In July 2001, a CSX freight train derailed in the Howard Street Tunnel in Baltimore. Four tanker cars spilled flammable chemicals and acids in a fire that reached 1,500 degrees Celsius -- hot enough to melt fiber-optic cables running through the tunnel. The result was catastrophic: the Northeast U.S. internet went down for hours, cell phone service in Maryland failed, and email even in Africa was disrupted. A local accident costing $12 million revealed something unsettling about the digital infrastructure that now dominates the world: it is infinitely more fragile than it appears.
Almost 25 years later, Pulitzer-winning journalist John West published "The Internet Will Die, and So Will You" -- a book that encapsulates a thesis echoing through weeks of Wired's Ghost Stories column: the internet as we know it is already dying, and much of it doesn't deserve to be saved. What West and other observers are describing is not a single infrastructure failure event, but a slow, cumulative, and largely voluntary process of degradation that is transforming the network into an ecosystem of synthetic content, dead links, and hollow interfaces.
The central paradox is simple, but uncomfortable. When Tailai Hang, digital editor of the Global Times, said in 2024 that "the internet is going to disappear," the rest of Western media laughed. But West's analysis -- and the data backing it up -- suggest that the internet won't disappear on a specific day, or through a single point of failure. It will disappear by degrees, millimetrically, like a building whose foundation is being corroded drop by drop.
The collapse has already begun
Research has already documented alarming numbers. According to a 2016 study cited by West, three out of four citations in scholarly articles have drifted -- the websites they purport to cite no longer say what they once said. More than half of Wikipedia's entries contain at least one dead link in their references. A 2024 Pew Research report confirmed that a quarter of New York Times links are now completely dead.
But dead links are only the visible symptom of a deeper structural disease. The internet is composed of layers of dependency, and each layer has failed in predictable ways. DNS, BGP routing protocols, data centers, content delivery networks -- everything depends on physical infrastructure, energy, and human decisions. And each of these layers has accumulated single points of failure that make partial collapses not just possible, but probable.
The Baltimore tunnel case was only the first major warning. Since then, we've seen failures of similar proportions: the 2025 BGP routing collapse that knocked 90% of global internet connections offline for hours; the Namecheap data center outage in August 2026 that affected 24 million domains; the power failure at Google's London data center that paralyzed services for companies like Revolut. None of these events destroyed the internet -- but each one reminded us that the internet is not an intangible cloud service. It is a collection of cables, chips, and servers just as subject to fires, electrical failures, and human error as any other physical infrastructure.
AI contamination
If physical fragility was the left side of the equation, the right side is content contamination. What Wired called "AI slop" -- mass-generated artificial intelligence content of questionable quality and often unverified -- is transforming the internet from a library of humanity into a river of digital sewage.
In August 2026, Google researchers presented data on how YouTube eliminated 130,000 low-quality content channels over six months. The scale is brutal. This is not a few bots generating spam: it is entire ecosystems of creators, many powered by automatic generation tools, who have built entire careers on content produced at industrial scale by algorithms. YouTube purged the channels, yes -- but the same logic is happening everywhere: Spotify, LinkedIn, TikTok, Pinterest, Facebook, Instagram, Threads. Entire platforms are dealing with the same problem: monetization incentivized mass production, AI made that production cheap, and the result is an ecosystem saturated with content lacking value.
The study "Retrieval Collapses When AI Pollutes the Web," presented at the WWW 2026 conference, quantified the risk rigorously. The authors, Hongyeon Yu, Dongchan Kim, and Young-Bum Kim, defined a new and disturbing concept: "Retrieval Collapse" -- a two-phase collapse where, first, AI-generated content dominates search results, eroding source diversity; and second, low-quality or adversarial content infiltrates the retrieval pipeline. In their experiments, 67% pool contamination generated over 80% exposure to contaminated results. The outcome is a homogenized but apparently healthy state: answer accuracy remains stable, but the sources are no longer human or authentic.
This is not theory. It is already happening. Search engines, Retrieval-Augmented Generation (RAG) systems, and even LLMs themselves are being trained on data that contains increasingly AI-generated content. It is a feedback cycle -- or, as the researchers call it, a self-reinforcing quality degradation cycle.
What is lost when the internet dies
There is something West raises that very few people mention: the vast majority of what exists on the internet was never truly human in the first place. The very notion of a "digital vastness" is an illusion. The internet, in its essence, occupies only a tiny fraction of human experience -- and of that fraction, much was already ephemeral communities that never came back. GeoCities disappeared. Tumblr died. Early-2000s forums evaporated. And what remains are zombie pages, full of spam and obsolete ads.
West recalls that his own middle-school anime site survives as a zombie page full of spam, while his college blog disappeared when he forgot to renew the domain. It is almost a perfect metaphor for the state of the internet: it is not that the internet has died. It is that as original content disappears and is replaced by synthetic waste, the network becomes less useful, less authentic, less alive.
The problem is that the internet no longer has a centralized control point that can decide what content to preserve. The Web has no owner -- and this lack of ownership is simultaneously its greatest strength and its greatest weakness. Any preservation effort requires someone to decide what is worth saving, and in that filtering process there will always be bias, error, and exclusion.
The future: less internet, more human
The question is not whether the internet will die, but what kind of internet will replace the current one. There are signs that platforms are trying to push back. YouTube eliminated 130,000 channels. TikTok requires labeling of AI-generated content. Reddit implemented bot protections. But each of these actions is reactive: they address the symptom, not the cause.
The root cause is economic. Social platform monetization has depended on screen time, and screen time depends on content. When content is made by humans, it is expensive. When it is generated by AI, it is virtually free. And no one, so far, has created a sustainable economic model that pays human creators to produce at scale while simultaneously preventing bots from flooding the platform.
The fragility of the internet is therefore dual: physical and digital. Physical infrastructure -- fiber-optic cables, data centers, chips -- has already shown, in Baltimore and in many more places, that it is infinitely more fragile than the "cloud" metaphor suggests. And the content infrastructure -- the network of sources, articles, images, and comments that makes the internet a useful resource -- is being corroded by mass contamination of synthetic content.
What makes the collapse inevitable is not a catastrophic failure, but the sum of millions of small decisions: a forgotten domain expiring, a data center without redundancy, an academic article linking to a page that has changed its URL, a creator generating 10,000 videos per month with AI, a search engine indexing results that no longer reflect reality. Each of these events, taken alone, is irrelevant. Together, they add up to an irreversible process of degradation.
And, perhaps ironically, that is precisely what makes the death of the internet as we know it inevitable: no one wants to save it, because saving it would require slowing down the entire digital economy, and the digital economy does not slow down.
Sources: WIRED, The New York Times, arXiv / WWW Conference 2026
✓ Independent sources cross-checked and verified before publishing
