On August 9, OpenAI switched off Atlas, its AI-native browser — 292 days after launch. It was the most decisive of four retreats: Arc was discontinued, Sidekick was absorbed by Perplexity and shut down, The Browser Company sold itself to Atlassian for $610 million. The "Chrome killer" wave of 2025 ended with Chrome's global share exactly where it started: two-thirds.
The most telling detail is the smallest one. The last instruction OpenAI gave Atlas users was how to export their bookmarks to an HTML file before August 9. A company that set out to reinvent the browser ended up teaching people how to migrate back to Chrome.
Four exits, four ways to die
The four AI browsers of the wave died four different deaths. Arc: discontinued in May 2025 when founder Josh Miller announced the team's pivot to Dia. Sidekick: team absorbed by Perplexity, product shuttered that August — its homepage now redirects to comet.perplexity.ai. The Browser Company: acquired by Atlassian for $610M in September, widely read as "a dignified exit." Atlas: simply switched off — Mac-only to the end, never shipped Windows, never shipped mobile.
The numbers behind the exits were never close. Chrome's global share stayed at roughly two-thirds all year, never moving more than two points. Every AI browser combined never crossed 1%. Comet, the biggest, was at roughly 10 million monthly actives — against the 800 million weekly actives of ChatGPT, the entry it was supposed to leverage. Eight hundred million people at the door, and a few million trickled through.
Why Chrome never had to fight
Chrome won a war it never fought, for three compounding reasons.
Switching cost versus plugin install. Changing browsers means migrating bookmarks, passwords, extensions, payment data, and relearning every interaction. Installing an AI plugin in Chrome takes two clicks. The challenger asked users to move house; the incumbent asked them to open a window. The entry-point business has always been distribution, not product — Google pays Apple roughly $20 billion a year just to stay the default search box, and that economics hasn't changed in thirty years.
Mobile absence. The challengers were desktop-first, and Atlas was Mac-only to the end. Browsing lives on phones. You can't siege a city you never visited.
Absorption speed. Gemini landed inside Chrome in September 2025; Copilot mode spread through Edge at the same time. The features startups spent six months polishing — sidebar Q&A, page summarization, agentic execution — became a single version update in the incumbent's product. The innovator's dilemma never kicked in, because its premise requires incumbents to be afraid of cannibalizing their own business. Browsers don't directly monetize, and AI features don't cannibalize anything, so Google and Microsoft followed without hesitation. The window that challengers normally exploit never existed. Chrome turned AI into a setting and the moats dissolved.
The survivors prove the point by how they live
The names still standing are not browser companies. Comet is Perplexity's acquisition funnel — it went free for everyone. Dia is an extension of Atlassian's office suite. Opera Neon charges $19.9 a month on top of Opera's two decades of installed base. Strawberry runs on VC money. None of them makes money from the browser itself: the browser is an acquisition channel, a feature carrier, or a piece of someone else's map — never a standalone business.
OpenAI's shutdown statement doubles as the category's epitaph: "AI browser should be a feature, not a destination." The company that spent the most on this direction — built the team, shipped the product, ran 292 days under "Chrome killer" expectations — admitted the direction itself was wrong. This wasn't a product failure. It was a route failure.
China had already previewed this ending a year earlier: nobody built a new browser from scratch. Kuaike grew into an AI workbench inside Alibaba, QQ Browser attached Hunyuan and drifted toward an office assistant, 360 stuffed AI into its security browser. Everywhere, the pattern was the same: existing products plus AI, zero greenfield browsers.
The framework: the browser is a distribution layer, not a product layer
Strip the story down and the structure is clear. The entry-point business has always been a distribution auction — default search slots are worth tens of billions a year — and AI didn't change those economics. It changed where intelligence attaches. When the intelligence layer can attach to any surface, the shell stops mattering: AI as a feature gets absorbed into the incumbent's product in one release cycle, indistinguishable from a settings toggle. The category didn't lose to Chrome. It lost to "AI is now a feature of Chrome."
What the entrance war actually did
OpenAI didn't abandon the entrance war — it left the browser shell. Atlas's capabilities were split in two: part went into the ChatGPT desktop app, including the cloud browser that operates web pages for you; part became a Chrome extension. The next battle isn't about better tabs; it's about who completes the task for the user. The opponent is still Google. The weapon is now the agent.
Two lessons fall out of the wreckage. For startups: any AI product that is a feature in search of a destination will be absorbed, because absorption is free for the incumbent and fatal for you. For users: the one-sentence summary of this entire year is — don't move for AI. The plugin comes to you.
What to do this week
- If you're building: decide early whether you're a feature or a destination. A feature should attach to the largest distribution surface available. A destination needs a monetization moat that feature-absorbers can't copy — a subscription to a workflow, not a UI.
- If you're buying: don't switch tools for AI features. Evaluate the AI additions in the tools you already use; the capability gap is closing in release cycles, not quarters.
- Watch the agent layer, not the browser: the entrance war has relocated to agentic execution — desktop agents, cloud browsers, extensions that act. That's where the next "Chrome killer" story will actually play out.