A 115-year-old giant lost a quarter of itself in a single day

On Tuesday, July 14, something broke in IBM's stock the moment U.S. markets opened. In one session it fell more than 25%, closing near $217. And this wasn't a garden-variety "ouch, that hurt" day. It was the single worst trading day in IBM's 115-year history. Want a reference point? October 19, 1987 — the infamous Black Monday. IBM dropped 23% that day, and this crash topped it. Roughly $67 billion in market value simply evaporated, dragging the company's worth below $205 billion.

What makes it even stranger is that this didn't happen on an official earnings date. IBM's full Q2 results were scheduled for July 22. But rather than wait, CEO Arvind Krishna got ahead of it with an unscheduled public letter admitting the preliminary numbers came in "worse than our expectations." The letter's opening line was blunt: "This quarter we faltered." It's not common for a large-cap CEO to fire off a confessional letter before the official print. That tells you how bad the numbers were — and how much he wanted to soften the blow in advance. Instead, the letter became the trigger.

And here's the truly unnerving part: the reason. IBM didn't stumble "because of AI" in the usual sense. It stumbled because AI made clients reroute the money they'd have spent with IBM toward hardware. That's not one company's accident — it may be the opening signal that the rush into AI infrastructure has started draining the wallet of the entire software industry. Let me unpack why this one matters, piece by piece.

The players — IBM, Krishna, and the hardware camp that smiled

IBM (International Business Machines) needs no introduction. For over a century it's been synonymous with enterprise IT, living on mainframes, enterprise software, and consulting. In recent years IBM has re-cast its identity as "hybrid cloud + AI software + consulting." Buying Red Hat for $34 billion in 2019 was the symbol of that pivot, and it has been pushing watsonx, its generative-AI platform, with a clear message: "We don't sell chips like Nvidia — we sell the software and services layer that actually lets enterprises use AI." That's exactly why IBM had been classified as an AI beneficiary. The logic was that the AI era would lift demand for software and consulting.

Arvind Krishna, an India-born engineer, took IBM's helm in 2020 and has been credited with reshaping the company around cloud and AI. So it was striking to watch him say, in effect, "we faltered." His explanation is the heart of this story, so here it is verbatim: "In the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases." He added: "While we anticipated some supply chain related impact in our expectations, we did not anticipate the magnitude of the capex reprioritization."

In plain terms: clients have finite IT budgets, and that money got vacuumed straight into a hardware land-grab. Exploding AI-datacenter demand made servers, storage, and memory scarce, and memory prices in particular spiked from late 2025. Memory heavyweights — Samsung, SK Hynix, Micron — shifted production toward high-value AI chips (HBM and the like), which effectively sold out ordinary server and storage memory through 2026 under long-term contracts. So enterprises rushed to buy hardware "before prices rise, before supply dries up," and pushed their software deals to the back burner.

That's why the real winners on this day sat on the other side of the table — the hardware and semiconductor camp. On the very day IBM collapsed, Nvidia rose about 4.1% and Intel about 4.5%. The market showed, in real time, that money flows toward whoever controls the AI-infrastructure bottleneck. The same news was a death sentence for one group and a fireworks show for another.

The core — by the numbers, this was a disaster

Let's start with the figures. IBM guided to roughly $17.2 billion in preliminary Q2 revenue. The problem is that Wall Street (the LSEG consensus) expected $17.86 billion — a shortfall of about $660 million. Adjusted earnings per share came in at $2.93, below the $3.01 consensus. On revenue that's about a 3.7% miss — and the fact that a miss that size sent the stock down 25% is the whole point. The market wasn't reacting to "this quarter's number." It was reacting to the structural threat underneath it.

Look at which businesses cracked and the picture sharpens. IBM flagged weakness specifically in software and infrastructure (mainframes and the like). Clients pouring money into securing hardware delayed IBM's new software contracts and renewals, and at the same time some customers went into wait-and-see mode — "we're not sure how AI will reshape our IT stack, so let's pause software buying." Those two forces converged to hit IBM's highest-margin businesses head-on.

And the shock didn't stay contained to IBM. The whole software sector wobbled with it. Salesforce (CRM) fell about 5%, ServiceNow (NOW) dropped nearly 7%, and large caps like Microsoft and Intuit slid between 3% and 5%. Consulting hurt even more: Accenture (ACN) fell about 8% and Cognizant (CTSH) about 7%. One analyst said IBM's warning would deliver a "devastating blow" to software and services names. Investors got scared that this capex rotation wasn't an IBM-only problem — that it could hit the entire software industry.

Item Figure / detail
July 14 stock drop ~-25% (close near $217) — worst day in 115-year history
Historical comparison Topped the -23% of Black Monday 1987
Market value wiped out ~$67 billion → valuation below $205 billion
Q2 preliminary revenue ~$17.2B (vs. $17.86B consensus, ~$660M short)
Q2 adjusted EPS $2.93 (below $3.01 consensus)
Software peers down ServiceNow -7%, Salesforce -5%, MSFT & Intuit -3~5%
Consulting peers down Accenture -8%, Cognizant -7%
Names that rose Nvidia +4.1%, Intel +4.5%
Official earnings date Originally July 22 (pre-disclosed via letter)

Let's be honest about one thing. IBM wants to frame this as a "temporary reprioritization" — the idea being that once clients finish stockpiling hardware, software spending eventually comes back. That could genuinely happen. But the reason the market lopped off 25% is fear that nobody can be sure whether this is temporary or structural. If it's a structural shift — if the center of gravity of IT budgets is moving permanently from software to hardware in the AI era — then it's not just IBM at risk; the entire valuation premise of the software industry gets shaky.

Who won and who lost

The biggest loser was IBM itself and its shareholders. Sixty-seven billion dollars of market value gone in a day says it all. IBM in particular attracts a lot of steady, income-focused investors thanks to its dividend, and a crash like this cracked its "safe old-tech" image. Krishna personally took a hit to his narrative of having put IBM back on a growth path for the AI era.

The software and consulting camp fell alongside it. Salesforce, ServiceNow, Accenture, Cognizant — none of them compete directly with IBM, yet all got beaten up together. The reason is simple: IBM's warning exposed a shared risk of "AI-infrastructure rush → delayed software spend." Investors basically asked, "So are you going to say the same thing next quarter?" and sold ahead of time.

On the flip side, the semiconductor and hardware camp cheered. Nvidia (+4.1%) and Intel (+4.5%) rose on the same day. That's symbolic. IBM's misfortune read as proof that "AI hardware demand is real." Whoever makes the servers, storage, and memory — and the chips inside them — is currently surrounded by clients saying "we'll pay whatever, just give us the units." The memory trio (Samsung, SK Hynix, Micron) are indirect beneficiaries too, since their shift toward AI-grade high-value memory like HBM is what set off this scarcity and price surge in the first place.

Hyperscalers and server makers are quiet winners as well. Server vendors like Dell and Supermicro, and the cloud giants building out their own datacenters, sit at the front line of the "infrastructure land-grab." Enterprise clients racing to hoard hardware translates directly into orders for them. Sure, they also carry the cost pressure of rising memory and chip prices — but demand itself is overflowing.

History rhymes — capex rushes make winners, and traps

We've seen this "capex rush" scene many times, and the endings keep diverging.

The first is the dot-com bubble (1999–2001). Back then companies threw money madly at telecom gear, servers, and fiber, insisting "the internet changes everything." Infrastructure names like Cisco briefly became the most valuable company on the planet. But much of that demand was overbuild — "lay it now, figure it out later" — and when the bubble burst, a huge chunk of that installed infrastructure sat idle for years (the infamous "dark fiber"). The lesson is clear: when the infrastructure land-grab is hot, the hardware sellers look like winners, but if that demand never converts into real revenue, an overcapacity hangover follows. Today's AI hardware rush isn't free of the same question: does this spending come back as actual AI profit?

The second is the 2018 semiconductor correction. In 2017–2018, a crypto-mining boom and datacenter buildout sent memory and GPU demand soaring, and a "buy before prices rise" mentality took over. Then demand cooled abruptly in the second half of 2018, memory prices crashed, and all that stockpiled inventory became a boomerang. It's a vivid case of how fast "panic buying on supply-shortage fear" can flip into "excess inventory." Companies pulling forward server and memory purchases ahead of expected price hikes right now could face the same inventory boomerang if AI demand disappoints.

There's a success story too: the 2020–2021 pandemic cloud supercycle. Remote work and a cloud surge spiked infrastructure spending, and those who invested pre-emptively in datacenters and chip capacity captured the lead for years afterward. In other words, if a capex rush is riding genuine structural demand, whoever invested early wins big. The trouble is that right now, nobody can say for certain whether this AI-infrastructure rush is dot-com-style (overbuild) or pandemic-style (structural). That uncertainty is exactly what got stamped onto IBM's stock as a 25% discount.

The counter-play — how the infrastructure camp locks this in

In the very trend that hammered IBM, the infrastructure camp will try to cement its advantage.

Microsoft sits in a delicate spot. MSFT also fell 3–5% on the day of IBM's warning, on worries that its software revenue (Office, Dynamics) faces the same threat. But Microsoft's counter is Azure. If it can get customers to rent AI infrastructure on Azure instead of hoarding hardware themselves, it can absorb the "infrastructure rush" as its own revenue. Microsoft is aggressively expanding AI-datacenter capex right now, effectively telling customers, "don't buy it — rent it from us."

AWS (Amazon) has a similar counter. The whole reason enterprises are scrambling to physically secure servers, storage, and memory is that they lack the infrastructure to run AI workloads. AWS wants to vacuum that demand into the cloud with its own chips (Trainium, Graviton) and massive datacenter footprint. The tighter the infrastructure squeeze, the stronger the "come to the cloud instead of buying" pitch from the big three. Ironically, the same trend that clobbered IBM can be an opportunity for the cloud majors.

Oracle plays a different angle. Oracle has been re-rated lately on AI infrastructure (OCI) and database cloud. Its bet is selling "AI-optimized infrastructure plus the software on top" as one package. When customers are weighing hardware against software and deferring the software, whoever bundles the two together may come out ahead.

There's a counter here IBM itself should learn. Its weakness was that "customers chose hardware in the hardware-vs-software trade-off." So IBM's comeback move is to reposition its own software (watsonx, Red Hat) as the essential layer that runs all that AI infrastructure more efficiently — "you hoarded the infrastructure? The software to actually run it is ours." Ultimately, the crux of this whole fight is one question: are hardware and software spending zero-sum, or does more hardware eventually pull software along with it? IBM wants to believe the latter. This quarter played out like the former.

So what actually changes

If you're an investor — the real message here isn't "one company missed earnings," it's "the AI-infrastructure rush can threaten software valuations." If you hold software or consulting names, watch closely for whether the next earnings season brings similar comments about "clients redirecting spend toward hardware." Conversely, remember that semis, servers, and memory are direct beneficiaries of this trend. But as history shows, everything hinges on whether this infrastructure demand is structural or a one-off stockpile.

If you're an enterprise IT decision-maker — you're right in the middle of that "infrastructure land-grab." What IBM's warning tells you is the reality: servers, storage, and memory are supply-constrained, prices are climbing, and everyone is pre-buying. If your organization plans to physically secure AI infrastructure, it's wise to bake lead times and price hikes into the budget now. At the same time, scrutinize whether "buy hardware first, software later" is actually rational. Infrastructure with no software or people to run it can turn into very expensive scrap metal.

If you're a developer or engineer — this news reconfirms that AI infrastructure is a genuine bottleneck. Severe server and memory shortages mean the fight for cloud GPUs and instances will stay fierce for a while. And the fact that even a company like IBM says "customers are pausing software purchases" signals that enterprises still aren't sure how to use AI — which means big opportunity for whoever builds AI software that actually earns its keep.

If you're a general user — no direct impact on your day today. But in the big picture it matters. Companies worldwide rerouting entire budgets to secure AI hardware means they feel AI infrastructure isn't overheated yet but rather insufficient. If that turns into genuinely useful AI services, great; if it ends in overbuild like the dot-com era, the eventual hangover (price corrections, pullbacks in investment) could ripple into the services you use.

🥄 Three Things You're Probably Wondering

— So why should I care? If you don't own IBM, it looks irrelevant. But this is the first major evidence that the AI-hardware land-grab has started draining the software industry's wallet. The price and speed of the cloud services you use, and the AI tools your company adopts, are shaped by this infrastructure rush.

— A 25% crash over a mere 3.7% revenue miss? The number alone looks excessive. But what the market sold wasn't "this quarter's figure" — it was the structural threat underneath. If clients picking hardware over software isn't temporary but an ongoing shift, the growth premise of the whole software industry wobbles, not just IBM's. That fear is what's packed into the 25%.

— Does this mean software companies are done? Nobody knows yet. IBM calls it a "temporary reprioritization" and expects software spend to return. Others worry the AI era permanently moves the center of gravity of IT budgets toward hardware. The winner of this debate will be decided by the next few quarters — especially whether Microsoft, Salesforce, and Oracle start saying the same thing.

References

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!