Try to remember the last time you chose a browser.Not opened one. Chose one. Sat there with an actual decision in front of you, weighed the options, and picked. For most people, if the memory exists at all, it is somewhere around 2012, on a machine that has since been thrown away, and the decision took about four seconds because someone said Chrome was faster and it was.Since then you have replaced the laptop twice, changed jobs, changed phones, changed operating systems, and Chrome came with you every time. Not because you re-chose it. Because it was already there, signed in, with your bookmarks and your saved cards and that folder of tabs from 2021 you cannot bring yourself to close. Four billion people are living inside a decision they made once and have never revisited.That is the whole reason nobody can kill Chrome. Almost nobody attacking it has been willing to say so. OpenAI just found out the hard way, killing ChatGPT Atlas nine and a half months after launching it. But Atlas is only the newest name on a long list, and the list keeps getting longer because everyone on it keeps attacking the browser, which is a different object entirely.
The better mousetrap doesn’t work
Atlas, for what it’s worth, was a good browser. Cleaner than Chrome. OpenAI built a whole architecture for it, OWL, which pulled Chromium’s browser process out of the main app so tabs would open instantly and an engine crash wouldn’t take the window down with it, then rebuilt the interface in SwiftUI rather than reskinning Chromium’s, then published a long and pleased engineering post about the work. None of that saved it. It shipped on macOS in October 2025, never made it to Windows or iOS or Android despite promising all three, and dies on August 9.Arc is the cleaner proof, because Arc was better still. The Browser Company built the most inventive browser in fifteen years: tabs in a sidebar and treated as disposable, Spaces to keep work from bleeding into everything else, an interface that looked like somebody had gone back and asked what a browser window is actually for. People didn’t merely use Arc, they evangelised it. Users quadrupled in a year. Then Josh Miller said the sentence founders don’t say, that Arc was too much novelty and change to ever reach the numbers the company needed, and built Dia instead, an AI browser that looks almost apologetically like Chrome. Atlassian bought the lot for $610 million.Opera has been at this longest and has the least to show. It invented the omnibox that Chrome later took and made standard. It has since shipped Neon at $19.90 a month and Air with breathing exercises built in, and it holds about two percent, roughly what it held before any of this began. Brave sells privacy and pays you in crypto to watch ads. It does fine. It is not close.The pattern is not subtle. Better does not move people, because a browser is infrastructure, and nobody replaces infrastructure because a nicer one exists. They replace it when the one they have stops working. Chrome, for four billion people, has not stopped working.
Neither does owning the machine
The obvious rebuttal is that those are small companies without distribution, and distribution is the whole game. So look at the one that has it.Microsoft owns Windows. It ships Edge preinstalled and set as default, which means that on a new PC the only way to get Chrome is to open Edge and type “Chrome” into it. It has defended that checkpoint for a decade with pop-ups that look enough like malware for users to file support tickets, Windows updates that reboot the machine and open Edge with your Chrome tabs already imported, and a Bing page for the query “Google” dressed up to resemble Google, doodle and all. Microsoft also got to AI first, which everyone has forgotten. The Bing sidebar landed in Edge in early 2023, when ChatGPT was four months old and Google was still deciding whether to panic.Edge sits in low double digits.That number is more damning than anything in the Atlas obituary. Microsoft owns the operating system, ships the browser, controls the search box, got to AI a year before Google, and has been willing to behave like adware throughout. It still cannot move the market. Not because Edge is bad, because Edge has been good for years, but because owning the machine is not the same as owning the moment of choice, and the moment of choice is long over.There is a small comedy in there that nobody says out loud. Edge’s Copilot ran on OpenAI’s models. So when OpenAI built Atlas, the browser it was actually cannibalising was Edge, the one that had been carrying its technology to Windows users since 2023. Atlas was sold as a Chrome killer. In practice it was aimed at the browser that had already lost to Chrome while running OpenAI’s own engine.Perplexity is the last data point, and the most honest. Aravind Srinivas has been the most articulate person in this fight about why a browser matters: it is a containerised operating system, the only place an agent can reach your logged-in accounts and take an action rather than answer a question. Answering questions is a commodity, he said. Actions are the next advantage. He was right about all of it. Then Comet launched behind a $200-a-month plan, invite only, and by August Perplexity was offering $34.5 billion for Chrome. A company valued at $18 billion, bidding twice its own worth for the browser it was theoretically out-competing.That bid was not a strategy. It was the man who understood the prize best admitting the only way to own a browser is to buy the one that already has the users.
The idea wasn’t stupid, which is what makes this painful
Which is where I should stop counting bodies, because the case for AI browsers was a good one, and pretending otherwise would be the lazy version of this argument.The pitch was that this time is different in kind, not degree. Not a nicer browser but a different animal, where an agent reads across your tabs, books the table, fills the cart, and does the tedious middle of the web while you supervise. That is genuinely not a feature the way sidebar tabs are a feature. It is a change in what the software is for, and a change in what software is for is the only thing that has ever unfrozen a market this frozen. It happened once already, when the web went from documents to applications, and Chrome is the thing that won that shift. If you believed a comparable shift was underway, building a browser wasn’t vanity. Not building one would have been negligence.The problem isn’t that the idea was wrong. It’s that a genuine new category and a very good feature look identical right up until the incumbent ships it.In January, Chrome shipped it. A persistent Gemini sidebar. Tab-group awareness, which was Atlas’s marquee trick, lifted more or less intact. Personal intelligence reaching into Gmail and Photos. Auto-browse, which will go and buy the thing for you and hunt the discount code. Every headline capability Atlas launched with, arriving inside the browser four billion people already had open, requiring nobody to install or migrate or learn anything.Ninety days later, Atlas was cancelled.Srinivas saw this coming and said so a year out: if your feature can throw off serious revenue, assume a model company will copy it. Then he built Comet anyway, because the alternative was standing in front of investors and saying the word “feature.”
You cannot burn down the house you are standing in
Here is the fact that makes the category unwinnable, and it goes largely unsaid because saying it embarrasses everyone in the room.Atlas was Chromium. Comet is Chromium. Dia is Chromium. Neon is Chromium. Brave is Chromium. Edge is Chromium, ever since Microsoft gave up on its own rendering engine in 2020 and rebuilt on Google’s.Almost every browser built to escape Google’s browser is built on Google’s browser, because the alternative is reconstructing rendering, memory management, sandboxing, encryption, video playback and a security-patching operation that never ends. Ladybird is attempting it from scratch and has been in alpha for years with no meaningful ship date. Srinivas has conceded that Comet wouldn’t exist without Chromium. OpenAI’s proud engineering post about OWL is, read plainly, a long description of how carefully it wrapped Google’s engine.Consider what that concedes before the challenger has written a line of its own code. It cannot compete on rendering. Or speed. Or web compatibility. Or security. All of those belong to Google, and all of those are, inconveniently, what a browser fundamentally is.What is left is the layer bolted on top. The sidebar, the agent, the box in the corner. And that layer is precisely what Google can reproduce in a single release cycle.To fight Chrome you must first adopt Chrome. Having adopted it, the only ground left is ground the incumbent can retake whenever it feels like it.
None of it was built for you
There is a second reason underneath the first, and it explains why users didn’t move even when the products were good.Listen to what these people say when they are talking to each other rather than to customers. The Browser Company’s CTO explained that the browser owns Cmd-T and the omnibox, the most-used text box on your computer, where a person expresses intent before anything else happens. An OpenAI product lead called the browser the operating system for your life. Srinivas said the point of a Perplexity browser was context from outside Perplexity’s own app, a fuller profile of you, with ads to follow through the Discover feed.Line those three up and the strategy resolves. The browser was never the product. It was the way in. What was wanted was the text box, the cookie jar behind it, and standing permission to act inside every session you are already logged into.Each of those is an excellent reason to build a browser. Not one of them is a reason to use one.Users never read a word of it and worked it out anyway, in the only way that shows up on a balance sheet. They were asked to abandon a browser that works, adopt one that leaks credentials in security tests, and accept an agent that books the wrong restaurant slowly. They looked at the offer, found nothing addressed to them, and stayed put. In November 2025, with Atlas and Comet and Dia and Neon all live and all funded and all being written about as the end of Google’s reign, Chrome set a market share record.
So try to remember
Here is the uncomfortable thing about a decision nobody remembers making. It cannot be argued with, because there is no argument to answer. Nobody sat down and reasoned their way to Chrome, so nobody can be reasoned out of it. There is no case to rebut, no belief to change, no rival claim to defeat. There is only inertia, and inertia is not persuaded. It is interrupted.And it does get interrupted. That is the part the challengers were right about. Netscape went down when the web arrived on machines that shipped with Internet Explorer. Explorer went down when the web became applications and its old engine simply could not run them. In both cases the browser people had did not merely fall behind, it stopped working, and stopped working in a way they could feel. That is the only force that has ever moved this market: not a better browser, but a broken one.Which is what nobody attacking Chrome has been able to manufacture. Arc made a nicer browser. Atlas made a smarter one. Comet made a more ambitious one. Edge made a browser you have to physically walk past to install anything else. Not one of them made Chrome stop working, because Chrome does not stop working, and every capability they arrived with was a capability Google could ship into Chrome before the switching cost was ever paid. Auto-browse landed in January. Atlas died in July.So the browsers survive as niches, which is not nothing. Brave keeps its privacy crowd. Dia keeps the knowledge workers Atlassian paid $610 million to reach. Comet keeps whoever is paying $200 a month. Real products, real audiences. Just not the audience anyone was actually after, and none of it adds up to a reason for the person who has never once thought about their browser to start.The wall in front of them is not Google’s engineering. It is not even Google’s money. It is you, and the last time you gave the browser any thought at all, and the fact that it was over a decade ago and took four seconds and you cannot bring the moment to mind.Go on. Try to remember choosing.You cannot beat a decision nobody remembers making, because you cannot win an argument nobody is having.
