On July 14, IBM CEO Arvind Krishna took the unusual step of pre-announcing second-quarter results ahead of the scheduled July 22 report:
The chain runs: AI infrastructure supply is tight (memory and servers expected to get pricier) → enterprise clients shift capex from software subscriptions and mainframe maintenance toward stockpiling hardware first → a large share of IBM's expected deals don't close, revenue misses by 3.7% → the market reads that 3.7% miss as evidence IBM's AI/software strategy is broken rather than a one-quarter timing shift → the stock drops nearly seven times the size of the revenue miss, a textbook case of positioning-driven cascading selling, not a linear valuation repricing.
The detail that matters: this is not "AI disrupting IBM's software" — it is "the AI-hardware land-grab temporarily redirecting corporate budgets away from software." Krishna offered a more specific, far less-repeated detail in an interview: Anthropic's AI cybersecurity product Mythos is making clients pause and ask "how much do I actually need to spend on cyber" — a much narrower, more precise, and more fragile threat than the sweeping "AI is eating software companies" narrative.
Krishna told CNBC: "We don't see our software being disrupted by AI at all." That same week, an HSBC analyst reached the opposite conclusion, downgrading IBM to Reduce because its 22.0x CY27 multiple looked stretched against a 16.9x sector median, and suggested investors could instead buy a "synthetic IBM" basket of SAP, Accenture, HP and IonQ.
One analysis probability-weighted bull/base/bear scenarios to a fair-value midpoint near $232, about 8% above the roughly $215 intraday price — a result the analysis said was not especially sensitive to the weighting, which itself signals how narrow the bull-bear spread is (HSBC's $191 low versus Oppenheimer's $350 high, an ~83% gap). Prediction market Polymarket priced only a 25.5% probability that IBM beats consensus at its official July 22 report.
Gains: chipmakers absorbing the AI-infrastructure hardware demand, Intel and Nvidia; short-term arbitrageurs long the HSBC-style "synthetic IBM" basket (SAP, Accenture, HP, IonQ); IBM's quantum-computing narrative, which the crisis let management pivot the conversation toward — over $10 billion committed to quantum.
Losses: IBM shareholders, down $68.8B in a day; the entire enterprise software sector, dragged down alongside IBM (Microsoft, ServiceNow, Salesforce, Intuit, Accenture, DXC) even though their subscription revenue models with contracted backlog are a fundamentally different risk profile than IBM's legacy mainframe business; traditional IT consultancies exposed to clients now reordering priorities toward "stockpile hardware first, discuss software later."
This is largely an IBM-specific problem: the shortfall is concentrated in the aging mainframe and Transaction Processing lines, while rivals like ServiceNow sell subscription revenue with contracted backlog, a different kind of fragility entirely.
The revenue miss was only 3.7%; a 25% drop clearly overdrew on the bad news, and the probability-weighted valuation already shows the price broke below fair value.
Software revenue itself is still growing +5%, and management insists the core business isn't being disrupted by AI.
This isn't the first time — ServiceNow and IBM fell together on the exact same "AI eating software" narrative back in April. A second run of the same story suggests something structural, not one quarter's noise.
HSBC's "synthetic IBM" logic lands a real blow: if the same business exposure can be replicated with a cheaper basket, IBM's valuation premium doesn't hold.
Krishna himself admitted "numerous deals failed to close on time" — that's an execution problem, not just a timing shift, and execution problems don't usually vanish after a single quarter.
Overstated: "AI is disrupting IBM's software business." The actual shortfall concentrates in mainframe and the associated Transaction Processing line; software revenue overall is still growing +5%. Reading a one-quarter capex timing shift as a historic inflection point for "AI eating software" runs ahead of the data.
Overlooked: the exact same "AI eating software" script already played out this April, when ServiceNow and IBM fell together on the same theme. The market's reaction this time is nearly a carbon copy, yet almost no coverage flagged that this is the second instance, not new evidence appearing for the first time — a repeated narrative deserves more skepticism the second time around, not less.
What the crash really exposes is not "AI is killing software stocks," but that enterprise IT budgets are being temporarily rerouted by the AI-infrastructure land-grab — supply bottlenecks and expected price hikes in servers, memory and storage are pulling money that would have gone to software and consulting into hardware first. Whether that rerouting is a one-quarter blip or persists into next year determines whether this is one-time noise or a structural repricing. The first hard evidence arrives at IBM's official Q2 report on July 22, with ServiceNow and peers' next earnings serving as a second check. Until that answer arrives, the market already placed a 25%-sized bet on it — based on nothing more than a preliminary 3.7% miss and a CEO's letter.
When a single-quarter miss triggers a 25% crash, separate two things first: a timing shift (where this quarter's money went) and a structural repricing (whether that money ever comes back). Markets routinely trade the former as if it were the latter, especially when a ready-made grand narrative — "AI disrupts everything" — is available to reach for. The same discipline applies to your own decisions: when a sector falls a second time on "the same story," ask whether the story played out as advertised the first time. If it didn't, this is probably just a rerun of the same false alarm.