Somewhere right now, a person on a three-year-old phone is waiting for a news article to load. The article is about eight hundred words long. The words themselves, if you saved them to a file, would take up less space than a single photo of the journalist’s cat. But before those words can appear, the phone must download JavaScript frameworks, font files, advertising scripts, analytics beacons, a consent-management platform, and the machinery to auction the reader’s attention in real time. The person gives up. Nobody notices. The analytics dashboard simply records one fewer session, and the dashboard cannot say why.
This is the quiet trick the modern web has pulled off: it got enormously heavier, and because it got heavier gradually, on connections that were getting faster at the same time, almost nobody felt the weight arrive.
The numbers are not ambiguous. The HTTP Archive’s Web Almanac 2025, which crawls millions of pages and measures what they actually transfer, puts the median home page at 2,862 KB on desktop and 2,559 KB on mobile. In October 2014, the mobile median was 505 KB. The median page has roughly quintupled on mobile in a decade, and developer Alex Russell, working from the same underlying data, notes that the typical mobile page passed the size of the entire 1993 video game DOOM — all its levels, monsters and music — sometime in 2025. A standard floppy disk held 1.44 MB. A web page now outweighs it comfortably.
Here is the part that makes the weight interesting rather than merely annoying: it is not the content. The Almanac’s breakdown of the median page is a small scandal in four figures. Images account for 911 KB. JavaScript for 632 KB. The HTML — the actual document, the thing the page ostensibly exists to deliver — is 22 KB, under one percent of the bytes. The text is the lightest thing on the page by an order of magnitude. Everything else is apparatus: the scaffolding erected around the words to style them, track who reads them, monetize them and render them in a font the reader did not ask for.
The apparatus, not the article
One blogger who recently weighed a flagship news site’s front page clocked it at over 50 MB transferred — one person’s measurement, not a census, but a vivid one — while observing that the article text on such sites typically runs under 50 KB. That is a thousand-to-one ratio of machinery to message. The Web Almanac’s 2024 third-parties chapter found that nearly all websites load code from at least one outside party, which is how a paragraph of reporting ends up chaperoned by ad-tech, tag managers, A/B testing suites and session-recording scripts written by companies the reader has never heard of.
The most perverse fixture of this apparatus is the consent banner. A site, wishing to comply with privacy law, installs a consent-management platform — itself a third-party vendor script, loaded before anything else, interrogating the reader about their preferences and frequently phoning home to record the answer. The system erected to protect you from surveillance is, structurally, another piece of surveillance infrastructure, and it must execute before the words can. Nobody has published a clean figure for what the average banner weighs, which is its own kind of telling: we measure the trackers obsessively and the tracker-permission slip hardly at all.
Who pays
Page weight is usually discussed as a user-experience problem, a matter of bounce rates and Core Web Vitals, as if the stakes were a mildly irritated reader. The Almanac itself states the real stakes better than most critics do: “A 5 MB page isn’t just a slow experience; it is an exclusive one. If a user cannot afford the data to load your page, your site is, by definition, inaccessible.”
That sentence deserves to be read slowly, because it identifies the mechanism by which exclusion became invisible. Weight is not distributed evenly across the population, and neither is its cost. For someone on a new phone and an unlimited plan, the median page’s extra megabytes are absorbed silently — a half-second here, a warm pocket there. For someone on a metered data plan, an aging device or a rural connection, the same page is a toll. And when that person stops visiting, they do not file a complaint. They simply vanish from the metrics that the site’s owners use to justify the weight. The analytics cannot see the people the analytics drove away. It is a feedback loop with no sensor for its own damage: the heavier the web gets, the more its heaviest costs fall on exactly the users least likely to be counted.
This is how an entire medium can exclude people while every dashboard shows green. The industry measures performance on the devices it buys for its own employees, on the connections in its own offices, and concludes that things are fine. The median page quintupled in weight over a decade and the median product meeting never came up.
There is a standard worth proposing, and it is not nostalgia for the text-only web. It is simply this: the document should be the heaviest thing a document is asked to carry, and everything else should have to justify itself against the reader’s data plan rather than the developer’s. The web did not get heavy because people demanded heavier pages. It got heavy because the people who paid for the weight and the people who decided on the weight were never the same people — and the former left without saying goodbye.