HubSpot Is Quitting a Business It Invented, and the Stock Knows It

On a September 9 conference stage, HubSpot CEO Yamini Rangan did something a little unusual for a company worth $11.5 billion. She described the next version of HubSpot as not a better version of the last one. The company, she said, is moving from selling software features to selling "work" — an agent that resolves your support ticket, another that books the sales meeting, priced by the outcome instead of the seat. The stock, which has fallen about 43% this year, had already priced this in as a downgrade, not an upgrade.

That gap is the whole story, and it isn't an AI story or a valuation story. It's a category question: is this the same company with a new feature, or an early version of a different kind of company? Most investors have been sorting it as the former — a steady compounder whose growth is decelerating — while management is building the latter. One of them is wrong, and the difference matters a lot.

Here's the concrete move, because the abstraction is easy to wave at. HubSpot has sold software for its entire life the way everyone in its category does: you buy a seat, a human sits down, learns the menu, and uses it. You renew the seat every year. That model is beautiful precisely because it never breaks — the customer is doing the work. Now the company is doing something that, on its face, un-invents the business. It extended its free trials for the new AI agents from a week or so to 14 to 30 days, so a customer could drop the tool into their own data and actually watch it work before paying. Longer trials mean deals that slip, a "slow start" to the quarter, and a little air pocket in new-customer growth. She said this plainly: the transition creates "near-term trade-offs." A company that is confident in its future can afford to make its present look worse.

The second move is the one that rewrites the economics. HubSpot is shifting to what it calls a hybrid seat-plus-credits model, where the credits are tied to outcomes — a resolved ticket, not a clicked button. Customers can set a spend ceiling by use case. Think about what that does. Under the seat model, a customer who automates away their work still pays for the seat. Under outcome pricing, a customer who succeeds at the thing you're selling has a built-in motive to cap it. The company is, in effect, agreeing to get paid less per unit of value delivered, in exchange for owning the value-delivery layer itself. That's a real bet, and it's the opposite of the instinct you'd expect from a company that just hit ~300,000 customers.

Now the part that makes the future feel real, because it isn't just a deck. HubSpot says agentic reach across its base climbed from 37% at the start of 2025 to over 55% by the conference, that first-party agent adoption roughly doubled from 9% to about 18% in six or seven months, and that actual agentic actions grew more than 300% in the same stretch. Those are usage numbers, not revenue numbers. That distinction is the entire investment case. The usage is compounding. The money is not. One analyst summary put it cleanly: AI is "future tense," contributing to margins through cost takeout, not to the top line. The company cut inference costs to about 80% of what they were by switching some tasks to cheaper open-weight models — so the AI is helping the profit margin today while the revenue engine is still rewiring.

This is where the two framings stop agreeing. The stock fell from a high near $525 to the low $200s. It is still expensive, and it's worth being precise about why. GAAP earnings are thin — stock-based compensation keeps them low — so on that basis the stock runs around 78 times trailing earnings. On the non-GAAP basis software investors usually use, trailing earnings are about $11.70 and the multiple is closer to 20 times, with revenue growing about 18% a year. Either way, you are paying a premium for a company whose growth just stepped down. The market's verdict, in so many words, is: this is no longer a re-accelerator, it's a compounder, and you can't pay growth-stock prices for compounder numbers. The company's own guidance backs the deceleration: full-year 2026 revenue is guided up about 18%, a step down from the early-2026 pace. What it is not doing is guiding down the margin — non-GAAP operating margin is expanding, and management hit a 2027 margin target a year early. So the machine is getting more profitable while growing more slowly. That's a fine business. It is not, at 78x, a business that needs the AI story to be true to justify the price. It needs the AI story to be true to justify the future.

The honest version of the risk is a single, sharp one, and management surfaced it themselves. Their Customer Agent resolves about 70% of tickets on its own. Outcome-based pricing only works if that number is believable; if resolution quality slips, the customer's spend ceiling does exactly what it was designed to do and the whole thesis quietly deflates. This is the classic trap of selling a new kind of value: you can't explain it away with a marketing deck, because the number is the product.

So the test isn't a target price. It's two numbers the company has already promised to show you. One: management's own bar is that net-new ARR — new recurring revenue added each period — re-accelerates and exceeds constant-currency revenue growth by the end of the year. If the agents are pulling in real customers, that number moves before anything else does. Two: resolution rates on the Customer Agent. If that 70% holds or climbs while agentic actions keep growing, the outcome-pricing model has found its footing. If the usage curve is still steep but net-new ARR stays flat or resolution slips, then the transition is a story the usage is telling and the revenue is quietly disagreeing with.

The deeper question, and the one worth sitting with, is whether a company that sells work can keep selling it. Every seat-vendor in software has been told to "add AI." Almost none of them have rewritten the price tag to be paid by the outcome, because that means betting the renewal model on a number they don't fully control. HubSpot is doing that, and the market is paying it with a de-rating to find out whether the bet is a category shift or just a very expensive rebrand. The stock doesn't know which one this is, and neither do you yet — but it does tell you what you're being asked to buy: at 78x earnings, you're paying now, as if the future version were already the better company.