The multiple was a bet on the option, not the revenue
Airtable sold to Bending Spoons at a $1.285 billion enterprise value in August 2026 on about $480 million of ARR, roughly 2.7 times revenue. In 2021 it raised at about $11.7 billion, something like 40x forward revenue. The business did not fall apart between those two numbers. It still grows over 20% and serves about 80% of the Fortune 100. What changed is that a revenue multiple is a bet on what happens next, and the thing investors were betting on in 2021, that Airtable becomes the platform where a generation of business software gets built, stopped being a live option.
The mechanism is specific. Airtable's value to a customer was proportional to the effort that customer had already put into modeling their workflow into bases, views, automations, and interfaces. That effort was the switching cost, and the switching cost was the moat. Agents let a competent person describe the same workflow in a paragraph and get a working app, so a moat denominated in effort deflates at the rate effort gets cheaper. The enterprise plumbing, permissions, audit, governance, and integrations, is still hard, which is why the ARR is sticky. But sticky is a cash-flow property. Buyers pay 2.7x for sticky and 40x for the option, and Bending Spoons, a firm that buys cash flows and runs them lean, being the top bidder is the market saying the present value of the business now exceeds the present value of its story.
What I do with this on my own product
I run this as a concrete exercise, not a mood. Take your lock-in and name the unit it is denominated in. If it is hours of user labor, assume that unit deflates around 30% a year and reprice your strategy accordingly. If it is data gravity, regulatory approval, or being the system of record four other systems read from, it is more durable.
On Heidi, the agentic platform I build, that pushed my funding off the builder surface, which is the generatable half anyone can ship in six months, and onto three things a model does not have: memory that accumulates an organization's real decisions over years, trust infrastructure for long-running agents (permissions, reversibility, guard rails, and receipts for what an agent did and why), and being the system other AI tools read from. And it pushed me off per-seat pricing, because if work that took a person three weeks now takes an agent an afternoon, charging per seat is charging for the wrong thing and puts you on the wrong side of your own product's success.