• The Latte
  • Posts
  • The SaaSpocalypse: Why First Principles Lead Back to the Physical World

The SaaSpocalypse: Why First Principles Lead Back to the Physical World

šŸ“‰ What Just Happened in Software Land

Call it the SaaSpocalypse. The S&P North American software index just suffered a 15% monthly drop, its worst since ’08. Wall Street isn’t just skeptical about SaaS anymore; it’s panicking. ā€œGet me outā€ isn’t a meme, it’s the tape.

Software stocks? Dumped indiscriminately. Legaltech? Obliterated after Anthropic demoed its AI paralegal. Reporting and analytics names? Down double digits in a day. Even the giants, Microsoft $MSFT ( ā–¼ 1.99% ), ServiceNow $NOW ( ā–¼ 7.74% ) and SAP $SAP ( ā–¼ 4.61% )  , can’t find a bid. It’s the existential vibe that has people rattled.

šŸŖ‘ Seat Compression and Software Moat Shrinkage

Let’s start with the obvious: AI is destroying the ā€œper seatā€ SaaS model. Anthropic launches a tool that does the work of 10 in-house lawyers, goodbye to 9 software licenses. Google’s Project Genie lets you prompt an entire game world into existence: suddenly, that game studio software bundle doesn’t look so essential.

Nobody’s immune. Even Microsoft, the chart-topping cloud titan, dropped 10% in a single day over AI and cloud spend concerns. Wall Street is worried: If AI can write code (ā€œvibe codingā€), summarize knowledge, and automate workflows, how deep are these moats, really?

  • ā€œShallower Moatsā€: Margins under pressure. Competition goes up, pricing power goes down, and the old license-and-seat-count model crumbles.

  • ā€œSeat Compressionā€: Each dollar invested in AI shrinks human seats required. That’s fewer licenses: potentially permanent if AI’s trajectory holds.

  • Negative Spiral: Less revenue, looser moats, less reinvestment, more disruption.

Everyone from private equity to active managers is searching for what’s truly defensible. It’s a sector-wide reckoning.

🚨 Panic or Opportunity? Both.

Technicians will tell you software is ā€œoversold enough for a bounce.ā€ Maybe. But conviction is at historic lows. Even bulls don’t know where the ā€œhold-your-noseā€ bottom is.

Here’s the reality: a few names will survive, maybe even thrive, on the far side of the AI transition. But separating winners from zombies? That’s become the market’s hardest problem. If Microsoft is struggling, what hope for the long tail?

Yet where others see carnage, I see clarity: this selloff is an x-ray. It reveals what’s truly durable when the digital layer gets knocked off its pedestal.

🧱 First Principles: The World You Can’t Disrupt With Python

Let’s go back to basics. What survives when AI commoditizes digital workflows? The stuff AI can’t replace. The ā€œatoms, not just bitsā€ part.

  • Power. You can’t generate electrons with a prompt.

  • Cooling. Liquid can’t be virtualized.

  • Storage. Data must physically live somewhere.

  • Networks. Someone wires and maintains the pipes.

This is first principles investing. The digital stack is only as strong as its physical foundation.

ā€œYou can code a workflow away. But you can’t AI your way out of needing a transformer, a substation, or a rack full of NAND flash.ā€

I’m watching money rotate into companies providing the bricks-and-mortar (literally) of the AI economy.

šŸ—ļø Meet The Physical Moat Crowd

While software gets abandoned at any price, look at the tape for picks and shovels. Quanta Services $PWR ( ā–² 1.8% )  isn’t glamorous, but it’s at the heart of the US power grid upgrade: everyone from hyperscalers to manufacturers is shoving money its way. Here’s the difference:

  • Quanta’s backlog: Record highs, because you can’t deploy a single GPU unless someone strings a new high-voltage line first.

  • SanDisk $SNDK ( ā–² 2.96% ): Crushed its Q2 on datacenter flash demand. AI doesn’t run on air. It runs on high-performance NAND. The market gets it: stock up 550% last year.

  • Sanmina $SANM ( ā–² 2.09% )  : Unknown to most, but rolling in contracts to build out the hardware racks AI lives on. (Full deep dive here)

Contrast this with most SaaS names:

  • Moats getting shallower by the week

  • Pricing power evaporates as AI tools ā€œvibeā€ their way through code and support tickets

  • Growth, when it comes, is from a shrinking base

Physical businesses? Their moats get wider as digitization accelerates. No one’s figured out how to automate building a transformer.

šŸ“ˆ The Conviction Flip: Get Me Out vs. Hold Forever

There’s a reason you see ā€œget me outā€ panic in software but ā€œhold foreverā€ mentality in infrastructure. It’s not just narrative, it’s cash flows and risk:

Software/SaaS

  • Revenue driven by user count: vulnerable to AI compression

  • Switching costs falling as every competitor uses the same foundation models

  • Refactoring a vertical SaaS tool is a week’s work for the right AI dev

Physical AI Infrastructure

  • Decade-plus contracts, regulatory barriers, enormous switching costs

  • Asset bases that require capex and expertise AI can’t substitute

  • Growing demand as every AI deployment needs more grid, more storage

The market may swing too far in both directions: but I know which camp has sleep-at-night qualities.

šŸ” A Simple Framework: What Can AI Never Replace?

Thinking about where to invest? Ask yourself, ā€œCould an LLM or an API make this business less necessary, or is its product tied to the ground, the grid, the real world?ā€

  • Infrastructure as a moat is deeper when it’s physical.

  • Digital workflows get automated. Laying fiber, pouring concrete, building a rack: AI can help, but can’t eliminate.

  • AI raises demand for power and storage at an exponential curve, regardless of whose SaaS is being displaced.

That’s why my personal radar is tuned to infrastructure-adjacent growth stocks: the ones that show up in capex plans, not just IT budgets.

šŸ› ļø Building for the Real AI Cycle

Everyone wants to play the AI cycle. Here’s the twist:

  • Most software businesses were levered to the old digital transformation: more seats, more automation, endless upward slope.

  • The new AI buildout is physical: more power, more cooling, more storage. And the multiplier effect on the physical world is only getting started.

Want the details? I’ve built out a 2026 AI Infrastructure Playbook tracking every investable layer in the stack: from grid modernization to high-performance memory, to the contractors bolting it all together.

āš ļø Risks to Watch

Nothing is bulletproof: not even transformers. What could go wrong for physical infrastructure plays?

  • Delayed projects if rates spike, permitting slowdowns, or public pushback

  • Cyclical capex: if the AI cycle slows, so does infrastructure spend

  • Regulatory risk: big utilities and infrastructure must dance with red tape

  • Commoditization at the hardware layer (less likely but not impossible with flash or server gear)

Still, when everyone flees digital exposure, the physical layer keeps winning. History says…follow the pipes, not the platforms.

šŸ”— Key Reads From The Infrastructure Angle

Want to dig deeper? Check out:

šŸ¤ The Takeaway

The SaaSpocalypse might be ugly for software, but it’s a timely reminder for all of us: first principles win, especially when disruption goes exponential.

If you’re rethinking allocations:

  • Move up the stack to hard-to-replace infrastructure

  • Prefer real-world moats to digital abstraction

  • Watch how AI cycles up demand for power, cooling, and storage

Stay curious, stay grounded, and remember: when in doubt, look for what AI literally can’t do.

George ā˜•ļø