IBM’s 25% Single Day Plunge: The AI Boom’s Biggest Victim May Be The Market’s Anti-AI Trade

Enterprise capex is stampeding into the chips of today. IBM’s answer may be to double down on building the chips of tomorrow.

On Tuesday 14 July, IBM shares plunged just over 25 per cent in a single session, closing at around $217. That is the worst day in at least 58 years of trading records, and quite possibly the worst in the 110 years since the company’s predecessor listed on the New York Stock Exchange in 1916. The previous holder of the title was Black Monday in October 1987, when the stock fell 23.7 per cent amid a market-wide collapse. This time there was no market-wide collapse to hide behind. This was IBM’s problem alone, and roughly a quarter of a $272 billion company evaporated before lunch.

The trigger was unusual in itself. IBM preannounced its second-quarter results a full week ahead of its scheduled 22 July earnings call, something companies only do when the numbers are bad enough that sitting on them becomes a legal risk. The numbers were bad. Adjusted earnings per share came in at $2.93 against a consensus of $3.02, and revenue landed at $17.2 billion against expectations of $17.86 billion. A $660 million revenue miss is not, on its own, the stuff of historic collapses. What frightened the market was the reason behind it.

The memory shortage ate IBM’s quarter

In a letter to investors, CEO Arvind Krishna explained that in the final weeks of June, enterprise clients abruptly redirected their quarterly capital expenditure away from software and mainframe purchases and towards servers, storage and memory. Clients were racing to secure supply-constrained hardware ahead of expected price increases. Krishna admitted that while IBM had anticipated some supply-chain impact, it did not anticipate the sheer magnitude of the reprioritisation, and that numerous large deals slipped past the quarter’s close as a result. His phrasing was blunt for a CEO letter: “this quarter we faltered.”

The underlying mechanics are worth spelling out, because they matter well beyond IBM. The global memory market has been squeezed since late 2025, as Samsung, SK Hynix and Micron shifted fabrication capacity towards the high-bandwidth memory that feeds AI accelerators. That capacity is effectively sold out for 2026 under long-term contracts, leaving conventional DRAM for servers, PCs and phones in structural shortage. Micron’s CEO has said he expects tight conditions to persist beyond calendar 2027. Enterprise IT departments, watching hardware prices climb, did the rational thing: they raided their software and services budgets to stockpile compute and memory while they still could.

IBM’s z17 mainframe business, which the company had already pencilled in for a low single-digit decline, fared significantly worse than that. Software revenue still grew, but at 5 per cent rather than the 11 per cent posted in the first quarter, a deceleration sharp enough to unnerve anyone who had bought the “IBM as a software growth story” thesis. The pain spread immediately. Salesforce fell around 4 per cent on the day, Microsoft nearly 3 per cent, and the IGV software ETF dropped 2 per cent. Zoom out and the picture is grimmer still: Oracle is down roughly a third year to date, Microsoft around 20 per cent, and Accenture has halved. The AI infrastructure boom is not lifting all boats. It is actively draining budget from everything adjacent to it.

There is also a legal coda already forming. At least one securities firm announced an investigation within hours, probing whether IBM misrepresented the pace of new deal signings and the strength of its IBM Z outlook. Whatever comes of that, it adds friction at precisely the moment management needs freedom to manoeuvre. Nor is this IBM’s first bruising of the year: back in February the stock suffered a 13 per cent single-day fall, its worst since 2000, after Anthropic demonstrated an AI tool for modernising COBOL, the decades-old business language that still underpins a meaningful slice of IBM’s installed base. Twice in six months, the market has repriced Big Blue on the suspicion that AI is coming for its legacy franchises.

Big Blue is still the biggest name in quantum

Before turning to what this means for quantum, it is worth restating why the question matters at all. IBM is not a peripheral player in quantum computing; it is arguably the field’s central institution. It put the first quantum computer on the public cloud in 2016, open-sourced Qiskit the following year, and has since built Qiskit into the de facto standard software stack for gate-based quantum computing, with a developer community no rival comes close to matching. Its roadmap discipline is equally singular: milestone after milestone has landed on schedule, from the Heron generation through to last November’s unveiling of the 120-qubit Nighthawk architecture and the experimental Loon chip, which IBM says demonstrates every hardware component needed for fault tolerance.

That cadence has continued right through the current turmoil. Earlier this month IBM shipped Qiskit v2.5, a substantial release that introduces a new multi-representation compiler framework, expands the C API that lets Qiskit run natively inside high-performance computing environments, and delivers major transpiler speedups, alongside renaming Qiskit Runtime Service to IBM Quantum Compute Service. These are unglamorous engineering releases, but they are exactly the plumbing required to make quantum a genuine infrastructure product rather than a laboratory curiosity. The direction of travel is unambiguous: IBM is wiring quantum into the classical supercomputing stack, where the enterprise money now lives.

The timing of this week’s hardware news underlines the point. On the very day after the crash, IBM released Nighthawk R2, the second iteration of the square-lattice processor family, and the scheduled step on the path towards the 7,500-gate capability IBM has promised by the end of 2026, up from the 5,000 two-qubit gates the first Nighthawk supported at launch. The quantum division, in other words, is shipping on schedule in the same week the rest of the company is apologising for missed execution. That contrast will not be lost on anyone inside Armonk arguing over where the next capital allocation round should land.

Quantum had nothing to do with this, and everything to do with what comes next

Let us be clear about one thing first: quantum played no role in Tuesday’s collapse. IBM Quantum remains a rounding error against $17 billion of quarterly revenue, and no analyst model moved a decimal point over Nighthawk or Starling this week. The interesting question is the second-order one, which is what a wounded IBM does with its most credible long-horizon bet.

The bear case writes itself. A company that has just lost a quarter of its market value, faces a shareholder lawsuit machine spinning up, and is watching its core software and mainframe franchises leak budget to AI hardware is a company under pressure to cut. Quantum is exactly the kind of programme that CFOs circle in red ink during these moments: enormously capital-intensive, revenue-light, and with its principal payoff (the fault-tolerant Starling system, 200 logical qubits, 100 million operations per job) not due until 2029. The Poughkeepsie data centre build, the Albany 300mm fabrication line, the four-year Nighthawk iteration cycle: all of it is real money spent years ahead of real returns. If the July shortfall turns into a multi-quarter slump, investor patience for decade-scale physics projects gets thin quickly.

But I think the bear case, while plausible, reads the situation backwards. Consider what Tuesday’s collapse actually revealed about where enterprise money is going. Clients are not spending less; they are spending differently, and specifically they are spending on differentiated compute infrastructure. Generic software and consulting, the businesses IBM has leaned on for decades, are precisely what got raided. The lesson for IBM’s board is not that speculative infrastructure is a luxury. It is that speculative infrastructure is the only thing enterprises are currently willing to pay a premium for, and quantum is the one category of differentiated compute where IBM plausibly leads the world rather than trailing Nvidia by a country mile.

My suspicion is that Tuesday strengthens the hand of the quantum programme within IBM’s own ranks rather than weakening it. The crash was a brutal demonstration that the software and consulting franchises, however profitable today, are exposed the moment enterprise capex swings towards chips and infrastructure. Quantum cannot be the whole answer, and it would be reckless to pretend a pre-revenue division can backstop a $17 billion quarter. But as a hedge against a world where clients pay for differentiated compute rather than generic services, it is the best card IBM holds, and boards tend to notice that sort of thing after a 25 per cent day.

The chips of tomorrow

There is an operating thesis buried in all of this that deserves to be stated plainly. IBM cannot win the fight for today’s chips. It exited commodity semiconductor manufacturing a decade ago when it paid GlobalFoundries to take its fabs, and it holds no meaningful position in the merchant memory and accelerator markets that are currently hoovering up every spare dollar of enterprise capex. Watching client money flood into other people’s silicon while your own software deals slip is a painful way to learn where you sit in the value chain.

The rational response is not to re-enter a commodity war it cannot win, but to make sure it owns the chips of tomorrow. That is precisely what the quantum programme now represents. IBM has moved primary quantum processor fabrication to the advanced 300mm wafer line at the Albany NanoTech Complex, doubling its development speed and increasing chip complexity tenfold, and it demonstrated real-time qLDPC error decoding in under 480 nanoseconds a full year ahead of schedule. The roadmap runs through Kookaburra this year, Cockatoo’s chip-to-chip entanglement in 2027, and Starling’s fault-tolerant debut in 2029, with the Blue Jay quantum-centric supercomputer beyond that. If the next compute scarcity cycle is quantum rather than memory, IBM intends to be the company everyone else is stockpiling from. Seen through that lens, Tuesday was not an argument for retrenchment; it was a preview of the market position IBM is trying to buy itself a decade early.

The harder-edged commercial point cuts the same way. Consulting and conventional software are melting under AI pressure across the entire sector, and IBM knows it better than anyone after February’s COBOL episode. A structurally shrinking legacy business raises, rather than lowers, the strategic value of the one franchise where IBM owns the full stack: chips, cryogenics, error correction, software and a claimed 2026 advantage milestone. If Krishna wants a narrative for the 22 July earnings call that is not simply an apology, “we are redeploying towards the compute categories our clients are actually fighting over” is the obvious candidate, and quantum sits squarely inside it.

A contrarian hedge against the AI trade

Whether Tuesday represents a buying opportunity remains to be seen, and nothing here should be read as investment advice. But there is a contrarian thesis worth airing, because it inverts the way most investors currently frame IBM. The company is genuinely diversified across software, infrastructure, consulting and financing, with revenue anchored in long-standing relationships with governments, banks and other institutions that change vendors roughly once a generation. Draw the Venn diagram of what IBM’s earnings actually depend on and, for all the watsonx marketing, the one circle conspicuously absent is the AI infrastructure trade itself. Big Blue sells around the AI boom, not on top of it.

There is a rich irony here, because IBM was earlier to AI than almost anyone. Watson beat the best human players at Jeopardy in 2011, a full decade before ChatGPT, and IBM spent years marketing Watson as the future of everything from oncology to weather forecasting. The bet never made it into the mainstream: Watson Health underdelivered and was eventually sold off, and the brand quietly receded. Today IBM does maintain an AI line, with watsonx and its Granite family of deliberately small, enterprise-focused models, but nobody counts it among the frontier model providers, and its valuation carries none of the premium attached to those that are. Having been burned by one AI hype cycle already, IBM has effectively sat out the current one, which is exactly what makes it interesting as the antithesis of an AI investment. Ask what IBM’s AI defence actually is and there is no confident answer, but that absence may prove to be one of its stronger cards: a company with no AI narrative baked into its price has no AI narrative to unwind, no multiple to give back, and nothing to confess on the day the music stops.

Ask what IBM’s AI defence actually is and there is no crisp answer, which sounds damning until you notice that most of its peers are spending fortunes constructing theirs. A company whose earnings do not depend on winning the AI race does not need a moat against losing it. In a sector where every strategy deck now opens with an AI slide, having nothing much to defend may quietly prove one of the stronger positions of all.

Tuesday demonstrated the cost of that positioning in a market where AI capex crowds out everything else, but the same positioning cuts the other way if the AI trade deflates. IBM’s quarter was wrecked because client money stampeded towards AI servers and memory; if that build-out proves overdone and the stampede reverses, the budgets flow back towards exactly the software, mainframe and consulting spend that IBM lives on, and the enterprise IT environment reverts to the one in which Big Blue has quietly compounded for decades. In a portfolio stuffed with names whose valuations assume the AI build-out continues indefinitely, a beaten-down IBM starts to look like a hedge: a large, dividend-paying, systemically embedded enterprise vendor whose fortunes improve if the bubble thesis proves correct, and which carries a genuine post-AI frontier bet in quantum as a kicker. For investors who believe AI is overhyped and overbought, there are worse places to look than the stock the AI boom just punished hardest. All the standard caveats apply, and a falling knife this sharp deserves respect either way.

What I will be watching on 22 July

The full earnings call next week is now the most important event on IBM’s calendar in years, and quantum watchers should pay attention alongside the market. Any language about capital discipline, portfolio prioritisation or “focusing investment” will need parsing carefully, because that is where a quiet quantum slowdown would first show up. Conversely, if Krishna leans into the infrastructure pivot, expect the 2026 quantum advantage target and the Starling timeline to feature prominently as proof that IBM still owns a frontier.

My working view is that the quantum roadmap survives intact and gains internal priority, with the Nighthawk R2 release and Qiskit v2.5 serving as well-timed reminders that this is the one division delivering exactly what it promised, exactly when it promised it. But the margin for error has vanished. The end-of-2026 verified advantage claim was always ambitious. It is now something closer to load-bearing. A company that has just told the market “we faltered” cannot afford to miss the one deadline it set for itself in the one field where it claims undisputed leadership. For the quantum sector as a whole, that makes the next six months unusually interesting: the industry’s biggest player now needs a win, publicly, on a clock of its own making, and history suggests that a cornered Big Blue is at its most dangerous when it has something to prove.