Why Is Everyone Building AI Browsers

About 2 months ago I peeped into the future, saw what was coming, and decided to build an AI browser for kids. Last week OpenAI released their own AI browser called Atlas for everyone. If I had pursued that idea, I might have been in trouble or not. We’d never know.

But why AI browsers now?

• Data
• Distribution FOMO

The data angle is relatively straightforward; if you have enough data on a group of users, you can better communicate to, build for, and sell to them. But distribution FOMO is a different beast, and I will explain.

You see, for the longest time, Google has lived inside other people’s browsers; in fact, it even paid huge tickets to remain in these browsers as the default search engine. Without this, Google would never be where it is today.

The Google Playbook

Starting in the year 2000: Chrome became the most popular browser.

But when it first came out, Google Search was nothing more than a website inside of browsers made by Microsoft, Mozilla, Apple, etc.

Initially, Google offered the Google Toolbar, which was essentially a precursor to Chrome; it integrated Google Search into Internet Explorer and Firefox. Although toolbars are ineffective, Sundar Pichai’s initial role at Google was to develop one. Your kiosk will never be truly secure if the landlord owns the building, unless you either purchase it or find a different location for your business.

Google released Chrome in 2008.

Building Chrome turned out to be a very good investment.

Chrome, by default, uses Google Search, and most users stick with it, ensuring billions in ad revenue. This setup allows Google to track a wealth of user activity through the browser. By 2025, Chrome held a commanding 60-70% of the global browser market across all devices, with Safari trailing behind. Controlling the browser landscape proved to be a savvy move.

How does Google’s budget for competitive maintenance look?

We know the following, even though Google does not publish the “Chrome budget” as a separate line item:

Google continues to pay Apple between $18 and $20 billion annually to maintain Safari as the default search engine on Macs and iPhones.

That is not Chrome itself, but rather Safari. In retrospect, it simply meant:

“Securing default search in a browser is worth tens of billions per year to us.”

That makes one think:

• Chrome and Chromium cost billions of dollars a year to maintain and require a large engineering and infrastructure team, with thousands of engineers working on desktop, Android, iOS, security, performance, user interface, and standards.
• It’s actually a fairly reasonable investment because the combined revenue from search and ads has consistently exceeded $100 billion annually.

If they are willing to pay Apple $20 billion annually to borrow one (Safari), you can imagine the value of owning one (Chrome).

Smartphones, Android, and the “Play”

Google had a new issue after the world switched from PCs to phones, and this is all well documented; in fact, it coincides with the time when they had to deal with numerous lawsuits:

What if the phone maker controls search and ads instead of us?

As a result, Google acquired Android, a tiny startup, in 2005. Why? Because mobile platform lockdowns by Nokia, Microsoft, BlackBerry, and eventually Apple would put Google search and ads on the back burner.

Why Android is so important

Right now:

• Approximately 70–75% of mobile operating systems worldwide are Android
• Chrome (or a Chromium-based browser) comes preinstalled
• The OS is intricately integrated with Google Search and other Google services

Because of this stack, Google got:

Guaranteed distribution: On most Android phones (outside China), Google services are the default.

Data and ads at the operating system level: This includes telemetry from Android, Google Play, and Chrome.

Regulators see this too; the DOJ explicitly floated forcing Google to spin off Android as part of search remedies.

What Does “We Want Your Data” Actually Mean?

Let’s time-travel a bit.

50 years ago (1970s–80s): data = paper records + early databases

“Data” mostly meant:

• Bank account balances
• Credit histories (think early credit bureaus)
• Retail purchase logs and loyalty stamps
• TV ratings from panels like Nielsen

It was mostly structured (rows and columns), collected relatively slowly (monthly statements, manual forms). No one was tracking every click you made. The tech didn’t exist.

1990s–2000s: the web, cookies, and ad tech

Now “data” starts to mean:

• Web server logs: what pages people visit, when, and from where
• Cookies: tiny browser identifiers that let sites recognize you
• Search queries: what you type into Google, Yahoo, etc.
• Email metadata: who emails whom, when

Companies use this to:

• Build interest profiles: “likes sports,” “shopping for car,” etc.
• Optimize websites (A/B tests)
• Run targeted ads across sites through ad networks

This is when “if you’re not paying, you’re the product” starts to become a thing.

2010s–2020s: AI, social media, and smartphones

The concept of “your data” has evolved into:

• Real-time behavioral data points
• Location/mobility
• Social graph
• Device + sensor data
• Financial & transaction data
• Training data for AI models

When a company today says, “We want your data,” in practice it often means:

• We want every interaction you have with our product
• We want to link it to a persistent identity
• We want to use it for profiling, optimization, and AI training
• Occasionally, we want to share or sell it

It’s the raw events (clicks, taps, and searches), the profiles or segments (“likely car buyer,” “high-value gamer”), and AI models that provide insights like “users like you tend to favor layout X.”

The AI Browser Landscape

Everyone is building the AI browsers because why not make a bet against Chrome and for data? Perplexity has already launched its browser, which I find to be a much better experience than Atlas, by the way.

A data point I am interested in getting is the ratio of desktop to mobile users using OpenAI’s customer-facing apps. It won’t surprise me much but will confirm a couple of things about how users can make adjustments if the product is excellent.

Why AI Browsers Are a Poor Investment

Building browsers is a technical job, and the raw engineering and ecosystem effort that goes into it gets very little praise. The development is riddled with thousands of dependencies, and things go wrong all the time.

If OpenAI follows through to continue building their own browser, they’d immediately carry the cost of the maintenance overhead, as I can see that they have added a layer on top of the Chromium engine clone and extended the work with Swift (created by Chris Lattner for Apple in 2010). This simply means that Windows users will not receive any updates in the near future, especially considering the overhead costs they are inheriting.

In terms of products, Google Chrome already includes some AI components in its development area; users can utilize the AI SDK integrated into the browser for voice transcription and text-to-speech (TTS) functions. I discovered this when I was building random personal software. They have also released AntiGravity, which I will review at some point. I believe the long-term goal would be to do more AI-native user-facing features beyond dev tools.

Google is currently in the best position to win the escalating AI browser war, due to their extensive organizational expertise in building browsers and their current strategy for deploying AI applications, such as NotebookLLM and AntiGravity. Also, word on the street is that a couple of ex-senior Google employees have moved back from Microsoft to Google, and they are really excited about what they are building. Quite encouraging.

The experience with the Atlas browser is quite disappointing. And I am uncertain if they are A/B testing their way into figuring out what we already figured out a long time ago.

• We want search to be blazingly fast
• There are different types of searches, but the one we do most times doesn’t need to be quick

On Atlas, it appears that there are plans to rewrite that feature, but the automated mode still makes many mistakes. Personally, I believe they will figure out all these issues; if you’ve been around long enough, you know that technology always catches up.

I am more interested in…

What Do I Think Is Better Than Browsers?

Personal Automations.

The rapid growth of personal software leads me to believe that a standard marketplace for personal automations is imminent.

OpenAI’s other acquisition, “Sky,” serves as a good example of solid design and a strong team that knows their way around the Mac, as it folds natural speech into Mac workflows and floats around your Mac like a little companion. Second evidence on why I know OpenAI doesn’t give a toss about Windows.

I plan to delve deeper into this topic the next time I tackle this vertical.

On automations and why they work, a good example is being able to send my code to GitHub in record speed. In the past I’d have to handcode my branches—branches are the way we add new features to an existing codebase.

Yesterday, I wrote a script that takes my commits and creates a new branch based on them. Now, I no longer have to think about branching, and the script provides me with three suggestions to choose from; the only drawback is that I have to wait one second for the suggestions to return.

Regular folk will want to do things like this and do this for everything. It’s also extremely sticky, probably stickier than personal websites. I have not opened most of my personal websites in decades.

Originally published on LinkedIn, November 2025.