency attached, on a clock measured in quarters rather than weeks, into precisely the layers where IBM has spent seven years concentrating its portfolio.

Transitions now arrive at the speed of a purchase order, and the vendors that endure will be those whose reflexes, commercial as much as productive, run at that tempo.

The company faltered on timing and said so plainly. The market answered by erasing a quarter of its value over a quarter in which it grew. One of those two reactions was proportionate.

Ten months ago the same market was pouring champagne over a backlog it had mistaken for a bank balance, and we said so while the market was still applauding. Today it is holding a wake over a deferral it has mistaken for a death. We are not switching sides. We are applying the same rule we applied then, the only rule that survives both moods: applause is not delivery, and panic is not diagnosis. Count what converts.

We remain bullish on IBM. Not on its share price, which is not our trade, but on its franchise, which is our field. Hardware can be hoarded. Outcomes cannot. And the outcomes era has only just presented its bill.

Important Disclaimer

Greyhound Research are industry analysts, not equity analysts. This note is a demand-side assessment of IBM’s franchise and strategy, built on IBM’s preliminary disclosure; final figures may differ, and we will update this position after the company’s full results on July 22. It is not investment advice, and nothing in it should be read as a view on securities.

This is a strictly independent analysis. IBM has not commissioned, paid for, reviewed, or influenced this note in any form, and Greyhound Research has received no compensation from IBM or any other party for its production. The views expressed are Greyhound Research’s alone.

This note draws on Greyhound Research’s published IBM coverage, including our Q3 2025 results note, and on Backlog Isn’t Bank Balance (September 2025). Buyer evidence is drawn from Greyhound CIO Pulse 2026 and Greyhound Fieldnotes. Our full archive is available at greyhoundresearch.com.

Sources And References

Every load-bearing claim in this note traces to one of the sources below, presented in Harvard reference format and grouped by class. The class matters: company disclosures and government records carry different evidential weight from press reporting, and readers are entitled to see which is which. In-text author-date citations are deliberately omitted, as they serve academic writing rather than research intended for the executive desk. All sources accessed 15 and 16 July 2026.

Company and vendor disclosures

IBM (2026a) Arvind Krishna’s Letter to IBM Investors: Selected Preliminary Second-Quarter 2026 Financial Results. Form 8-K exhibit filed with the US Securities and Exchange Commission, 14 July. Available at: https://www.sec.gov/Archives/edgar/data/51143/000005114326000070/ibm-20260714xex991.htm

IBM (2026b) IBM Releases Fourth-Quarter Results. Form 8-K exhibit filed with the US Securities and Exchange Commission, 28 January. Available at: https://www.sec.gov/Archives/edgar/data/51143/000005114326000004/ibm-20260128xex991.htm

IBM (2026c) IBM Releases First-Quarter Results. IBM Newsroom, 22 April. Available at: https://newsroom.ibm.com/2026-04-22-IBM-RELEASES-FIRST-QUARTER-RESULTS

IBM (2026d) IBM to Announce Second-Quarter 2026 Financial Results. IBM Newsroom, 8 July. Available at: https://newsroom.ibm.com/2026-07-08-IBM-to-Announce-Second-Quarter-2026-Financial-Results

IBM (2026e) IBM Launches New Power Systems and Software Built for Enterprises to Address Risk, Productivity, and Flexibility. IBM Newsroom, 15 July. Available at: https://newsroom.ibm.com/

IBM (2026f) Introducing IBM Bob: AI Development Partner that Takes Enterprises from AI-Assisted Coding to Production-Ready Software. IBM Newsroom, 28 April. Available at: https://newsroom.ibm.com/2026-04-28-introducing-ibm-bob-ai-development-partner-that-takes-enterprises-from-ai-assisted-coding-to-production-ready-software

IBM (2026g) Agent control plane, IBM watsonx Orchestrate, product documentation. Available at: https://www.ibm.com/products/watsonx-orchestrate/agent-control-plane

IBM (2026h) Announcing the IBM Bob Premium Package for Z: Accelerating Enterprise Mainframe Application Modernization. IBM Announcements, July. Available at: https://www.ibm.com/new/announcements/announcing-the-ibm-bob-premium-package-for-z

IBM and Red Hat (2026) IBM and Red Hat Expand Lightwell with New Offerings to Build the Trust Infrastructure for AI-Era Open Source. IBM Newsroom, 8 July. Available at: https://newsroom.ibm.com/2026-07-08-ibm-and-red-hat-expand-lightwell-with-new-commercial-offerings-to-build-the-trust-infrastructure-for-ai-era-open-source

Red Hat (2026) IBM and Red Hat Expand Lightwell with New Offerings to Build the Trust Infrastructure for AI-Era Open Source. Red Hat Press Releases, 8 July. Available at: https://www.redhat.com/en/about/press-releases/ibm-and-red-hat-expand-lightwell-new-offerings-build-trust-infrastructure-ai-era-open-source

IBM and US Department of Commerce (2026) IBM and U.S. Department of Commerce Announce America’s First Purpose-Built Quantum Foundry, Supported by Proposed $1 Billion CHIPS Award. IBM Newsroom, 21 May. Available at: https://newsroom.ibm.com/ibm-and-u-s-department-of-commerce-announce-americas-first-purpose-built-quantum-foundry

Oracle (2025) Oracle Announces Fiscal Year 2026 First Quarter Financial Results. Oracle Investor Relations, 9 September. Available at: https://investor.oracle.com/investor-news/news-details/2025/Oracle-Announces-Fiscal-Year-2026-First-Quarter-Financial-Results/default.aspx

OpenAI (2025) Stargate Advances with Partnership with Oracle. Available at: https://openai.com/index/stargate-advances-with-partnership-with-oracle/

Cognizant (2026) Cognizant to Acquire Astreya, Deepening Its AI-First Managed Services Capabilities at Scale. Cognizant Investor Relations, 29 April. Available at: https://investors.cognizant.com/news-and-events/news/news-details/2026/Cognizant-to-Acquire-Astreya-Deepening-Its-AI-First-Managed-Services-Capabilities-at-Scale/default.aspx

Government and official records

Board of Governors of the Federal Reserve System (2026) Monitoring AI Adoption in the U.S. Economy. FEDS Notes, 3 April. Available at: https://www.federalreserve.gov/econres/notes/feds-notes/monitoring-ai-adoption-in-the-u-s-economy-20260403.html

Electric Reliability Council of Texas (2026) ERCOT’s New Batch Connection Process for Large Electricity Users. ERCOT Trending Topics, June. Available at: https://www.ercot.com/files/docs/2026/06/18/ERCOT-Trending-Topic-New-Batch-Connection-Process-for-Large-Electricity-Users.pdf

Greyhound Research published coverage

Greyhound Research (2025a) Backlog Isn’t Bank Balance: A Reality Check on Oracle’s AI Story, 12 September. Available at: https://greyhoundresearch.com/backlog-isnt-bank-balance-a-reality-check-on-oracles-ai-story/

Greyhound Research (2025g) IBM’s Q2 2025 Results: Buyer Confidence Grows, But Field-Level Complexity Persists, 13 August. Available at: https://greyhoundresearch.com/ibms-q2-2025-results-buyer-confidence-grows-but-field-level-complexity-persists/

Greyhound Research (2025b) IBM’s Q3 2025 Results: Delivery Strength Holds, But Integration Demands Intensify, 16 November. Available at: https://greyhoundresearch.com/ibms-q3-2025-results-delivery-strength-holds-but-integration-demands-intensify/

Greyhound Research (2025c) What IBM Power11 Really Means for Enterprise Architecture and Operations. Available at: https://greyhoundresearch.com/what-ibm-power11-really-means-for-enterprise-architecture-and-operations/

Greyhound Research (2025d) Streaming, Sovereignty, And The New AI Stack: Decoding The IBM-Confluent Deal. Available at: https://greyhoundresearch.com/streaming-sovereignty-and-the-new-ai-stack-decoding-the-ibm-confluent-deal/

Greyhound Research (2025e) What’s Next for the Microsoft OpenAI Relationship? Available at: https://greyhoundresearch.com/whats-next-for-the-microsoft-openai-relationship/

Greyhound Research (2026a) IBM coverage archive. Available at: https://greyhoundresearch.com/category/by-brand/ibm/

Proprietary Greyhound Research evidence

Greyhound Research (2026b) Greyhound CIO Pulse 2026. Survey of more than 1,000 enterprise technology leaders. Cited for mega-announcement scepticism at 82 per cent, the shift from AI adoption to AI accountability at 84 per cent, energy cost and availability as the leading barrier to AI scale at 55 per cent, and delivery of measurable outcomes across a majority of materially funded AI initiatives at 32 per cent.

Greyhound Research (2025f) Greyhound CIO Pulse 2025. Survey of more than 800 enterprise technology leaders. Cited for the corresponding prior-year figures of 62 per cent, 78 per cent, 47 per cent, and 25 per cent.

Greyhound Research (2026c) Greyhound Fieldnotes. Advisory engagements with enterprise technology buyers across the Americas, Europe, and Asia Pacific.

Supply chain and data centre build-out

Tom’s Hardware (2026a) Micron inks long-term supply agreements worth $100 billion, says it has no idea when RAM crisis will end, June. Available at: https://www.tomshardware.com/pc-components/dram/micron-inks-long-term-supply-agreements-worth-usd100-billion-says-it-has-no-idea-when-ram-crisis-will-end

Reuters (2026a) SK Hynix CEO sees worst-ever memory supply shortage in 2027, says demand to outstrip supply beyond 2030, 10 July. Reproduced by US News. Available at: https://money.usnews.com/investing/news/articles/2026-07-10/sk-hynix-ceo-sees-worst-ever-memory-supply-shortage-in-2027-says-demand-to-outstrip-supply-beyond-2030

Reuters (2026b) Micron tops estimates, touts $22 billion in customer deals for memory chips, 24 June. Reproduced by The Standard. Cited for Qualcomm on AI silicon designed for cheaper memory tiers. Available at: https://www.thestandard.com.hk/innovation/article/335568/Micron-tops-estimates-touts-US22-bln-in-customer-deals-for-memory-chips

Reuters (2026c) Dell expects AI server revenue to double in fiscal 2027 on data center boom, 26 February. Reproduced by Yahoo Finance. Available at: https://finance.yahoo.com/news/dell-forecasts-fiscal-2027-revenue-210929667.html

Reuters (2026d) Ericsson warns of rising memory chip costs driven by surging AI demand, 14 July. Reproduced by Global Banking and Finance Review. Available at: https://www.globalbankingandfinance.com/ericsson-profit-beats-forecast-second-quarter/

Reuters (2026e) TCS plans up to 8,900 AI deployment engineers, seeks AI acquisitions, 12 July. Reproduced by Business Standard. Available at: https://www.business-standard.com/companies/news/tcs-plans-up-to-8-900-ai-deployment-engineers-seeks-ai-acquisitions-126071200332_1.html

TechRadar (2026) Nearly half of US data centers planned for 2026 canceled or delayed, 10 April. Reporting Bloomberg analysis of US data centre construction. Available at: https://www.techradar.com/pro/if-one-piece-of-your-supply-chain-is-delayed-then-your-whole-project-cant-deliver-nearly-half-of-us-data-centers-planned-for-2026-canceled-or-delayed-and-things-could-soon-get-much-worse

Tech-Insider (2026a) U.S. AI Data Center Delays: 7 GW Capacity Crisis, 4 June. Reporting forward pipeline capacity and electrical equipment lead times. Available at: https://tech-insider.org/us-ai-data-center-delays-cancellations-7gw-capacity-crisis-2026/

NBC News (2026) Data center opponents have blocked or delayed projects worth nearly $130 billion in 2026, study finds, 12 June. Available at: https://www.nbcnews.com/tech/tech-news/data-center-opposition-sharply-rising-2026-study-finds-rcna349728

Tom’s Hardware (2026b) More than 75 data center build-outs worth $130 billion have been successfully blocked in the first three months of 2026, 13 June. Available at: https://www.tomshardware.com/tech-industry/artificial-intelligence/more-than-75-data-center-build-outs-worth-usd130-billion-have-been-successfully-blocked-in-the-first-four-months-of-2026-bipartisan-opposition-mounts-nationwide-over-fears-of-soaring-power-and-water-costs

The Irish Times (2026) What the world can learn from Ireland’s battle to power data centres, 17 March. Available at: https://www.irishtimes.com/business/2026/03/17/what-the-world-can-learn-from-irelands-battle-to-power-data-centres/

Philip Lee (2026) Re-balancing the Digital Bargain: Ireland’s New CRU Large Energy User Connection Policy, 8 January. Available at: https://www.philiplee.ie/re-balancing-the-digital-bargain-irelands-new-cru-large-energy-user-connection-policy/

Market and industry reporting

The Wall Street Journal (2025) OpenAI’s Funding Challenges Loom Over Oracle, Broadcom Deal Spree. Available at: https://www.wsj.com/tech/ai/openais-funding-challenges-loom-over-oracle-broadcom-deal-spree-be353399

CNBC (2026) IBM stock craters 25%, the worst day on record, after company issues second-quarter earnings warning, 14 July. Available at: https://www.cnbc.com/2026/07/14/ibm-warns-second-quarter-earnings-fell-short-of-expectations.html

Tech-Insider (2026b) OpenAI IPO: $850B Valuation, $25B Revenue. Cited for the restructured Microsoft and OpenAI arrangement, OpenAI’s projected cash burn, and Oracle backlog concentration. Available at: https://tech-insider.org/openai-ipo-850-billion-valuation-2026/

The Register (2025) Oracle insists its $300B contract with OpenAI is on schedule, 15 December. Cited for reported site delivery slippage, Oracle’s denial, and Oracle debt. Available at: https://www.theregister.com/2025/12/15/oracle_denies_openai_delays/

Yahoo Finance (2026a) Oracle cuts 21,000 jobs as it prepares $50 billion AI buildout. Available at: https://finance.yahoo.com/technology/ai/articles/oracle-orcl-cuts-21-000-221026147.html

Bloomberg Opinion (2026) Salesforce and Other SaaS Companies Deserve Their AI Reckoning, 12 February. Available at: https://www.bloomberg.com/opinion/articles/2026-02-12/salesforce-and-other-saas-companies-deserve-ai-disruption

Tech-Insider (2026c) AI Agents Just Erased $2T in SaaS Value, 4 June. Cited for software market value erased since January 2026 and drawdowns across horizontal software. Available at: https://tech-insider.org/saas-stock-crash-ai-agents-2-trillion-2026/

Forbes (2026) AI Fears Keep Hammering Software Stocks, Even Those Reporting Good Earnings, 23 April. Available at: https://www.forbes.com/sites/aliciapark/2026/04/23/ai-fears-keep-hammering-software-stocks-even-those-reporting-good-earnings/

StocksToTrade (2026) Workday Stock Draws Upgrade As AI Risks And Insider Selling Mount, 13 July. Available at: https://stockstotrade.com/news/workday-inc-wday-news-2026_07_13/

Stockpil (2026) Software Stocks Got Crushed. Did They Have It Coming? Cited for enterprise software forward multiple compression. Available at: https://stockpil.com/software-stocks-selloff-ai-overreaction

Yahoo Finance (2026b) Salesforce, Workday, and PagerDuty Stocks Trade Up, July. Cited for the rotation from infrastructure into application software. Available at: https://finance.yahoo.com/markets/stocks/articles/salesforce-workday-pagerduty-stocks-trade-012639189.html

TheStreet (2026) IBM just answered a $5 billion cybersecurity question, 9 July. Cited for the Palo Alto Networks and Lightwell pairing and the Lightwell commercial model. Available at: https://www.thestreet.com/technology/ibm-red-hat-lightwell-open-source-security-launch

The Quantum Insider (2026) IBM Plans $10 Billion Quantum Push as Efforts to Commercialize Quantum Intensifies, 28 May. Available at: https://thequantuminsider.com/2026/05/28/ibm-plans-10-billion-quantum-push-as-efforts-to-commercialize-quantum-intensifies/

GeekWire (2013) After losing CIA bid, IBM swipes at Amazon in new campaign, 4 November. Available at: https://www.geekwire.com/2013/losing-cia-bid-ibm-swipes-amazon-cloud-campaign/

Data Center Dynamics (2017) IBM retires Bluemix brand, November. Available at: https://www.datacenterdynamics.com/en/news/ibm-retires-bluemix-brand/

Copyright Policy. All content contained on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. We request our readers not to copy Greyhound Research content and not republish or redistribute them (in whole or partially) via emails or republishing them in any media, including websites, newsletters, or intranets. We understand that you may want to share this content with others, so we’ve added tools under each content piece that allow you to share the content. If you have any questions, please get in touch with our Community Relations Team at connect@thofgr.com.