Why is OpenAI making a browser and unsettling a war that had ended years ago? The easy answer is “data collection”. But, there’s more to it than that. I don’t think Google left them any choice!

If OpenAI had built an operating system, they could theoretically collect everything there is. Why not do that, then? Because getting users to switch OSs is nearly impossible. The market share of operating systems has stayed stable for decades. The last time a new OS had any chance was when an entirely new class of devices — smartphones — came into being.

Past that impossible task lies the next insurmountable challenge: apps. Developers have no incentive to make apps for an OS without users and users have no incentive to adopt an OS without apps. Even if OpenAI used AI to make those apps themselves, changing users’ minds takes time. Time on the order of decades. Time that no AI company has.

There is a quicker path. About 80% of our digital lives are online now — we search, check email, write documents, create spreadsheets, meet, learn and entertain ourselves using web apps. AI companies can gain access to all that data by controlling the one app we use to navigate our online worlds with: the browser.

However, it begs the question: did OpenAI really have to create a new browser?

Relying on Chrome would be the most dangerous thing for OpenAI to do.

With more than two-thirds of the market share, Chrome is far and away the most popular browser in the world today. It’s fair to ask why OpenAI would want to pick a fight with this behemoth by creating a competing browser? Why not do what Manus AI did instead, and ship a plugin that can capture users’ data and watch their behaviour?

Because Manus AI is missing the point completely.

Users’ personal and behavioural data is only valuable if it is passed to yourLLM. Google owns Chrome. Google owns Gemini as well. As a plugin, OpenAI will be completely at Google’s mercy. Google controls all API surfaces. If threatened, they can — and most certainly will — stop any valuable data from being sent to any other model than their own.

If OpenAI went the legal route to get Google to play nice, it would take them years to get a result. That’s enough time for Google to cement any lead it has. And, there’s no guarantee that it will tilt things in OpenAI’s favour, or even if they win, that Google will not hobble access to data in some other way.

Making a browser of their own, while their brand identity still has momentum, is likely the only viable path left to OpenAI.

User and behavioural data is nothing to scoff at. However, I suspect there is a bigger plan afoot. Google is building Gemini directly into Chrome. Let the import of that sink in. Firstly, they make sure that all requests for intelligence flows to their model. Next, every request handled in the browser is a request that is not being sent to their servers, saving them staggering amounts of compute and energy.

Then comes the real play: what’s to stop Google from federating inference, or distributing the work of refining smaller sub-models, across the billions of devices running Chrome? Google will have access to a planet-scale supercomputer, for free.

More nefarious than Chrome’s long history of over-collecting user data? Or that it used to keep users logged in even when the interface said they weren’t?

Regardless of their capacity for brazenness, there is a possibility that we will happily hand them compute. Consider the work that Google DeepMind has done. It had taken humanity more than 50 years to deduce the 3D structures of about 100,000 proteins. Google DeepMind’s AlphaFold blew past that record and predicted the structures of more than 200 million proteins in about a year. This information and technology was then given away for free. That kind of generosity is staggering. AlphaFold didn’t just leapfrog humanity, it pole-vaulted it! Given this, if Google ever asked me to donate my idle compute cycles for scientific discovery, I will gladly switch on the option. I suspect many others will as well.

Self-interest can get us to switch on the option too. If any of these companies agreed to give us more access to their best models in return for some idle compute cycles, I’d probably do it since access to this intelligence is far more valuable to me.

They don’t have to strong arm us to get what they want. Altruism, barter, or even just good PR could do the trick.

Chrome is a dead end for OpenAI. They have little choice than to build their own browser. Fortunately for them, browsers had been stagnating for years. Arc, a new browser, proved that the market was ripe for reinvention. It is making a name for itself and has a loyal user base. Dia, another browser, introduced AI into the browser. OpenAI could introduce a new browser into this fray, especially one that introduced Computer Use.

I had occasion to use OpenAI Atlas’ ability to operate autonomously when I had to register for a series of online events. Quite unthinkingly, the organizers required me to register for each event individually, filling in the same details over and over again. Atlas got me registered while I, literally, made lunch!

Inroads are possible. Given that people don’t change their browsers very often, the users that OpenAI wins over will be theirs to keep. If Google does indeed have plans of federating inference, then eating into Chrome’s market share is not just a defensive strategy but an offensive one as well.

Google is poised to skyrocket its AI advances by building a global, distributed compute infrastructure atop Chrome. They are not going to cede their advantage to OpenAI by shipping any model but their own. OpenAI has no choice but to make their own browser. The browser market has been shifting for the first time in a decade. And, OpenAI has chosen the perfect moment to put a chink in Google’s armour.

This article was written in Thinkery, a writing tool with a built-in intellectual partner. Take a look!