SaaS Eats Itself

For most of the last two decades, software-as-a-service was the cleanest business model in modern capitalism. Investors paid twenty and thirty times revenue for it because the economics were that good and the moat was that wide. Per-seat licensing turned every customer’s headcount growth into vendor revenue growth. The product compounded as the company grew. Annual prepaid contracts were effectively free loans. The whole arrangement was beautiful.

In a single week this February, more than a trillion dollars of software market capitalization was wiped out. JP Morgan called it the largest non-recessionary drawdown in software in more than thirty years. Goldman Sachs compared software’s future to the newspaper industry, warning this may be “the end of the beginning” of a long decline. Jefferies analyst Jeffrey Favuzza had a name ready for it. The SaaSpocalypse.

The headlines treated it as a shock. It was not. The math had been visible to anyone willing to do it for two years.

Here is what I keep seeing.

The same vendors that built their valuations on per-seat licensing are racing to ship autonomous AI agents that perform the work of the seats. The agents are pitched as productivity multipliers. They are described as helpers. The customer reads the pitch and does the math anyway. If one agent can do the work of ten coordinators, the customer does not buy a hundred extra licenses. The customer cancels eighty-five of the existing ones.

That is not a hypothetical. It is already happening at quiet scale across logistics, support, recruiting, sales operations, and back-office finance. The contract that used to be worth a million dollars a year is now worth a hundred and fifty thousand. The vendor’s reward for engineering the breakthrough is an eighty-five percent contract haircut.

Tomasz Tunguz at Theory Ventures saw this and asked the right question more than a year ago, in a piece titled bluntly “No SaaS! How AI Agents Will Change Software Pricing.” What does a software seat mean when a human is no longer operating the software? He laid out the alternatives — triple the per-seat price, move to usage-based, charge for performance — and was honest about the fact that each one breaks something the SaaS model had quietly depended on for twenty years. Predictability. CFO modelability. The clean compound of growing customers producing growing contracts. The reason multiples were so high was that revenue was so smooth. Lose the smoothness, lose the multiple.

This is the per-seat suicide trap. The product that justifies the next round of pricing also destroys the pricing model the company is currently sitting on. Every vendor that ships an effective agent triggers a cancellation cycle from its own customers.

The pivot the industry is attempting is consumption pricing. Pay per automated conversation. Pay per resolved ticket. Pay per outcome. The math sounds rational. The reality is chaotic. A per-seat license is predictable to the penny — the CFO can model it for the year. A consumption price is a variable input sitting on top of a compute cost the vendor cannot fully predict. Both sides of the contract end up watching a meter run. Neither side likes it.

Worse, consumption pricing turns enterprise software companies into utilities. Utilities have thin margins. They get regulated. They compete on price. They do not get twenty times revenue multiples. The financial markets understand this even if the press releases do not. That is part of what Goldman was naming when they reached for the newspaper comparison. The newspapers did not die because nobody wanted news. They died because the unit economics they were sitting on stopped existing.

There is a deeper pattern underneath the pricing turmoil.

Sarah Tavel at Benchmark named it early. Sell work, not software. The argument is that LLMs let founders sell the work itself rather than software to improve an end-user’s productivity. Sequoia Capital made it institutional this March in their thesis paper “Services: The New Software.” The contractual surface stops being software access and starts being finalized outcome. EvenUp prices on a per-case basis. Harper captures brokerage commission outright instead of selling a CRM to the human broker. Intercom charges only for resolved tickets. The vendor is paid when the work is delivered. The software is incidental.

This sounds like progress to a venture capitalist. To the incumbent enterprise software company, it is an obituary written in someone else’s voice.

The thing that used to make a piece of SaaS valuable was that the code was hard to build. The code was hard to build because writing software was an expensive, multi-person, multi-quarter undertaking. That assumption is now broken. A capable developer with a capable model can ship in a weekend what used to take a team six months. The proprietary application layer — the part of the stack vendors charged the most to protect — is becoming the cheapest part of the stack.

When code becomes cheap, the application layer becomes a commodity. When the application layer becomes a commodity, the only durable value sits underneath it. The data the customer owns. The workflows the customer runs. The context that describes how the customer’s business actually operates. That layer cannot be replicated by a vendor in a weekend, because that layer was never the vendor’s to begin with.

Tech critic Ed Zitron has been writing this story under a different label for two years. He calls it the Rot Economy. The argument is that growth-at-all-costs has replaced genuine value creation, and that the rot eventually reaches the income statement. The SaaSpocalypse is the rot reaching the income statement. It is what happens when the gap between what the dashboard reports and what the customer actually receives gets large enough that the market notices. The customer notices first. The market lags. Then the market crashes.

Risk theorist Nassim Nicholas Taleb has long argued that hyper-centralized, hyper-efficient systems do not collapse gracefully. They collapse all at once. Antifragility comes from distributed exposure to small shocks. Fragility comes from concentrating exposure in a single dominant model. The enterprise SaaS sector spent two decades concentrating exposure in a single pricing model. The shock arrived in February. It will not be the last one.

The companies that understand this are quietly stopping their next big SaaS purchase. They are looking at the seats they still pay for and asking which ones they would actually replace if they had to. They are starting to build the layer underneath their stack on infrastructure they own, or rent on their own terms. They are not anti-software. They are anti-being-a-permanent-tenant in someone else’s system.

The two-decade arrangement is reversing. The vendor used to hold the upper hand because the code was expensive. Now the customer holds the upper hand because the code is not. The per-seat model that financed the cloud era is about to be cancelled by the cloud era’s own next product.

Here is what I keep coming back to.

The customer who notices this early sets up differently. They keep their data close. They use the frontier models where the frontier earns its place. They run the rest of their operation on infrastructure they own, with a context layer that actually knows the company. They treat per-seat SaaS as a cost line to manage down, not a category to expand. They are operationally calmer for it, and they are not paying for an eighty-five percent contract their counterpart vendor has not yet figured out is going away.

The supply-side shock is here. The pricing pivot is chaotic. The customers are doing the math. The vendors are running out of room to pretend.

Jefferies named the moment. Goldman named the analog. JP Morgan named the scale. Sequoia named the structural shift. Tavel named the pricing pivot two years early. Tunguz named the per-seat question before the market cared. Zitron has been documenting the rot inside the dashboards. Taleb has been explaining for fifteen years why concentrated systems end the way February ended.

The cloud era’s strongest business model is the one that built it. The next era’s strongest business model has not been named yet. The vendors who survive will be the ones who name it first and build a business honest to it. The ones who do not will keep using the old language until their income statements stop letting them.