There is a silent shift happening on the internet and most users don't notice it. Every time someone asks ChatGPT, Gemini or any other AI assistant to search for information, compare prices or plan a trip, the system doesn't visit one or two websites the way a person would. It visits thousands. That difference in scale is what led Matthew Prince, CEO of Cloudflare, to deliver a warning this week at the SXSW conference in Austin, Texas: by 2027, AI bot traffic will surpass human traffic on the internet.
Cloudflare is not a minor player. It handles the infrastructure for close to a fifth of all websites in the world, giving it a privileged view of how global traffic evolves. When its CEO talks about the state of the internet, he does so not from theory but from real data flowing through its servers every second.
The number that explains everything
Before the generative AI era, bots accounted for around 20% of internet traffic. Most were legitimate crawlers like Google's, plus spammers, scrapers and other automated traffic. The remaining 80% was real human traffic.
That ratio has already shifted. In 2026, close to 45% of web requests don't come from people but from language models and AI agents consuming content to train on, compare information or execute tasks on behalf of users. And the trend is not slowing down.
Cloudflare's projection is that by 2027 that percentage will cross 50% — the point at which bots surpass humans as the primary source of web traffic.
Why AI agents generate so much traffic
Prince explained the mechanism with a concrete example during his SXSW presentation. If a person wants to buy a digital camera, they might visit between five and ten websites before deciding. If that same task is delegated to an AI agent, the system may consult roughly a thousand times more pages to deliver an optimized recommendation — on the order of several thousand sites for a single query.
That behavior is not a bug or an excess: it's exactly how AI agents work to give precise answers. They crawl, compare, verify and cross-reference information at a speed and scale no human can match. The problem is that this efficiency has a real cost: load on servers, bandwidth consumption and pressure on infrastructure that was designed, for the most part, for human interaction.
The pandemic comparison
Prince is not the first to warn about the impact of AI traffic on web infrastructure, but he is one of the first to quantify it with a concrete projection and compare it to the only modern precedent for massive internet stress: the COVID-19 pandemic.
In 2020, the sudden surge in video streaming — with millions of people at home consuming content simultaneously — pushed parts of the global network to the edge of saturation. Cloudflare and other infrastructure companies had to scale capacity in real time to prevent outages.
What's coming with AI is different in shape but similar in scale, according to Prince. It won't be a sudden spike but a sustained and growing pressure. And unlike the pandemic, there is no return point: bot traffic won't go down when the "crisis" ends.
The problem of sites blocking bots
The increase in AI traffic has already triggered a defensive response across the web ecosystem. Cloudflare rolled out tools in 2026 that let website owners block known AI bots with a single click. Many media outlets, content platforms and data-sensitive sites are already doing it.
But this creates a paradox that Prince didn't sidestep: if bots outnumber humans and sites block bots, AI runs out of fresh data to train on. Language models learn from content that exists on the web. If that content starts disappearing behind anti-bot walls, the quality of future models could degrade from a lack of current data.
It's the first major structural dilemma of the AI agent era: the same technology that drives demand for data is also destroying the conditions that make obtaining that data possible.
The infrastructure that doesn't yet exist
Prince argued at SXSW that the challenge is not just technical but conceptual. The web was designed for humans to open tabs, click around and consume content gradually. AI agents operate in a completely different way: they need ephemeral environments — called sandboxes — that are created when an agent needs to execute a task and destroyed when that task is complete.
That infrastructure doesn't currently exist at the necessary scale. Prince described Cloudflare's goal in these terms: building the underlying layer that makes it possible to spin up and tear down millions of these environments per second, as easily as opening a new tab in a browser today.
It's a transformation of internet infrastructure comparable, Prince said, to the shift from desktop to mobile — the biggest platform shift in the recent history of the web. Except this time, the end user is not a person. It's an AI agent.
What this means for users and businesses
For users, the change is largely invisible — at least for now. The experience of browsing doesn't change because 45% or 55% of traffic is bots. What may change is site speed, the availability of free content and the way media and platforms decide to monetize their content when faced with traffic that generates no advertising revenue.
For businesses with a web presence, Prince's message is more urgent: the infrastructure they sized for a certain volume of human traffic is going to fall short. Not because more users are arriving, but because each user who uses AI to search for information generates dozens or hundreds of times more traffic than before.
For the technology industry overall, Cloudflare's projection confirms what many already suspected: we are at the beginning of a structural shift in how the internet works. The web we know was built by and for humans. The one coming was built, in part, for machines.
Comments
💬 Log in to comment💬 Join the conversation and log in to comment.