The financial talking heads are staring at a "divergence" between hardware and software and calling it a temporary rotation. They’re wrong. Jim Cramer and the CNBC crowd want you to believe that software is merely catching its breath while Nvidia runs the marathon. They tell you to wait for the "catch-up trade."
Stop waiting. It isn't coming.
What we are witnessing isn't a lag. It’s a structural execution. The traditional SaaS (Software as a Service) model is being dismantled by the very intelligence it claims to sell. If you’re holding onto legacy seat-based software stocks expecting a 2021-style rebound, you aren't an investor. You're a museum curator.
The Seat Based Model is a Dead Man Walking
The "lazy consensus" argues that software is struggling because of high interest rates and cautious enterprise spending. That’s a surface-level symptom, not the disease. The real pathology is the death of the "seat."
For two decades, software companies grew by charging per user. Salesforce, ServiceNow, and Adobe built empires on the premise that more employees equals more licenses. But we’ve entered the era of the Degenerative Employee Count.
If an AI agent can do the work of five mid-level marketing managers, the enterprise doesn't need five Adobe licenses. They need one. Or zero, if they build a custom internal wrapper around an open-source model. The math for SaaS companies has flipped from $f(x) = growth$ to $f(x) = cannibalization$.
Every time a software CEO brags about integrating AI to "increase productivity," they are effectively telling you they are shrinking their own addressable market. They are trading $50,000 in recurring seat revenue for a $5,000 API credit. This isn't a "soft patch." It’s a fire sale of the old world.
The Hardware Supremacy is Not a Bubble
Critics love to compare the current AI infrastructure build-out to the fiber-optic craze of 1999. They see Nvidia’s margins and scream "unsustainable."
Here is what they miss: Fiber optics were passive. You laid the cable, and then you prayed someone would use it. Compute is active. GPU clusters aren't just "pipes"; they are the refineries of the new economy.
The divergence exists because hardware is capturing the Value of Intent, while software is stuck selling the Value of Interface.
When a company spends $100 million on an H100 cluster, they are buying the ability to create. When they spend $100 million on a legacy CRM, they are buying a place to store things they’ve already created. In a world of infinite generation, storage is a commodity. Compute is the only true hard currency.
I’ve seen CTOs at Fortune 500 firms slash their "digital transformation" budgets—the bread and butter of firms like Accenture and Salesforce—just to find the cash for more compute. They aren't "lagging" on software; they are abandoning it in favor of raw power.
The Fallacy of the AI Moat
The biggest lie in the valley right now is the "Data Moat."
Software companies claim they will win because they have proprietary customer data. "We have twenty years of your sales history," they say. "Therefore, our AI will be better."
This is nonsense. LLMs (Large Language Models) have proven that reasoning capability is more important than specific data silos. A generic, high-reasoning model can parse a messy CSV file in seconds and provide better insights than a rigid, proprietary AI tool locked inside a legacy software ecosystem.
The "moat" is actually a cage. These companies are desperate to keep you locked into their UI, but the UI is exactly what AI replaces. We are moving toward a Headless Enterprise. You won't log into a dashboard to see your churn rate. You will ask a local agent to pull the data, analyze it, and execute a retention campaign.
The software company that used to charge you for the dashboard is now irrelevant. They are a database with a dying business model.
Gross Margins are a Fantasy
Software was the darling of Wall Street because of those beautiful 80% plus gross margins. You build it once, you sell it a million times.
AI incinerates those margins.
Running a traditional SaaS platform costs pennies in server time. Running a generative AI feature costs a fortune in inference. Every time a user interacts with an "AI Assistant" inside a software tool, the provider’s margin takes a hit.
The Unit Economics of the Collapse
Imagine a scenario where a SaaS company charges $100 a month.
- Old World: Server costs are $2. Gross profit is $98.
- AI World: Inference costs for the "smart features" are $40. Gross profit is $60.
To maintain their old valuation multiples, these companies have to raise prices. But they can’t. Why? because the underlying models (GPT-4, Claude, Llama 3) are becoming cheaper and more accessible every six months.
They are caught in a pincer movement: rising COGS (Cost of Goods Sold) and falling pricing power. This is the "nuance" the bulls are ignoring. They see "AI integration" as a feature; I see it as a margin-eating parasite.
The Myth of the "Easy" Enterprise Transition
Cramer likes to talk about the "Blue Chip" software names as if they are too big to fail. But size is a liability during a shift in physics.
I’ve spent time in the trenches with enterprise sales teams. It takes them 18 months to close a deal. By the time they’ve installed a "new" AI-powered module for a client, the underlying technology is three generations obsolete.
The "lag" isn't a delay in buying; it's a delay in realization. Enterprises are realizing they don't need a middleman to talk to an LLM.
How to Actually Play the Divergence
If you want to survive this, you have to stop looking for the "next Microsoft" and start looking for the "Software Killers."
- Vertical Integration over Horizontal SaaS: General-purpose software (spreadsheets, CRM, word processing) is dead. The only software that survives is that which is tied to physical-world complexity that AI can't yet simulate—think specialized CAD for aerospace or HIPAA-locked medical systems.
- The "Bring Your Own Model" (BYOM) Movement: Watch for companies that provide the infrastructure for businesses to run their own models. They are the new utilities.
- Short the Middlemen: Any company that earns a margin by simply "wrapping" an API and calling it a product is a zero.
The Brutal Reality of the P/E Ratio
People ask, "Is Salesforce cheap at 25x earnings?"
The answer is: It doesn't matter what the P/E is if the 'E' is headed for a cliff. We are valuing these companies based on a world where humans do the clicking. That world is ending.
The divergence isn't a gap to be closed. It’s a cliff. On one side, you have the providers of the new fire (Chips, Power, Cooling). On the other, you have the people who used to sell candles.
The candles are on sale. That doesn't mean you should buy them.
Stop listening to analysts who treat "Tech" as a monolithic block. Tech is currently at war with itself. Hardware is winning because it is the only thing that is tangibly real in an increasingly synthetic economy. Software is losing because it was built for a labor force that is being automated out of existence.
If you’re still holding the laggards, you aren't "investing in value." You are subsidizing a funeral.
The software "lag" will continue until the valuations reflect the new reality: that most of these companies are features, not platforms, and their lunch is currently being eaten by a line of code written by a machine.
Sell the "recovery." Buy the disruption.
Move your capital to the source of power, or watch it evaporate in the "cloud."
