Did IBM Miss?

The Wrong Kind of Good News

In the AI-obsessed market of 2025, there are apparently two kinds of good news. There is "good news," where a company beats earnings estimates and raises its forecast. And then there is "AI-cloud-hyper-growth good news," which is the only kind anyone actually cares about.

On Thursday, IBM had the misfortune of only delivering the first kind. The company reported a solid quarter, powered by a 17% surge in its mainframe business. It was strong enough that IBM lifted its outlook for the entire year. The market's initial, knee-jerk reaction was to send the stock dropping more than 6%.

Which seems harsh, because for most of this year, Wall Street had been enthusiastically buying into IBM's specific and surprisingly plausible AI story. The pitch goes like this: IBM isn't trying to be another generic cloud provider for startups. It's leveraging its deep, historical strength in mainframes and high-security computing for the enterprise clients that still run the world. When a big bank wants to run an AI model for real-time fraud detection, it doesn't want to send its most sensitive transactional data to a public cloud. It wants to do it right there, on the box it already trusts. And IBM's new Z mainframe, with its custom Telum II AI accelerator chip, is designed to do exactly that.

You wrap this specialized hardware story in a software layer with its watsonx platform, and suddenly you have a generative AI "book of business" that has quietly grown to over $9.5 billion. And for most of 2025, this narrative worked beautifully. It was a coherent strategy for a legacy giant to find new relevance in the AI boom, and it's why the stock was up nearly 30% year-to-date before the earnings report. It was a bet on a very specific kind of "AI-ready-infrastructure good news."

So if the story was so good, why did the market, after cheering for months, suddenly send the stock tumbling 6%? Because what the market heard on Thursday was just... "good news." And it came with a hint of bad news in the one area it cares about most. To understand the disappointment, you have to ignore the successful parts of the report and focus on the one number that mattered. As Reuters noted, the anxiety was all about a slight deceleration in the cloud:

Sluggish expansion in IBM's cloud business, housed within the software segment, raised concerns about the company's ability to take advantage of demand for AI... The company's sales growth in the hybrid cloud unit - known as Red Hat - decreased to 14% from 16% in the previous quarter.

"IBM's software performance and outlook tend to carry more weight than the rest of the business due to the level of earnings contribution and value the business represents," said J.P. Morgan analysts.

There you have it. IBM's profitable, legacy mainframe business is booming, but the part of its business that competes in the AI-driven cloud arms race decelerated from a robust 16% to a still-robust 14%. In the current environment, this is a fatal flaw.

And that brings us to SAP. Europe's largest software maker also reported solid, if not spectacular, numbers. Its cloud backlog—a measure of future contracted revenue—grew a very healthy 27%. But its stock also slipped initially. Why? Because while its results were good, they weren't Oracle good.

Here's Reuters:

SAP said it expected cloud revenue for 2025 to be at the lower end of its forecast of 21.6 billion to 21.9 billion euros...

"We did not see an 'Oracle moment', but some fears about a weak report looming did not come true either," a local trader said, referring to Oracle's upbeat view on its cloud infrastructure revenue.

This is the new, impossibly high bar for success. An "Oracle moment" refers to Oracle's recent blockbuster report, where it revealed a massive $300 billion AI contract with OpenAI, sending its stock into the stratosphere. The new standard for an enterprise software company isn't just "Are you growing?" It's "Are you growing at a dizzying, AI-fueled, hyper-scale pace that suggests you are the primary landlord for the coming AI revolution?"

If the answer is no, the market's first instinct will be disinterest. (The fact that both IBM and SAP's stocks recovered most of their losses in the following days is its own story; it suggests that while the market's initial passion is reserved for the AI hyper-growth narrative, its second, more sober thought is that a profitable, growing software giant is still a pretty good business.) But that first, reflexive reaction tells you everything about where Wall Street's heart is right now. The only question that matters for the old guard of enterprise tech is, "How's your AI cloud business?" And if the answer isn't "growing to the moon," you're telling a perfectly good story to the wrong audience.

This also raises another question: is this a temporary market obsession about AI, or is a fundamental economic divide opening up between the companies that build the AI "stack" and those that sell "apps" on top of it? For our premium subscribers, we dig into why all the value is currently flowing to the bottom of the stack.

Extended Analysis: The Great AI Divide

The Friday Download

Connecting the micro to the macro

AI Workforce Restructuring

• The Trend: The impact of automation on the workforce is rapidly moving from theory to reality, as internal documents reveal Amazon's long-term plan to replace over 500,000 jobs with robots (NYT), Meta explicitly tells white-collar staff their roles are being eliminated by technology (Business Insider), and Amazon simultaneously tests AR glasses to augment its human drivers (Reuters).
• ARPU's Take: These signals reveal the complex nature of the AI-driven labor shift. Amazon's long-term plan confirms the strategic goal is mass replacement of blue-collar roles to decouple growth from headcount. Meta's memo is a watershed moment, providing a transparent case of direct white-collar displacement by automation. Meanwhile, Amazon's AR glasses show the "augmentation" track, where remaining human workers will be transformed into hyper-efficient, algorithmically-guided operators.
• The Implication: The future of work is not a simple binary of "human vs. robot" but a deep restructuring involving simultaneous replacement, augmentation, and displacement. This signals that the market value of routine cognitive and manual labor is set to plummet, forcing a painful transition for the workforce. For corporations, the key competitive battleground will be mastering the blend of human and machine labor to achieve maximum efficiency, while navigating the immense social and political backlash that this large-scale transformation will inevitably create.

Semiconductors' Custom Silicon Counter-Attack

• The Trend: The AI chip market is bifurcating as hyperscalers validate their custom silicon, highlighted by Google's massive deal to supply up to a million TPU chips to Anthropic (Bloomberg) and biotech firm Metagenomi tapping Amazon's Inferentia chips for a significant cost advantage in scientific computing (Reuters).
• ARPU's Take: These are proof points for the multi-billion dollar, decade-long custom chip strategies at Google and Amazon. The Anthropic deal is a "kingmaker" moment, publicly validating Google's TPUs as a credible, at-scale alternative for training frontier AI models. Metagenomi's use of AWS Inferentia is a crucial beachhead into a high-value scientific vertical, proving custom chips offer compelling price-performance beyond just large language models. This is a direct response to the market's desperate need for non-Nvidia compute.
• The Implication: The AI accelerator market is evolving into a duopoly: Nvidia's general-purpose GPUs versus the hyperscalers' specialized, vertically integrated custom silicon (ASICs). This creates a powerful competitive moat for Google Cloud and AWS, as they are no longer just renting servers but offering access to unique, cost-advantaged hardware. This will give customers real choice for the first time, increase pricing pressure on Nvidia, and make vertical integration—from the silicon to the software—the new standard for competing at the top of the AI infrastructure market

You've just seen a sample of our work. The ARPU Intelligence Desk is the complete system for professionals who need to go deeper. Members get the weekly Signal Stream, the monthly Competitive Radar, our Briefing Library, and M&A Monitor. If this level of intelligence is a fit for your needs, we invite you to request a brief demo.

P.S. Got a specific intelligence challenge not covered here? Drop us an email at arpu@hedder.com—we occasionally explore ideas to inform our core coverage.