Per-seat SaaS pricing is giving way to outcome-based models, and the revenue pool is an order of magnitude larger

Per-seat pricing built the SaaS industry. It also assumed that software was used by humans who logged in, worked, and logged out. Investor Elad Gil argues that assumption is collapsing. As he puts it, the industry is moving from seats and SaaS into a world where it is selling human labor equivalents. That framing, from access to labor, captures the commercial logic that several builders and investors are now acting on.

Kareem Amin, who leads Clay, has already made the move. Clay charges for usage rather than per seat, a deliberate choice to tie the price to productivity delivered rather than headcount enrolled. The same logic is spreading into enterprise incumbents. Philipp Herzig, who oversees artificial intelligence at SAP, describes a planned stepwise transition: first from seat-based to consumptive pricing, then, once the company has greater verifiability in its systems, toward an outcome-based license model. The condition Herzig attaches is worth noting. Outcome pricing, in his framing, requires confidence that the system actually did what it was supposed to do. The pricing shift and the reliability problem are the same problem.

Laura Burkhauser and Nathan Labenz arrive at the same destination from the product side. Burkhauser observes that the consensus direction for pricing is outcome-based charging tied to something like a successful export, rather than a per-use cost. Labenz explains why: reasoning models and agentic workloads make token consumption hard to predict and harder to reason about, so the industry is moving toward effective compute consumption as the relevant unit. Token-based pricing, he argues, breaks down precisely where AI is growing fastest.

Matt Barrie, who runs Freelancer.com, has already implemented the alternative. His platform charges per outcome rather than per token, positioning it against the AI model pricing convention rather than alongside it. Anton Osika adds a revealing demand-side signal: customers are increasingly willing to pay overage charges because the value they receive is high enough to justify it. That willingness matters. It suggests the market will bear outcome pricing not just as a vendor preference but as a buyer preference, once the value is legible.

Investor Gavin Baker puts a dollar figure on what the transition implies at the infrastructure layer. “The shift to usage based pricing is probably why you will see OpenAI and Anthropic exceed well over $200 billion in ARR this year.” That forecast is Baker’s own and should be read as such. But the mechanism he names is the same one every other speaker identifies: when pricing follows consumption rather than seats, revenue scales with use rather than with the size of a licensed roster.

The benchmark question follows naturally. Riley Brown predicts that within a year, AI model evaluations will stop measuring token efficiency and start measuring cost and time per task. That is the pricing logic applied to model selection: buyers will optimize for what they actually pay for, and if they are paying for outcomes, they will compare models on outcome cost. The evaluation layer tends to lag the commercial layer; Brown is suggesting the gap is closing fast.

What connects these positions is a structural observation that practitioners are already acting on: outcomes are paid out of work budgets, which are an order of magnitude larger than software budgets. Seat licensing competed for a share of a software line item. Outcome licensing competes for a share of labor spend. The timeline for full market repricing is debatable. The direction, given what builders and investors are already doing, is not. The companies that reprice earliest are not just changing a billing line. They are repositioning into a larger pool of money than the one per-seat SaaS ever competed for.