Cloudflare: AI bots could surpass humans in internet traffic by 2027

Cloudflare is warning of a shift that could quickly change the economics and architecture of the internet. According to the company's CEO Matthew Prince, as early as 2027, AI bots will begin generating more web traffic than live users.

Why Traffic Is Growing

The main reason is the agentic format of work. When a person searches for, say, a digital camera, they typically open several websites, compare prices and specifications, and then make a decision. An AI agent works differently: it can visit thousands of pages for a single task, collect options, check availability, delivery terms and reviews, and then return a ready answer. In terms of network load, one such assistant turns out to be much "heavier" than an ordinary visitor.

"Your agent or bot will often visit 1,000 times more sites than a real person,"

Prince said at SXSW.

According to Prince, before the generative AI boom, bots accounted for about 20% of internet traffic. The largest legitimate bot was Google's search crawler, while a notable portion of the remaining requests came from fraudsters and other bad actors. Now the structure is changing: more and more bots are working not for indexing or attacks, but to fulfill user tasks. For Cloudflare, whose infrastructure serves roughly one-fifth of the web, this shift already looks not like a theory, but an operational reality.

New Network Load

The key problem here is not the fact of automation itself, but its scale. When a single agent, in order to choose a product, takes a path that previously required dozens of people or hours of manual work, the load is distributed across thousands of sites, APIs and servers. Prince compared the situation to the pandemic period, when the internet rapidly swelled due to video services like YouTube and Netflix. The difference is that back then there was a sharp spike followed by a plateau, whereas now Cloudflare sees more gradual but continuous growth with no clear signs of slowing down.

• More requests to product pages, knowledge bases and reference sections
• Higher load on CDNs, caches, anti-bot systems and DDoS protection
• Greater need for new servers, racks and data centers
• More complex analytics: sites need to distinguish useful agents from junk traffic

If this trend continues, the internet will have to scale not only logically but also physically. This is no longer just about smarter models, but about additional computing infrastructure, communication channels and fault-tolerance mechanisms. Otherwise, site owners will find themselves in a situation where they need to simultaneously let in useful AI agents, block malicious bots and not lose performance for ordinary people. For network platforms, this is turning from a research topic into a task for the nearest releases and budgets.

Sandboxes for Agents

Another idea that Prince mentioned is special "sandboxes" for AI agents. The logic is simple: if a user asks the system to do something on their behalf — for example, plan a vacation, find flights or collect accommodation options — the agent needs an isolated environment where it can run code, browse websites and complete work without unnecessary risks to the main infrastructure. According to Cloudflare's vision, such environments should be spun up almost as easily as a new browser tab, and destroyed just as quickly after the task is complete.

In this model, the web begins to look more like a factory of short-lived processes than a set of static pages. Prince allows that in the future, millions of such sandboxes will be created every second. If the forecast proves correct, this will change not only the infrastructure layer, but also the rules of content access, authorization, request limits and methods of payment for resource usage.

Simply put, the internet will have to learn to work not only with people and search crawlers, but also with an army of autonomous executors acting on behalf of users.

What This Means

Cloudflare's forecast is important not only as a trend assessment. It describes a shift in which AI agents become normal participants of the web, not an exception. For media, e-commerce, SaaS and infrastructure companies, this is a signal to already reassess protection, performance and access rules, because the next round of competition will go not only for human attention, but also for bandwidth for machines.