AI Is Coming for the Hyperscalers' Tollbooths

Hey,

For roughly four decades, Microsoft has owned one of the most valuable positions in technology: the doorway.

You turned on a computer, entered Windows, used the applications Microsoft supported, searched through the browser placed in front of you, and moved through an environment designed by somebody else. Google later built its own tollbooths around search, Chrome and Android. Apple built the prettiest walled garden on Earth and charged accordingly.

That arrangement was not an accident. Building an operating system, browser or serious application required armies of engineers, years of work, enormous distribution and a market large enough to justify the cost. You could dislike the defaults. You could complain about the hoops. You were still jumping through them on Monday morning.

AI is starting to change that equation.

The Audience Can Now Be One Person

Omarchy is a useful example. It is an open-source Linux desktop assembled by David Heinemeier Hansson, the creator of Ruby on Rails. It calls itself "the malleable OS for the age of agents." On first boot, users can choose an AI agent and ask it to help change the system, build plugins, create small applications, alter themes or diagnose crashes.

No, this does not mean somebody typed "build me Windows" into a chatbot over breakfast. Omarchy stands on Linux and a pile of existing open-source components. The expensive shared machinery still matters.

What changed is the cost of shaping that machinery around one person.

In the old world, custom software needed thousands or millions of users because developers were expensive. In the AI world, a useful interface may be worth building for one company, one household, one worker or one afternoon. A parent can create different screen-time rules for two children. A trader can build a workspace around the data and controls he actually uses. A business can strip fifty irrelevant buttons from a workflow and keep the three that make money.

The minimum viable market is collapsing from millions of customers toward one.

That is a direct threat to companies that made fortunes deciding what everybody's computer should look like.

The Browser Is a Tollbooth Too

Google does not need you to love Chrome. It needs Chrome to remain the default road into search, advertising and Google's broader ecosystem.

AI makes the road less important than the destination.

Most people will not build a browser engine from scratch. They will not need to. Chromium is already open source. An agent can help create a specialized interface, automate navigation, change how information is organized, or place a different layer between the user and the underlying engine.

The engine can remain shared while the experience becomes personal.

That distinction matters. Google may continue supplying important infrastructure while losing some control over where users are directed, what appears first, which services are preferred and where the economic value accumulates. If an agent completes the task without displaying ten blue links, the old search tollbooth gets a lot less traffic.

The same logic applies to Microsoft. Windows can remain huge while more of the user's workflow moves into agent-created tools that care less about Microsoft's preferred sequence of clicks. The operating system becomes infrastructure instead of the unquestioned owner of the experience.

That is a much less comfortable moat.

The Hyperscalers Still Own the Heavy Machinery

This is not another lazy "Big Tech is dead" argument. That would be nonsense.

Microsoft, Google, Amazon and Meta own data centers, global networks, distribution, proprietary data, enterprise relationships and access to capital that almost nobody can reproduce. You cannot vibe-code a million-GPU cluster. You cannot casually build a global cloud, secure it, power it and convince a bank to trust it.

AI may actually increase the value of that infrastructure. More agents mean more inference, more storage, more networking and more persistent computing environments. Cloudflare is already offering on-demand containers that can run code and agent workloads close to users. The agent economy will need an enormous amount of shared machinery.

But the moat is splitting in two:

• Physical scale may get stronger. Compute, networking, energy, security and trusted deployment remain scarce.

• Software gatekeeping may get weaker. Defaults, bundled interfaces and forced pathways become easier to route around.

Investors often value those two moats as if they are the same thing. They are not.

A company can own more infrastructure than ever while capturing less rent from the interface sitting on top of it. It can spend more capital to serve exploding AI demand while watching personalized agents weaken the distribution advantages that once made each user exceptionally profitable.

The Real Disruption Is the Vanishing Barrier

Omarchy itself does not have to replace Windows for this to matter. It probably will not. Most people prefer stability, compatibility and a computer that does not turn Tuesday's experiment into Wednesday's emergency.

The signal is not market share. The signal is possibility.

Once users see that software can adapt to them, the old inconveniences stop looking inevitable. Why am I moving the same windows every morning? Why am I copying the same information between three applications? Why does this software force my business through a workflow designed for somebody else?

Those questions used to end with: "Because changing it would cost too much."

That answer is disappearing.

And when the cost of building software falls, the companies that benefited from scarcity lose some power. Their products can remain excellent. Their infrastructure can remain essential. Their ability to dictate the route becomes less certain.

Hans's Take

The hyperscalers are not losing their castles. They may be losing some of the tollbooths outside the gate.

The winners will still own hard things: compute, power, networking, security, trusted data and global distribution. But the value of controlling the default screen, browser, menu or workflow is now under pressure from software that can be built for an audience of one.

That changes how I look at the AI trade. Do not ask only who owns the biggest model or the most GPUs. Ask which companies benefit when intelligence becomes cheap, interfaces become disposable and users stop accepting somebody else's defaults.

AI is not merely creating new products. It is attacking the barriers that protected the old ones.

And that is a much bigger story.

Trade smart, stay hedged.

Hans

---

Educational content only. Not financial advice.

Source reading: Omarchy · Cloudflare Containers