IBM stock just posted its biggest single-day drop in 115 years — falling 25% in a single session. The reason? Customers are shifting their technology budgets away from IBM's mainframe business and redirecting that capital toward AI hardware, GPUs, and hyperscale cloud computing. IBM is simply not a major winner in the categories where AI spending is currently flowing, and the market punished the stock brutally for it.
Why Did IBM Stock Crash 25% in One Day?
The high-level answer is straightforward: IBM reset the narrative around its server and mainframe business, signaling that enterprise customers are reallocating their technology spending toward the physical AI buildout. AI dollars right now are flowing into GPUs, memory, networking, hyperscale cloud, and frontier model inference — and IBM is not a meaningful player in any of those categories.
0:08
IBM stock one-week chart showing the 25% single-day collapse
Watch at 0:08 →
To put it in the language that tech investors like Brad Gerstner and Gavin Baker use, IBM is not well-positioned on the token path. The Z17 mainframe cycle had been surprisingly solid heading into 2024, and the stock had actually nearly doubled since the launch of ChatGPT — a genuinely impressive run for a company many had written off. But today's reset revealed that the tailwind was running out. Customers aren't abandoning IBM entirely, but they are giving a shrinking share of their tech budgets to Big Blue. That's a serious problem for a business built on long-term customer lock-in.
How Did IBM Go From Punch Cards to AI Software?
To understand why today's drop matters, it helps to understand what IBM actually is — because it's been many different companies over its long history.
IBM, which stands for International Business Machines, started with a simple foundational insight: businesses will continuously pay to automate recordkeeping. The company began by building punch card tabulating machines, clocks, and mechanical information processors. That core thesis — sell businesses machines that handle data — has guided the company ever since.
The modern IBM everyone knows was born in 1964 with the System/360 — a compatible family of mainframe devices. The brilliance of System/360 wasn't just the hardware. It was the upgrade path. You didn't have to rip out your entire infrastructure to get more storage or more compute. You upgraded piece by piece, which created enormous switching costs and locked customers in for decades. Banks, insurers, airlines, manufacturers, and governments all became IBM's captive clients.
3:45
IBM's three business segments: Software, Consulting, and Infrastructure with margin breakdown
Watch at 3:45 →
Then came the PC era — and IBM accidentally handed the future to two smaller companies. The IBM PC launched in 1981, running Microsoft's Windows on Intel chips. At the time, IBM was doing $30 billion in revenue. Microsoft had about 120 employees and $17 million in sales. Intel was doing less than $1 billion. Within a decade, both companies would dwarf IBM. The PC market fractured, clone makers flooded in, and Apple launched its famous anti-IBM campaign. Suddenly, IBM's model was broken.
Lou Gerstner became CEO in 1993 and rejected proposals to break up the company. His famous line: we do not necessarily need to manufacture every piece of technology — we need to be the company that makes all of it work together. That pivoted IBM into a massive services and consulting organization, which is largely what it remained for the next three decades.
What Does IBM Actually Do Today?
Today's IBM is best understood as three distinct businesses operating under one roof:
- Software (44% of revenue): This is the crown jewel — 80% gross margins, anchored by the Red Hat acquisition and enterprise tools built around hybrid cloud and AI.
- Consulting (31% of revenue): The legacy of Gerstner's pivot. Decent revenues, but under 30% gross margins — lower quality than the software business and more exposed to headcount and pricing pressure.
- Infrastructure (23% of revenue): Just under 60% gross margins. This is the mainframe business — the one that just caused today's crash.
For the last three years, the stock was up 77% before dividends. The Red Hat bet was starting to pay off. But the infrastructure reset today revealed just how dependent the bull case was on a mainframe upgrade cycle that is now running out of fuel.
12:30
Demis Hassabis's proposed AI regulatory framework outlined on screen
Watch at 12:30 →
What Is IBM Red Hat and Was It Worth $34 Billion?
In 2019, IBM acquired Red Hat for $34 billion — one of the largest software acquisitions in history. Red Hat's flagship product is OpenShift, an enterprise Kubernetes platform that orchestrates workloads across multiple servers and cloud environments.
In plain terms: Red Hat helps large enterprises run software containers across hybrid cloud environments. That's genuinely useful in an AI world where companies are running workloads across on-premise servers, private clouds, and public clouds simultaneously.
IBM also spun off its traditional managed infrastructure outsourcing business in 2021, which was meant to clean up the portfolio and let the higher-margin software and consulting businesses shine. The Red Hat acquisition is IBM's strongest claim to AI-era relevance — but with so many companies offering AI capabilities up and down the stack, IBM is getting squeezed on all sides.
Is IBM Winning or Losing in the AI Era?
The honest answer: losing share, even if not losing outright. IBM does have real assets. Red Hat OpenShift is a legitimate enterprise platform with strong adoption. The software business has excellent margins. And mainframes, despite the current cycle slowdown, still process enormous amounts of the world's financial transactions — they aren't going away tomorrow.
But the AI spending boom is creating winners and losers fast, and IBM is not in the winner's column for the current wave. Nvidia, the hyperscalers, memory manufacturers, and networking companies are capturing the lion's share of AI capex. IBM is watching its customers redirect technology budgets away from mainframe refreshes and toward GPU clusters. That's a structural headwind that a consulting contract or a Red Hat deal can only partially offset.
Should the US Create a Frontier AI Regulatory Body?
Separately, Google DeepMind CEO Demis Hassabis — a Nobel laureate — published a detailed proposal calling for a US-led standards body to test frontier AI models before release. His argument: society has a narrow window to prepare for technology advancing at historic speed, and urgent action from international regulators is needed.
The specific proposals Hassabis outlined include:
- Create a US frontier AI standards body, likely a beefed-up version of the existing CAISI (Center for AI Standards and Innovation) under the Commerce Department
- Define and regularly update benchmarks to determine which models qualify as frontier-class
- Require frontier labs to submit models for testing up to 30 days before public release
- Test models for cyber, biological, nuclear, deception, autonomy, and guardrail-bypassing capabilities
- Apply rules to all frontier models deployed in the US, including foreign and open-source models, while exempting smaller models
- Coordinate slowdowns among frontier labs if testing reveals serious risks
- Develop an international system of shared frontier AI standards
The core tension here is familiar: concrete regulatory frameworks tend to favor large, well-resourced incumbents who can afford compliance teams and Washington lobbying offices. Small open-source projects often can't. There's also the thorny question of open-source models from China — if a model like Kimi K2 gets released openly, how exactly does a US regulatory body control its proliferation across GitHub and Hugging Face?
A more actionable approach, arguably, would be trigger-based policy proposals — the equivalent of saying: if the unemployment rate hits 10%, send checks of $X to households earning under $Y. Concrete scenarios with concrete responses tend to be more legislatively useful than general warnings about frontier risk. But Hassabis's proposal is at least the most detailed intervention yet from a major lab executive, and it carries weight coming from inside Google's AI operation.
Why Did New York Put a Moratorium on AI Data Centers?
New York Governor Kathy Hochul signed an executive order placing a one-year pause on new large-scale AI data centers in the state. The moratorium takes effect while New York develops a regulatory framework and conducts environmental impact assessments covering energy demand, water use, water quality, air quality, and effects on the electric grid.
The tech industry immediately pushed back, arguing the move will cost local communities jobs and weaken America's position in the global AI race. Critics have drawn parallels to past US policy errors — nuclear energy restrictions and manufacturing offshoring — where regulatory caution ultimately moved economic activity overseas rather than stopping it.
If the order stands, New York would become the first state to impose a broad moratorium on large-scale AI data centers. The irony is that New York hasn't seen a massive data center boom compared to states like Texas and Virginia — so the practical impact may be limited. But as a signal, it's significant. And as Ken Griffin noted recently, if these restrictions spread nationally, the economic activity doesn't disappear — it just moves to other states, or other countries.
On the aesthetic front: architects are now designing data centers that look like tech campuses and art museums rather than windowless concrete blocks. If data centers can be clean, quiet, and visually integrated into their communities, the NIMBY backlash may soften. Whether regulators wait to find out is another question entirely.
