Thomas Seifert: In the next five years, non-human traffic may reach 1000 times human traffic, making humans a negligible fraction of the internet
Cloudflare CFO Thomas Seifert predicts that in the next five years, non-human traffic could reach 1000 times that of human traffic, rendering humans a "negligible fraction" of internet request volumes. This assessment is based on the accelerating use of AI agents, crawlers, automated workflows, and machine-to-machine calls, rather than a simple increase in traditional bot traffic.
Cloudflare's data shows that non-human traffic has for the first time surpassed human traffic in its observed sample: automated requests account for approximately 57.4%, while human requests account for about 42.6%. Previously, the company’s management anticipated that bot traffic would exceed human traffic around 2027, but the expansion of AI agents has brought this crossover point forward; Cloudflare's second-quarter performance communication further projected a long-term gap of 1000 to 1.
The core reason for the multiplication of traffic is not the number of individual AI users, but rather the way agents access information. When a user purchases a camera, they might manually browse 5 websites; however, their AI agent may automatically visit 5000 websites to compare prices, specifications, stock, reviews, delivery, and discount information. A single human search, shopping, or research task can be broken down into hundreds to thousands of machine requests, thus detaching network load from the traditional measure of "human clicks."
Cloudflare believes that the binary classification of "human or bot" is becoming ineffective. The more important questions in the future will be: Is the traffic authorized? Is it aggressive? Does the crawling load correspond to the value returned? Is there ad fraud? Does it come from trusted agents? Does it comply with the access and payment rules set by the website? AI assistants, privacy agents, legitimate automation tools, and malicious crawlers may use similar technical characteristics, but their commercial value and security risks to websites are entirely different.
In market mechanisms, AI model companies, agent platforms, search and shopping assistants are the buyers of new request volumes; cloud service providers, CDNs, cybersecurity companies, anti-crawling service providers, and data authorization platforms sell computing power, bandwidth, verification, access control, and content interfaces. Website operators will incur higher costs for bandwidth, servers, databases, and security, yet may not receive corresponding human advertising revenue; platforms that master bot detection, API billing, content authorization, and rate limiting will benefit, while small publishers, e-commerce sites, and content communities relying on free public web scraping traffic will be under pressure.
Source: Public information
ABAB AI Insight
The past business model of the internet was built on "human visits to web pages." Search engines crawled web pages to provide links to users; users clicked on websites, which monetized through ads, subscriptions, e-commerce transactions, or data collection. Even though search bots and malicious crawlers have always existed, the main value chain of the web was still designed around human browsing behavior: page views, ad impressions, cookies, conversion funnels, click-through rates, and dwell time. AI agents are disrupting this chain. Users no longer need to open web pages one by one; models can read, summarize, compare, fill out forms, call tools, and complete transactions directly; what websites receive may be machine requests rather than human page views. For publishers, this means content may be consumed by AI systems without corresponding ad impressions; for e-commerce platforms, consumers may shift from "browsing product pages" to "having agents find the lowest prices and place orders on their behalf"; for the advertising industry, traditional display ads and click attribution will lose some visibility. The 1000 to 1 ratio described by Cloudflare is not a prediction of user numbers, but rather a prediction of network request density. Humans are limited by time, attention, and manual operations, and can only access limited information for each decision; agents can read thousands of pages in parallel, repeatedly call APIs, compare different sources, and continuously monitor prices or stock. The surge in traffic does not mean the internet has 1000 times more consumers, but rather that behind each consumer, there may be a large number of machine agents, automated scripts, and model inference chains. Historically, search engine crawlers forced websites to establish robots.txt, sitemaps, CDN caching, and a search optimization industry; the mobile internet forced websites to restructure into responsive pages, in-app payments, and push notifications. The agent internet will bring about a third infrastructure rewrite: websites need to shift from being "human browser-oriented" to simultaneously serving humans, trusted agents, unauthorized crawlers, and malicious automation programs. Authentication, machine-readable permissions, access quotas, content licensing, API payments, and agent accountability will become new issues at the network protocol layer. Cloudflare's proposal to "go beyond humans and bots" is essentially a transfer of pricing power. In the past, websites could assume all visitors were potential ad users; in the future, a large number of visitors will be machines acting on behalf of others, and websites must determine whether to serve these machines, whether to charge for access, and whether to impose restrictions. Those who can identify trusted agents, convert content and tool calls into billable services, and block worthless automated traffic will regain pricing power in the agent internet. ABAB News · Law of Cognition
- Humans create demand, agents amplify requests
- Traffic no longer equals users, requests no longer equal value
- When machines become visitors, authorization becomes the new ad space.