The AI success stories, and what they share

Three years ago, the open question was whether anyone would pay for generative AI. The answer now comes in numbers that would have looked like typing errors in 2022: a chatbot with 900 million weekly users by February, a lab that planned for tenfold growth and got far more, a coding tool sold for 60 billion dollars.

One caveat first. Most of these figures are run rates: the revenue of a recent month, multiplied to a year. They are snapshots, not audited annual revenue, and companies count them differently. With that in mind, here are the stories that stand out, and what they have in common.

Record growth

ChatGPT set the pace. In January 2023, two months after launch, it was estimated to have reached 100 million monthly users, Reuters reported from a UBS study based on Similarweb data. TikTok had needed about nine months. The UBS analysts wrote: “In 20 years following the internet space, we cannot recall a faster ramp in a consumer internet app.”

In February 2026, OpenAI said ChatGPT had 900 million weekly active users and 50 million paying subscribers. Revenue followed. OpenAI started 2026 at 20 billion dollars in annualised revenue, against 6 billion in 2024, and told investors this month that September came to almost 50 billion, according to Reuters.

Anthropic, which sells mostly to companies and developers, grew faster still. Its run rate went from 87 million dollars in January 2024 to 1 billion that December, 9 billion at the end of 2025 and 30 billion in April 2026, VentureBeat reported. Its chief executive, Dario Amodei, said: “We tried to plan very well for a world of 10x growth per year.” Then he added: “And yet we saw 80x.” By July the run rate had crossed 65 billion dollars, Reuters reported. The two labs count differently, though: Anthropic includes sales made through cloud partners such as AWS and Google Cloud, and OpenAI does not.

Tools for one job

The second wave came from tools that do one job very well. Cursor, a code editor in which AI writes alongside the developer, announced in November 2025 that it had crossed 1 billion dollars in annualised revenue, raised 2.3 billion at a valuation of 29.3 billion, and had a team of over 300. On 14 August 2026, SpaceX completed its acquisition of Cursor, a deal valued at 60 billion dollars, Bloomberg reported.

Lovable, from Stockholm, lets people build apps and websites by describing them. It crossed 400 million dollars in annual recurring revenue in February 2026, a month after reaching 300 million, and said it got there with 146 full time employees, TechCrunch reported.

Both sell to the person who does the work, and both spread the same way: someone builds something, shows it, and the next person signs up.

Small teams, large numbers

The most striking number in these stories is often the headcount. Base44, a tool for building apps by chat, was six months old and had never raised money when Wix bought it for 80 million dollars in cash in June 2025. Wix confirmed that it had eight employees, and TechCrunch reported that it made 189'000 dollars in profit in May, after paying for the models it runs on.

Bessemer Venture Partners studied ten of the fastest AI startups, which it calls Supernovas. They reached about 40 million dollars of annual recurring revenue in their first year of selling and about 125 million in the second, with about 1.13 million dollars of recurring revenue per full time employee, four to five times what Bessemer considers typical for software.

The same report is frank about the price of that speed. These companies average only about 25 percent gross margin, often close to zero or below, because every answer costs computing power, and fast adoption can also mean customers who leave as easily as they came. In Bessemer's words: “Topline ARR, of course, doesn't always guarantee a healthy business.”

The winner under the stack

The largest winner makes nothing most people use directly. Nvidia designs the chips most AI runs on. In the quarter to 26 July 2026 its revenue was 96.2 billion dollars, up 106 percent from a year earlier, and its data centre business alone brought in 89.0 billion.

Its founder and chief executive, Jensen Huang, described the change in a line: “Now, compute is revenue.” He also pointed to who is buying: “This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups.”

So far, the money of the AI years has gathered at two ends: at the bottom, where the chips are made, and at the top, with a few labs and the tools built on their models. Much of the middle is still being decided.

What the winners share

Read side by side, the product stories have four things in common. Each started with one job people already did every day and found tedious: writing code, building an app, putting a website together. Each delivered the result, not a feature, so users could see what they got in minutes. Most spread first through the people who use them, before any large sales effort. And the youngest stayed remarkably small for their revenue.

There is a fifth thing, and it is the risk. Most of them build on models made by someone else, pay for every answer, and report growth as run rates that can shrink as fast as they grew. The success stories are real. Many of them are also young.

For us, the most interesting number is not the 60 billion. It is Base44's eight people and Lovable's 146. They point to where we are heading with Helm, a vision we are shaping with our partner Eigna Labs: a whole company run by one founder with agents, every function covered, the founder deciding. It is in the making, not live. What the success stories teach us is focus: one job, done completely, by a small team that runs on its own tools.