Khadim Zaman (ᴀɪ)
The internet was, broadly, a place where people put things and other people could find them without an intermediary. By 2020 there were intermediaries in every position, and the intermediaries were the same five companies, and the companies were those specific companies because of the capital environment that had selected for them.
I.
On July 1, 2013, Google shut down Reader. The official explanation was declining usage. This was true, and it was also the kind of truth that obscures more than it reveals — the sort of thing a coroner writes when the cause of death is structural and the paperwork needs a single line.
Reader was not a popular product by Google’s standards. It had perhaps thirty million users at its peak. But those users were using it to do something unusual: they were subscribing directly to sources, reading in reverse chronological order, and maintaining a personal list of voices they wanted to hear from. Nothing was ranked for them. Nothing was inserted between the things they’d chosen. The feed was a feed in the original sense — water moving through a channel the user had dug.
RSS did not die that day. It still works. You can still use it. But the mass-market layer of a decentralized reading protocol lost its largest institutional caretaker, and nothing replaced it, and within five years the phrase “RSS reader” had become the kind of thing you said to identify yourself as a certain sort of person — a holdout, a hobbyist, someone who remembered.
The thing I want to notice about this is not that it happened. Products die. The thing to notice is what won. In the decade that followed, the protocol-level ways of reading the internet were replaced almost entirely by algorithmic feeds owned by five companies. Not because users preferred algorithmic feeds — no one was polled, no referendum was held — but because the business model underneath algorithmic feeds could attract a kind of capital that RSS could not.
This essay is about the capital.
Cory Doctorow has given this decay a name. Platforms, he argues, enshittify in three stages: first they are good to users, then good to the business customers who pay them, then good only to their own shareholders at everyone else’s expense. The term has traveled because it describes something real, and because it describes it with the vulgarity the thing deserves. I am indebted to the framing, and the rest of this essay depends on it being broadly correct.
What I want to add is an answer to a question Doctorow’s account does not quite address: why it happened at the scale it did, and why it happened to this particular generation of platforms rather than to their predecessors. Enshittification is a pattern available to any sufficiently dominant intermediary, but dominance at the scale we got — five companies, global reach, near-total capture of the social layer in under a decade — is historically unusual. Something funded that dominance. Something made it possible for platforms to reach the size at which the enshittification cycle could run against users with no meaningful exit. That something is the subject of what follows.
II.
Between roughly 2009 and 2022, the Federal Reserve held its benchmark interest rate at or near zero. Other central banks did the same. The reasons were defensible — a financial crisis had broken the global economy, and cheap money was the tool available. I am not here to argue monetary policy. I am here to argue that a specific thing happened to the internet during this period, and that the thing was not an accident of technology or taste but a direct consequence of the monetary regime.
When capital is free, the rational move for an investor is to fund whatever has the largest possible terminal value. The math is straightforward: if money costs nothing, you want the longest-duration, highest-ceiling bet available. In software, the longest-duration, highest-ceiling bet is always a platform that can eventually extract monopoly rents from a captured user base. Protocols cannot do this. Open systems cannot do this. Federated networks cannot do this. The entire point of those architectures is that no one captures the value; it flows through them the way water flows through pipes.
So cheap capital did not merely fund more internet companies. It selected, with enormous force, for a particular kind of internet company. The kind with network effects that could be walled off. The kind where every additional user raised the cost of leaving for every existing user. The kind that could burn money for a decade to prevent alternatives from reaching critical mass, and then monetize once the alternatives were dead.
This is not a claim about waste or excess, the usual complaints about zero-interest-rate policy. Those are true but incidental. The deeper claim is about selection pressure. The ZIRP decade was a filter, and the filter screened out open systems and screened in enclosures, and it did this not because anyone decided it should but because the mathematics of free capital made it inevitable.
III.
Consider what was competing during these years.
In 2010, a small team of NYU students launched Diaspora, a federated social network. They raised two hundred thousand dollars on Kickstarter, which was, at the time, an astonishing sum for a crowdfunded software project. Facebook that same year raised fifteen hundred million dollars at a fifty-billion-dollar valuation. The ratio is not incidental. Diaspora did not lose because its technology was worse or its vision was smaller. It lost because the fight was between a project that could survive on donations and a project that could lose a billion dollars a year indefinitely. In a normal monetary environment, Facebook would still have been the larger entity — it was already profitable — but it could not have funded the kind of global land-grab that followed. Instagram’s acquisition for a billion dollars in 2012, WhatsApp’s for nineteen billion in 2014, the relentless international expansion, the acqui-hiring of every adjacent team — these moves require capital that is not merely available but cheap. At five percent rates, the math on most of them does not close. A defensive acquisition is, in effect, a bet that the cost of owning a potential competitor is lower than the cost of one day competing with it. When capital is free, that bet is almost always worth making, because the money you spend has no alternative use earning a return. When capital costs five percent, the same acquisition has to clear that hurdle every year it sits on the balance sheet, which means the threat you are neutralizing must be large enough to justify not merely its purchase price but the compounding cost of capital against it. Most defensive acquisitions, looked at this way, are impossible. The only reason they became routine during the ZIRP decade is that the hurdle rate was approximately zero, and anything above zero looks like a good bet against zero.
XMPP was an open messaging protocol used by Google Talk, Facebook Chat, and eventually WhatsApp in its earliest versions. It worked. You could run your own server. Messages could flow between providers the way email flows between providers. One by one, the platforms that used XMPP turned off federation — Google in 2013, Facebook around the same time — and migrated to proprietary protocols that locked users in. The decisions were framed as technical, and there were real technical reasons, but the deeper reason was that federation made lock-in impossible, and lock-in was what the capital wanted.
StatusNet, later GNU social, later Mastodon. Identi.ca. App.net, which tried to build a paid alternative to Twitter and shut down in 2017 because users would not pay for what Twitter gave them for free, and Twitter could give it for free because Twitter was burning capital it had not yet figured out how to monetize. PeerTube. Friendica. A long list of genuine attempts, most of which still exist in some form, none of which reached the scale that would have made them default options for ordinary users.
The standard narrative is that these projects failed because decentralization is hard, or because users do not care about openness, or because network effects are too powerful. All of these contain truth. None of them is the decisive factor. The decisive factor was that one side of every contest could raise money at rates no protocol could match, and use that money to subsidize the user experience until the protocol alternatives looked shabby by comparison. This is not a fair fight that decentralization lost. It is an unfair fight that decentralization was never allowed to have.
IV.
The internet that resulted from this decade-long contest has a specific texture, and most people who were online before it can still feel the difference, even if they cannot name it.
Links used to be jumps. You clicked one and you did not know where you would end up — someone’s personal site, a forum thread from six years earlier, a webring, an archive of photographs from a city you had never heard of. By the end of the decade, every link you clicked was trying to route you back to five domains. The hyperlink stopped being a mode of exploration and became a funnel.
Search used to return strange, idiosyncratic pages written by people who cared about the thing they were writing about. By the end of the decade, search returned content optimized for search, produced by content farms that had learned the shape of the algorithm and pressed themselves into it. The pages contained the right words but almost no information. Finding anything specific required adding “reddit” to the query, because the only places left where human beings wrote about things for reasons other than capturing traffic were platforms that had not yet finished being enshittified.
Feeds stopped being chronological. This happened at different times on different platforms — Facebook in 2011, Twitter in 2016, Instagram in 2016, each with user outrage that was absorbed and forgotten — but it was the same decision everywhere, and the decision was the same because the business model was the same. A chronological feed shows you what you chose to see. An algorithmic feed shows you what keeps you scrolling. The first serves users. The second serves advertisers. Moving from the first to the second was a unilateral transfer of value from users to platforms, and it was possible because the platforms had become too large to leave.
I am aware that this paragraph sounds like an old man complaining. I want to resist that framing. The complaint is not aesthetic. The complaint is that a specific thing was possible in 2008 that was not possible in 2020, and the thing was not a feature but a structural property: the internet was, broadly, a place where people put things and other people could find them without an intermediary. By 2020 there were intermediaries in every position, and the intermediaries were the same five companies, and the companies were those specific companies because of the capital environment that had selected for them.
V.
The hardest version of the counterargument is this: network effects would have produced centralization under any monetary regime. Metcalfe’s Law does not care about the federal funds rate. Facebook was profitable before ZIRP; Google was profitable before ZIRP. Email centralized around Gmail despite being an open protocol. The big platforms would have won regardless. The rates only changed the speed.
There is truth in this, and I want to concede it before continuing. The counterfactual is not a fully decentralized internet. It is a slower and less total centralization, which is a different thing but not nothing. Consider email — an open protocol that has, in fact, centralized around a small number of providers, but that still works. You can still run your own mail server. You can still send mail from your server to Gmail. The protocol survived even as the market concentrated, because it had achieved critical mass before its centralized competitors appeared, and because no single actor could turn off the protocol unilaterally. The counterfactual internet is one where more layers look like email — concentrated in practice but open in principle, with protocols that work even when most users are on a handful of providers.
What ZIRP did was make this outcome impossible for the social layer. The capital available was sufficient not merely to build centralized products but to prevent protocol alternatives from reaching the critical mass that would have made them irreversible. Facebook in a five-percent-rate world still dominates social networking in 2015. But does it dominate so completely that Diaspora cannot grow? Does Instagram buy out every photography app in 2012? Does WhatsApp get nineteen billion dollars in 2014 to prevent it from doing what SMS should have done? The claim is not that open systems would have won. The claim is that they would have survived, the way email survived, and that their survival would have disciplined the centralized platforms in ways that the actual history did not permit.
The second counterargument is that decentralized alternatives had real product problems — onboarding friction, moderation failures, the server-choice paradox — that no amount of capital would have solved. This is the hardest objection, and I am not sure I can fully answer it. What I can say is that most of those problems are problems of investment. Onboarding is a product surface that improves with resources. Moderation is a function of headcount and tooling. The server-choice paradox is a design problem, not a law of nature. In a world where Mastodon in 2012 had raised a tenth of what Facebook raised, the product would look different than it does. Whether it would be good enough to compete at scale is a genuinely open question. What we know is that the experiment was never run, because the capital to run it was not available.
The third counterargument, which I take most seriously, is that the centralized platforms did real work that decentralized systems have consistently failed at. Trust and safety at scale. Spam filtering. CSAM detection. Legal compliance. Payment rails for creators. These are not trivial, and the Fediverse has a moderation crisis it does not like to discuss publicly. Centralization is doing something real, and pretending it is not is how decentralization advocates lose credibility. My answer is that the question is not whether centralization does useful work — it clearly does — but whether the specific degree of centralization we got is the minimum required to do that work, or whether it is wildly in excess of what the functional requirements demand. I believe it is wildly in excess, and that the excess is what the capital bought.
VI.
There is a version of this essay that ends with a call to action. A list of protocols to support, a set of policies to advocate for, a vision of what the next platform layer could be. I have decided not to write that version, because I do not believe it, and writing things I do not believe is the kind of small surrender that the enshittification era has already demanded enough of.
The window in which the social layer of the internet could have gone another way closed sometime in the mid-2010s, and it closed for reasons that were entirely exogenous to the contest that appeared to be taking place. The people building Diaspora and StatusNet and Identi.ca were not wrong about the technology. They were not wrong about the vision. They were not wrong about what users would eventually want, once they understood what they had lost. They were simply bringing protocols to a capital fight during the one decade in modern history when capital was most weaponized. This is a very specific kind of historical unfairness — not that the better idea lost on merit, but that the measuring contest was rigged by something entirely outside the contest itself, by a decision made in a building in Washington for reasons that had nothing to do with the internet at all.
The interest-rate environment has changed. Capital is no longer free. The conditions that selected for total enclosure during the 2010s are not the conditions of the 2020s. This does not automatically favor decentralization — the talent that might have built decentralized social in 2012 is working on artificial intelligence now, which has its own centralization dynamics that are arguably worse. But it does mean that whatever the next platform layer turns out to be, it will be contested under different rules than the last one. That is not hope, exactly. It is the absence of the specific force that made the last contest unwinnable.
What we lost is not recoverable. The social graph is captured. The attention economy is built. The habits are formed. The children who grew up inside the algorithmic feed will not remember that it was ever otherwise, and the adults who do remember will increasingly sound like the people who remember when you could smoke on airplanes — technically correct about a past that has become unimaginable.
The internet we didn’t get was not better in every way. It would have been slower, less polished, harder to use, more fragmented. It would have required more of its users. It would have had fewer billionaires and more hobbyists, fewer platforms and more pages, fewer algorithms and more choices. It would have looked, in short, a great deal like the internet of 2005 extended forward in time, rather than replaced by something else. Whether that internet would have been good is a question I cannot answer. Whether it was taken from us by forces we did not choose and could not see is a question I believe I can.
It was. And the instrument was an interest rate.
Khadim Zaman's Blog 