Is AI Killing SaaS?

On November 12, 2026, HubSpot brought back its co-founder, Brian Halligan, as chief executive, five years after he handed over the job. His statement ended: "This is a refounding, and I can’t wait to start!" The easy conclusion is that AI is killing software, and HubSpot has sent for its founder.

Halligan co-founded HubSpot in 2005 and ran it for 16 years. He takes over from Yamini Rangan, who led the company from September 2021 and, by HubSpot's count, took it from 135,000 customers to more than 311,000, and from $1.3 billion of revenue in 2021 to an expected $3.7 billion this year.

The numbers tell a stranger story. HubSpot's revenue grew 17% last quarter, faster than all but two of 14 software companies that sell mostly by the seat. Since January 2025 it has trailed the US market by 78%, worse than any of them. Both are true, and the gap between them is the story. AI is not killing software's revenue. It is breaking the unit software is sold by: the seat.

Eight days earlier, HubSpot had reported revenue of $947.6 million for the third quarter, up 17% on a year before. It guided the fourth quarter to between $945.0 million and $947.0 million. Even the top of that range is below the quarter it had just reported. In each of the 54 quarters in HubSpot's filings since 2013, revenue had grown on the quarter before.

The filings tell a more useful story. Most business software is sold by the seat: one login, one person, one fee. HubSpot is now repricing its own product around credits and outcomes, and its own quarterly report warns that the new spending "may substitute for, rather than add to, existing seat and hub spend." Most people think AI is killing software's revenue. In fact the revenue is still there; what the market has stopped trusting is the seat. That makes HubSpot's refounding a pricing and positioning problem before it is a growth problem, and the same is true for every company that sells software one person at a time.

The word arrived on February 3, 2026, when Bloomberg quoted a trader on Jefferies' equity desk: "We call it the ‘SaaSpocalypse,’ an apocalypse for software-as-a-service stocks." The trigger that week, Reuters reported, was an AI lab's launch of plug-ins that let its agent automate work in legal, sales, marketing and data analysis. By TechCrunch's count, "an investor sell-off wiped nearly $1 trillion in market value from software and services stocks" in early February.

The logic is simple. As TechCrunch put it, "SaaS companies currently price their software per seat — meaning by how many employees log in to use it." If an agent does the work of three people, the customer needs fewer logins. The chief executive of Rocket Software put it in one line to Fortune: "The SaaS companies with user‑based pricing have taken a hit, because if AI improves productivity, the number of users goes down." Fewer users, fewer seats, less revenue. Some customers will build their own tools; some will buy from companies born around agents.

Not everyone bought it. Nvidia's Jensen Huang called the idea that AI would replace software "illogical" and said "time will prove itself." So we asked three questions: is the revenue shrinking, what is the market actually pricing, and where would the damage show first?

Not yet. Take 14 US-listed application software companies that sell mostly by the seat or the account. Thirteen of the 14 grew revenue in their latest quarter against the same quarter a year before. Dropbox barely grew, by 0.5%, and only ZoomInfo shrank, by 2.9%. HubSpot, at 17.1%, was the third fastest, behind Atlassian and ServiceNow.

The spending that existing customers do tells the same story. HubSpot's net revenue retention, which measures whether last year's customers spend more this year, was 103.5% in 2025, up from 101.8% in 2024. Freshworks' rate was 108% at the end of 2025, up from 103% a year earlier. Figma, which sells both seats and AI credits, said its retention "remained strong at 136% as customers expanded both seats and AI credit add-ons." Microsoft reported that "Microsoft 365 Copilot reached over 30 million paid seats": an AI product, sold by the seat.

The fear is in the filings, though, written in the conditional. ZoomInfo's annual report warns that "This AI-driven seat compression could result in customers reducing seat counts at renewal, negatively impacting our revenue even as they derive equal or greater value from our products and services." The risk it names is not that customers get less value. It is that they get the same value from fewer seats, and pay accordingly. That is a pricing problem, and it has not yet arrived in the revenue line of most of these companies.

If revenue were the story, the market would have punished the slow growers. It did the opposite. Measured against the US market from January 2, 2025 to November 11, 2026, HubSpot did worse than any of the 14: a holding in it ended worth 78% less than the same money in the US market. Box, which grew its revenue at about half HubSpot's pace, came out 5% ahead.

Split the group in two by growth and the pattern holds. The seven fastest growers fell a median 53% against the market; the seven slowest, 33%. This is a correlation, not a verdict, and the market does not publish its reasons. But it fits a simple mechanism. A share price is a bet on the cash a company will produce for decades. A fast grower's value sits mostly in the future, in contracts it has not signed yet. A slow grower is priced on what it already earns. If you doubt that a seat will still be the unit customers pay for in 2030, the company whose value depends most on 2030 is the one you mark down hardest.

So the market is not saying these companies have stopped selling. It is saying it no longer knows what they will be selling, or at what price. That is a different problem, and it has a different cure.

Inside the companies, it shows first in price, not in volume. HubSpot is the clearest case. Under Rangan its customer count more than doubled, from 135,000 to 311,704 at the end of September. Over 2021 to 2025, its average subscription revenue per customer moved from $10,486 to $11,414: about 9% in four years. HubSpot grew by adding customers, not by charging each one more, and its own quarterly report says where the new customers came from: "The growth in Customers was primarily driven by increased demand for our lower-priced Starter products." The same report credits the rise in revenue per customer to "continued adoption of our Professional and Enterprise products, partially offset by continued purchases of our lower-priced Starter products." Two engines, then: more small customers, and some customers moving up.

Now look at the third quarter as arithmetic. Twelve percent more customers, each paying 4% more, is about 16% more subscription revenue, close to the 17% HubSpot reported. Most of the growth came from the first engine. The fourth-quarter guide, up 12%, leaves little room for the second: on the call, according to the published transcript, HubSpot said it expects revenue per customer to grow at a "low to mid-single-digit" rate in constant currency.

Why is the second engine slowing? HubSpot's quarterly report names the headwinds itself: "changes in customer buying behavior for AI solutions and the recent changes we made to our product, pricing and go-to-market approach." On the call it also expected upgrades to stay "pressured as a result of customer budget optimization." Part of the slowdown, in other words, comes from HubSpot's own repricing. At its Analyst Day on September 17, 2026, according to the published transcript, it described "A hybrid model of seats and credits. Seats provides access to platform data and AI capabilities. Credits, they scale with the value and the outcomes that we deliver." On the third-quarter call it reported a pilot of the new pricing in the Nordics and Benelux: "Compared to regions outside the pilot, credit usage is 2 to 3x higher. Pro+ customer volume is higher, while average selling price is lower." Usage up, price down. On the same call: "We now expect net new ARR to decline year-over-year in fiscal 2026," and net revenue retention "to be down about 2 points."

The change in HubSpot's own risk factors is the clearest marker of all. Its annual report for 2025, filed in February, said: "Our consumption-based pricing strategies are novel and evolving." Its quarterly report filed on November 4 says: "Our credits-based and outcome-based pricing strategies are novel and evolving. These changes have, and may continue to, lengthen sales cycles, result in lower initial contract values, reduce or delay revenue, and may not be accepted by customers, and credit spend may substitute for, rather than add to, existing seat and hub spend." In nine months the company went from experimenting with a meter to warning that the meter may replace the seat.

Seats are not disappearing at HubSpot, either. On the call the company said "Active core seats grew 34% year-over-year." The core seat dates from 2024, when, as Rangan told analysts, "We lowered the seat price, we removed the seat minimums that we had for Pro as well as Enterprise, and we introduced a new seat called the Core Seat." Seats are growing about twice as fast as revenue. Customers are using more of the product and paying less for each unit of it. That is what seat compression looks like when it arrives: not fewer users, but a lower price for each one, and so far the vendor is the one setting it.

HubSpot is not alone, and it was not first. Zendesk introduced outcome-based pricing for its AI agents in August 2024, billing for automated resolutions. Salesforce introduced credits for its Agentforce agents in May 2025, at $0.10 an action, after launching them at a flat $2 a conversation. Intercom prices its agent at $0.99 per resolution. In April 2026, according to a published transcript of the call, ServiceNow's chief executive said half of its net new business came from pricing not tied to seats, including tokens and connectors. ZoomInfo added a hybrid of platform fees, credits and seats in the third quarter of 2026.

The vocabulary moved even faster than the prices. In 2024, five quarterly filings by prepackaged-software companies used the word "agentic." In 2025 the count was 60. By November 12, 2026, it was 155. Mentions of "seat-based" went from none in 2024 to 12 this year. A mention is not an adoption, and many of these are risk factors about rivals. But a word that goes from five filings to 155 in two years tells you what the people writing the risk factors are worried about.

A repricing costs money before it makes any. HubSpot has seen this once already. Of the 2024 seat changes, it said on the third-quarter call: "While this initially lowered ASP, it has led to healthier customer cohorts with higher retention." The pilot shows the same shape: usage rises first, average selling price falls first, and revenue catches up only if the meter runs long enough. Meanwhile the company is paying for the change in two currencies.

The first is shares. HubSpot's stock-based compensation was $528.2 million in 2025, 16.9% of revenue. In every year from 2013 to 2024 HubSpot reported an operating loss under the accounting rules; 2025 brought its first operating profit since at least 2013, $7.4 million. At Analyst Day the company set a 2030 target of stock-based compensation at 10% of revenue, with operating margins of 30% on its adjusted basis and 20% under the accounting rules.

The second is cash. In the first nine months of 2026 HubSpot bought back $973.7 million of its own stock. At the end of September it held $1.3 billion in cash and investments. Its weighted share count fell to 49.5 million from 52.5 million a year earlier. Buybacks are a defensible use of cash when the business is stable and the shares look cheap to management. They are a harder case in the year the business is rewriting its price list and expects the new price to come in lower first. Karen Burgess, finance chief of Adobe's Creativity & Productivity unit, joins on November 30 and becomes HubSpot's chief financial officer on February 11, 2027. In the release naming her, she promised "a disciplined approach to capital allocation, support smart investments to drive our future growth." HubSpot's release notes that "Since joining Adobe in 2002, she supported the company’s business model transformation." HubSpot is starting one of its own, and her first test is the order in which its cash is spent.

In February, Workday's co-founder Aneel Bhusri also came back as chief executive, and, in TechCrunch's report, "Bhusri said in a statement that the company's next chapter would be focused on AI."

The record on returns is sobering. A study published in MIT Sloan Management Review in 2020 followed 167 "boomerang" chief executives at US public companies from 1992 to 2017 and compared them with more than 6,000 others. "On average, the annual stock performance of companies led by boomerang CEOs was 10.1% lower than their first-stint counterparts." Founders were overrepresented: "While founders made up only 4% of our overall CEO sample, they accounted for 44% of the boomerang CEOs." The authors' advice was blunt: "Be especially wary of bringing back a founder." Their explanation matters more than their verdict: "many boomerang CEOs barely recognize the company upon returning, because business conditions differ dramatically from those of their first stint as CEO." The authors also name exceptions, such as Ron Shaich's return at Panera Bread. And a study like this cannot separate the leader from the trouble that brought the leader back. It is a base rate, not a forecast.

Founders do bring something specific. In a study of founder-led companies in the Journal of Financial and Quantitative Analysis, Rüdiger Fahlenbrach found that "Founder-CEO firms invest more in research and development, have higher capital expenditures, and make more focused mergers and acquisitions." Focus and investment are what a repricing needs. The risk is the one the Sloan authors name: leading the company you remember rather than the one in front of you.

A return can start with a diagnosis written in plain words. On February 14, 2007, before he returned as Starbucks' chief executive, Howard Schultz sent Starbucks' executives an email with the subject line "The Commoditization of the Starbucks Experience." He named the company's own choices as the cause: "we solved a major problem in terms of speed of service and efficiency," he wrote. "At the same time, we overlooked the fact that we would remove much of the romance and theatre." Halligan's statement has the start of one. "Chapter three is building the customer relationship harness that will enable millions to grow better." A harness is a good word for it: the thing every agent has to be clipped into to do its work.

Put the tests together and the picture is sharper than either the panic or the dismissal. First, software revenue is not dying. Thirteen of 14 seat-heavy companies are still growing, retention is holding, and seats are still being sold, including for AI. Second, the market has stopped paying for the future of the seat, and it has charged the most to the companies whose value depended most on that future. Third, the damage arrives as price before volume. Customers keep using the product and pay less per unit of what they get, and so far it is the companies themselves that are cutting the price.

HubSpot is the cleanest example. Rangan described the company's purpose as "the opportunity to help millions of small and medium businesses grow." Its customer growth has come mainly from lower-priced Starter products, and it grew by adding customers rather than charging each one more. HubSpot's refounding is not a turnaround of a failing business. It is a decision about what the business sells, and to whom.

The seat is one example of a broader habit: pricing the input instead of the result. Law firms bill hours. Outsourcers bill heads. Agencies bill retainers sized to the team on the account. Each of those units made sense when the work took people, and each is exposed in the same way when an agent can do part of the work. The customer does not want fewer hours or fewer heads. It wants the same result for less, and the unit that counted the people stops measuring what the customer values.

The companies that move first get to choose the new unit. Zendesk and Intercom chose the resolution; Salesforce chose the action; HubSpot is choosing credits on top of seats. The risk in waiting is a unit chosen for you, by a customer asking why the bill did not fall when the headcount did.

There is a trap on the other side too. Pricing by outcome sounds like alignment, but an outcome has to be counted, and someone has to agree on what counts. A resolution that the customer reopens a day later, a lead that never buys, a meeting booked with the wrong person: each becomes a dispute over the invoice. The cleanest new units are the ones both sides can see in the same system. That is why the system of record matters more in this world, not less.

Pick the unit before the price. Seats, credits and outcomes are three different promises. A seat sells access, a credit sells work, an outcome sells a result. HubSpot should be explicit about which promise each product makes and price it that way: the seat as the floor that buys the data and the controls, credits as the meter for work the agents do, and outcome pricing only where HubSpot can see and prove the result. Selling outcomes it cannot measure would turn every renewal into a dispute.

Make the CRM the place every agent has to pass through, including agents HubSpot did not build. Halligan's word, harness, points the right way. The durable asset is not any one agent. It is the customer record, the permissions and the history that any agent needs to act safely on a customer's behalf. Charge for access to that record and for the actions taken through it, and HubSpot earns from the rivals' agents as well as its own.

Choose the customer. Growth from low-priced Starter plans brings in many small accounts at a low price. Moving upmarket, which the company says it has been doing since 2021, raises the price but puts HubSpot against Salesforce, which now sells its agents by the action, and ServiceNow, which says half its net new business is priced without seats. A refounding is the moment to say which customer the product is built around, and to accept slower growth elsewhere.

Fund the transition before the buyback. The pilot shows the price falls before usage revenue catches up. Nearly $974 million of buybacks in nine months, against $1.3 billion of cash, leaves less room for a dip that the company itself expects to last into 2027. A pause, with the reasons given, would tell customers and employees which bet the company is making.

Price in the customer's language. A small business does not budget in credits. It budgets in leads, deals and support tickets. Translate every credit into the result it usually buys, show the customer the conversion on the invoice, and cap the bill where the customer asks for a cap. A meter the customer cannot predict will be switched off, however good the work behind it.

Publish the new scorecard. Investors are pricing what they cannot see. HubSpot should report the measures that show whether the change is working: revenue per customer split between seats and credits, credit consumption, retention, and gross margin after the cost of running the models. A company that shows its meter is easier to value than one that asks to be trusted.

Refound, do not restart. HubSpot's strengths are real: the inbound marketing category Halligan says it created; a CRM that, in Halligan's words, "hundreds of thousands use every day"; and more than 311,000 customers. The Sloan finding is a warning about leading the company you remember. The strongest version of chapter three keeps chapter two's customers and changes what they pay for.

If you run a business that sells software by the seat, find out what share of your customers' value is tied to a person logging in, and what share comes from work your product does on its own. The second share is the one an agent can reprice. Decide what you will charge for it before a rival does.

If you buy software, renegotiate as if the seat is temporary. Ask every vendor for a credit or outcome option, tie seat counts to measured use, and avoid multi-year commitments to seat volumes your own agents may make unnecessary.

If you sit on a board weighing a founder's return, ask for the diagnosis first: a plain statement of what has changed since the founder last ran the company, and a short list of what will change now. Ask the returning leader to name the unit the company will sell in three years, and how it will be priced.

We read HubSpot's filings: the 8-K of November 12, 2026, its third-quarter results and quarterly report, its annual reports for 2021 and 2025, and the October 6, 2026 filing on its workforce reduction. Quarterly revenue comes from SEC XBRL data; fourth quarters are the annual figure less the first three quarters. Statements from HubSpot's third-quarter call and Analyst Day, and ServiceNow's first-quarter call, are from published transcripts, not company documents, and are identified as such.

The peer group is 14 US-listed application software companies that file annual reports with the SEC and sell mostly by the seat or the account. Growth is each company's latest reported quarter against the same quarter a year before, as reported. Market performance is total return relative to SPY, a fund that tracks large US companies, from Tiingo adjusted closes; we show ratios, not prices. Word counts are the number of 10-Q filings by prepackaged-software companies that SEC EDGAR full-text search returns for each term. The study of returning chief executives is Bingham, Hendricks, Howell and Kolev in MIT Sloan Management Review; the study of founder-led firms is Fahlenbrach in the Journal of Financial and Quantitative Analysis.

What we could not see: HubSpot does not report revenue from credits separately, does not split revenue between small and large customers, and has not published Halligan's plans beyond his statement. The market's reasons are inferred from prices, not stated by anyone.

When revenue and value disagree, the market is not grading this year's sales. It is grading the unit you sell. Fix the unit, and the price of everything else follows.