Users.
Users demonstrate, by their behavior, again and again, that they want simple, they want turnkey, they want inter-operation, and they want stability. And the big corps generally succeed in providing those things. Web 1.0 and most of Web 2.0 was none of those things.
The easiest way to engineer those goals is to consolidate. Consolidated software developed under one roof can use incentive of money to induce engineers to figure out the "grimy" problems to make pieces of the tooling work with each other (such as GMail interoperating with Docs, or Facebook Messenger and Facebook itself). And the centralized dev houses force consolidation; for all the jokes (justifiable) about Google having too many chat options, how does it compare to the world outside their garden? How many people actually go through the hassle to get XMPP inter-operating, since Google dropping support for it didn't mean it actually went away? Not enough. A consolidated dev house doesn't have to have multiple competing social network platforms or image hosts, etc... They can maintain one, maybe float the occasional experiment, and users don't have to worry too much about change.
Users actually hate many kinds of change because they build their lives atop the predictability of systems they don't control and they get angry when those systems are altered without their consent. And to head off the argument "But the FAANGs change shit all the time..." They sure do. Now compare that rate of churn to the rate of churn of how one stands up a web server, or how many non-FAANG online services have come up, run out of money, and evaporated.
Inter-operation, at first glance, is the thing the walled gardens fight against. But ironically, by consolidating to one solution and letting just about everyone on the internet use it, they create an "interior inter-operation" to rival the inter-operation you can get with diverse solutions. Consider how much of Twitter you see doing a Twitter search vs. a Mastodon search, or how easy it is to find your coworkers on LinkedIn vs. finding (and using, successfully, through their spam filters) their email addresses. The single-service, one-API, few-clients model consolidates multiple things that users don't want multiple of... chat clients, search corpuses, histories, administration stories, username / password combinations, privacy policies. And this all carries consequences (get locked out for legitimate or illegitimate reasons and you're now locked out of a very big garden), but they're consequences that hit the few; for the near totality, "it just works."
Regarding stability: Google has been around a quarter of a century at this point. Facebook about two decades. They may not last forever, but users understandably are more likely to stick with them than to try new things. Every new thing is more likely to be a pets-dot-com than the next Amazon, and who has time to build familiarity with someone's bespoke solution just to get that inevitable "We never hoped to have to send this message..." turndown email?
And this all culminates in the lack of proliferation of browsers, because the browser enables all these features and, at the end of the day, no user (rounding for error) wants to run two or more browsers and no web developer (rounding for error) wants to accomodate the undocumented implementation quirks of two or more browsers (quirks that always exist no matter how spec-compliant they are... If they didn't exist, they'd be the same browser). So there's the occasional cry of "Why aren't there dozens of browsers" without anyone really answering who wants that?
Nothing about a "rewilded" net is impossible, but users don't want to do it and don't really want to use it. Not nearly enough to make it into something as big as the consolidated service providers. It's a nice hobby for people who really care about how the machinery works to dabble in (I, too, run a Mastodon node...), but I don't see anything pushing people out of their Facebooks, their Amazons, or their Googles, in the near future.
Not if the alternative is for regular users to have to do more work than they do today.