"In five years, humans will be a 'rounding error' on the internet." This one line from the Polymarket account passed 190,000 views and 1,400 likes and spread fast. Read on its own, the sentence sounds like an apocalyptic declaration that human users are being pushed off the internet. But when you check the context of the actual remark, the story is much more specific and points in a different direction.
The person who said it is Thomas Seifert, Cloudflare's chief financial officer. On the earnings conference call, he drew a clear line: "Humans become a rounding error not because human traffic is shrinking, but because non-human traffic is growing that fast." According to The Register's report, if current trends continue, non-human traffic is projected to reach 1,000 times human traffic within five years.
The Reversal Has Already Happened, a Year Earlier Than Expected
The more striking fact isn't a forecast but something that has already happened. Non-human traffic actually overtook human traffic in May 2026, more than a year ahead of the 2027 estimate Cloudflare had previously given. TechSpot's report points to this reversal as a direct result of the spread of AI agents.
The reason lies in a structural difference in consumption patterns. When a person shops, they browse about 5 sites on average. An AI agent, by contrast, responds to a single prompt by checking up to 5,000 sites on the user's behalf, firing off thousands of requests in an instant along the way. Cybernews's report also explains this gap as the core cause of the traffic surge. It's a structure where the footprint one user leaves is amplified from one into hundreds or thousands of requests.
More Important Than the Viral Line Is What Site Operators Do Next
For developers and site operators, this figure isn't apocalypse. It's an infrastructure issue. It means the vast majority of the traffic that should drive server costs and request-handling capacity design is already bots and agents, not people. Services with open APIs that need no login, and sites that have controlled access only through robots.txt, have reached the point where they need to separate out agent traffic and rewrite their pricing and rate-limit policies. The decision about whether to show content only to people or also open it to agents in a structured form is no longer something that can be put off.
But there's something to flag before accepting this figure as absolute fact. Seifert himself added a caveat: "my predictions have been wrong before." The "1,000x in five years" number isn't a confirmed result but an estimate that simply extends the current growth curve. He also stressed again that "this doesn't mean human activity is shrinking. Human traffic keeps growing too, but machine traffic is growing much faster."
In the end, there's a difference in temperature between Polymarket's provocative line and the actual remark. A more accurate summary is that humans aren't disappearing from the internet, but the nature of the traffic filling the internet is changing. And the pace of that change, at least judging by the past year's data, is arriving faster than predicted.