For ten years I ran an e-commerce business. Three warehouses — central storage, a distribution center, a physical store. I needed inventory management across all three. Nothing more. A clean Shopify app built by a small team in Singapore handled it. Seventy dollars a month. It did exactly what I needed and nothing I didn’t.
Then the Singapore company got acquired by Intuit. Worked fine for about a year. One day an email arrived: they were “concentrating on features in the US suite.” My inventory system was being sunset. I was left scrambling.
I could not find a standalone inventory tool to replace it. The replacement came bundled with production management features I would never use: a hundred and eighty dollars a month for the “multiple warehouses plan.” I was paying for someone else’s roadmap.
Then the second acquisition. The new owner wanted four hundred dollars a month for the “full suite” — analytics dashboards I didn’t need, factory assembly BOMs and further functionality that solved problems I didn’t have. They initially respected the old price as a “courtesy.” Two years later: “We need to charge you the full amount.”
I pressed cancel. Spent a weekend. Built my own inventory system.
It was better because it fit my processes, my workers, and my warehouses. The difference wasn’t my ability as a developer. Those teams were building for a different set of needs. A weekend with AI let me replace a decade of vendor dependency with something that did the job I needed.
That experience changed how I assess the value of SaaS products.
Two Trillion Dollars of Agreement
The market figured this out before most SaaS CEOs did. In February 2026, over $1 trillion in market capitalization vanished from software stocks in a single week. Within thirty days, approximately $2 trillion had evaporated. Median EV/Revenue multiples for public SaaS companies collapsed to 5.1x, down from pandemic peaks of 18-19x. Not a correction. A category repricing.
Buyer and vendor assessments differ: 80% of software buyers now cite AI-driven commoditization as the top risk to SaaS valuations. Only 25% of SaaS CEOs perceive the same threat. Most of the CEOs surveyed did not share the buyers’ assessment of the threat.
The Middleware Trap
When model providers collapse the stack above them, every company building on models it does not control gets squeezed. Your product sits between the customer and the platform. The platform gets more capable every quarter. Your layer gets thinner.
Could a competent engineer rebuild your core value with a foundation model API and a weekend of work? For a growing number of feature-layer SaaS products, the answer is yes. A business in that position depends on a feature its customer may now be able to build. Deeply embedded systems of record face a different, slower reckoning: years of accumulated data and integrations make replacement a much larger undertaking.
The economics break from both directions. When one AI-equipped user accomplishes the work of five, per-seat pricing — the model that underpinned two decades of SaaS economics — starts to collapse. Customers reduce seats. Every quarter, the same workforce needs fewer licenses. Meanwhile, CIOs consolidate vendor portfolios. The “best of breed” era is over. How many SaaS pitch decks still say “powered by GPT-4” as though that is a moat rather than a confession?
The Chegg Postmortem
Chegg provides one example of this exposure. Publicly traded EdTech company valued at $14.7 billion at its 2021 peak. Its product: answering students’ homework questions for $14.95 a month.
ChatGPT launched. Students got the same answers instantly, for free, with better explanations. Chegg’s stock collapsed 99% — from $113 to roughly $1. Subscribers dropped 31% year-over-year. Revenue fell 30%.
Chegg’s response was to launch CheggMate, powered by GPT-4. Its response depended on the platform competing with its core service. It bought time. It changed nothing.
The sectors closest to the automation frontier — Sales & Marketing, HR, Communication & Collaboration — are showing the strongest repricing reactions. The test is simple: if a general-purpose AI platform can replicate your core value proposition, you are Chegg.
The Catch
Global SaaS spending is projected to grow from $318 billion in 2025 to $576 billion by 2029 — an 81% increase. Calling this a “collapse” looks wrong on the surface.
But the distinction that kills companies is composition. The winners inside that $576 billion look nothing like the SaaS of 2015-2023. The growth includes AI-native platforms and vertical SaaS with deep domain expertise. It does not save the thousands of point-solution companies whose entire value proposition is a feature that foundation models now offer as a capability.
Switching costs are real friction. Enterprise customers with years of accumulated data, custom integrations, and trained users do not migrate overnight. But switching costs delay the collapse — they do not prevent it. Every renewal cycle becomes a decision point where the AI-native alternative is stronger than the last time the CIO evaluated it.
And yes, AI creates new SaaS categories. Creative destruction is real. But here is the velocity difference that previous platform shifts did not have: mobile-era SaaS categories remained defensible for five to ten years. AI-era categories may commoditize in twelve to eighteen months — foundation model providers absorb new capabilities faster than startups can build moats. The market grows. The composition changes. That distinction is where companies die.
Domain Knowledge: The Only Surviving Moat
Some businesses can benefit from the falling cost of software development.
The Jevons Paradox — as resource efficiency increases, total consumption rises rather than falls — offers a useful analogy for software markets, though not a proven law for this domain. As AI lowers the cost of building software, it becomes viable to serve markets that were previously too small to staff. The niches too small for platform economics become the opportunity for domain experts who understand the specific workflow, the specific regulation, the specific physical process.
Veeva Systems is the proof. Pharmaceutical regulatory intelligence delivered via SaaS. Revenue of $2.75 billion, growing 16% year-over-year, with 45% non-GAAP operating margins. While the SaaSpocalypse wiped trillions from horizontal SaaS valuations, Veeva grew — because its value is not features. It is encoded knowledge of FDA regulations, drug development workflows, and pharmaceutical commercialization that no foundation model can bootstrap.
Procore took the opposite approach and arrived at the same destination. Rather than waiting for AI to absorb its construction management platform, it acquired Datagrid — a vertical AI firm specializing in AI agents for contractors. Domain experts becoming AI acquirers, not AI victims.
The vertical SaaS market is projected to grow from $133.5 billion to $194.0 billion while horizontal SaaS faces absorption. Investors are reallocating capital accordingly: away from thin workflow layers, toward businesses that own workflows, proprietary data, and domain expertise.
But domain knowledge alone is not enough. Static domain knowledge — the kind that can be codified and taught — is replicable by AI-equipped competitors. The moat exists only when domain knowledge compounds through customer interaction: proprietary data flywheels, regulatory edge cases accumulated over years, workflow exceptions no competitor can bootstrap from zero. Gartner predicts 35% of point-product SaaS tools will be replaced by AI agents or absorbed within larger ecosystems by 2030.
Assess each advantage against what a new competitor could reproduce.
The SaaS Absorption Index
Use five questions to assess exposure to platform competition.
Feature Replicability. Could a competent engineer rebuild your core value with a foundation model API and a weekend? If more than 60% of your user-facing features map to general AI capabilities, your absorption risk is critical. A beautifully designed wrapper around a commodity is still a commodity.
Data Compounding. What proprietary data does your product generate that does not exist in any public dataset? If a new competitor could accumulate the equivalent in under eighteen months, the moat is insufficient.
Switching Cost Depth. Are your switching costs genuine workflow embedding or contractual friction? A three-year enterprise contract delays churn at renewal — that is a reprieve, not a moat.
Niche Specificity. Does your niche require specialized regulatory knowledge, physical-world process integration, or relationship-dependent workflows? If your TAM exceeds $500 million, major AI platforms will eventually target it. Serving “marketing agencies” is a label, not a moat.
Pricing Model Resilience. If your average customer reduces headcount by 40% through AI adoption, what happens to your revenue? If it drops proportionally, your pricing model is fatally coupled to the seat-count economy.
The Singapore plugin is gone. The $180 replacement is gone. The $400 “full suite” is gone. The weekend build runs perfectly.
Use the five questions to identify what customers would still pay for if the feature layer became widely available. My warehouses still needed an inventory system. What changed was who could build one that fit.