Cloudflare Wants to Charge Agents at the Edge

Cloudflare’s network now handles more non-human than human traffic. On September 30, 2026, it launched two beta products designed to bill the agents instead of the publishers.

Every figure below is Cloudflare’s own, measured on its own network and published on its blog on 30 September 2026. Its network is a very large sample and it is not the whole Internet. The 63, 115 and 150 million figures are TOTAL HTTP requests per second. The 1,700 per cent growth figure is a separate measure of daily AI-agent requests, and the two should not be read together. Both products described are in beta. Nothing here is investment advice.

What Happened

On September 30, 2026, Cloudflare published a blog post stating that, for the first time, more than half of Internet traffic on its network is not human. All figures in this piece come from Cloudflare’s own measurements on its own network — a very large sample, but not the whole Internet.

Two traffic metrics appear in the report, and they measure different things. Total HTTP requests across Cloudflare’s network have roughly doubled — from 63 million per second at the end of 2024 to 115 million today, with peaks above 150 million. Separately, daily requests attributable specifically to AI agents on the network grew by more than 1,700 per cent over the past year. These are not the same figure and should not be read together.

In the most heavily crawled categories — software, IT and services, and financial services — Cloudflare reports that human traffic has declined by as much as 40 per cent in less than a year. That is an upper bound for those named categories, not an average across the web. Cloudflare’s own summary of the commercial effect is blunt: revenue per request is falling while costs are rising.

The key insight: The web’s monetization model was built on a human arriving, seeing something, and converting. When the visitor is an agent that reads a page and hands the user a summary, the publisher pays the bandwidth cost and the human never shows up. Cloudflare’s proposal is to move the billing to the request itself — before the content is served.

These are total requests per second across Cloudflare, not agent requests. The 1,700% figure is a separate mea
These are total requests per second across Cloudflare, not agent requests. The 1,700% figure is a separate measure of daily AI-agent requests, and the two should not be read together. All figures are Cloudflare’s own, measured on its own network.

The Structural Read

For thirty years, the web ran on a single bargain. Publishers let search engines crawl their content. Search engines sent human visitors back. Publishers turned those visitors into revenue. Being discoverable and getting paid were the same motion.

Cloudflare describes what broke it precisely. Answer engines read the page and return a summary. The reader gets the information. The publisher absorbs the serving cost. The human visit — the one that could be monetized — never happens.

The company’s response is architectural rather than advisory. It sits at the edge of its customers’ infrastructure, between the incoming request and the content being served. That position means it can see the traffic, identify the agent, and intercept the transaction before delivery. That chokepoint is the asset. The pricing features are downstream of it.

Matthew Conroy, Cloudflare — September 30, 2026

“For businesses that learn to serve agents and capture value from them, agents are additive. For those that do not, they are extractive.”

That framing matters because it is honest about the optionality. Cloudflare is not arguing that agents are bad for publishers. It is arguing that passivity is. The two products are its answer to what active looks like.

How the Two Products Divide the Problem

Cloudflare describes the split cleanly. Some resources are consumed entirely in a single call — APIs, tools, data endpoints — where every request is the use. The Monetization Gateway targets those. Publishers set pricing rules using the same Rules language already in Cloudflare. When a rule matches, the gateway returns an HTTP 402 using the open x402 protocol. The agent pays before content is served.

Other content is different. A page might be crawled once and inform thousands of downstream responses. Pay Per Use addresses that. In Cloudflare’s description: a buyer offers a price for a specific use of your content, you choose whether to accept, the buyer reports each use, and Cloudflare bills the buyer and pays the publisher.

One customer case is offered. API2PDF previously required account creation, then a credit card after the first month — a friction sequence that produced a conversion drop-off of more than 50 per cent. It now returns an HTTP 402 instead. That is one company’s experience, not a pattern.

Both products are in beta. The Monetization Gateway is a closed beta available only to eligible United States Cloudflare customers.

What Is Not There Yet

The commercial side is almost entirely unquantified. There is no published pricing, no list of participating publishers, no number of enrolled buyers, no transaction volume, no revenue figure, and no disclosure of what Cloudflare takes as its share.

Nothing in the announcement establishes that a single AI company has agreed to pay. A payment rail with no disclosed counterparties is an architecture. Calling it a market requires evidence that has not yet appeared.

Three Implications

FOR PUBLISHERS

The cost structure of serving content has changed regardless of whether any payment product succeeds. Human traffic is declining in heavily crawled categories. Bandwidth consumed by agents is real and rising. Publishers carry that cost today without a recovery mechanism. Whether Cloudflare’s model is the right mechanism is a separate question from whether a mechanism is needed.

FOR AI DEVELOPERS AND AGENT BUILDERS

An HTTP 402 response at the edge is a different kind of friction than a login wall or a rate limit. It is machine-readable and designed to be paid by a machine. Developers building agents that consume third-party data at scale will need to account for per-request costs that did not previously exist in their unit economics, if this model spreads.

FOR CLOUDFLARE’S COMPETITIVE POSITION

No other infrastructure layer has announced an equivalent mechanism.

Business Engineer Framework

The FDE Framework: Founders, Distributors, Enablers

Cloudflare is a textbook Distributor in the FDE model — it does not build the AI and it is not the content publisher. It controls the pipe between them. This piece is a direct application of how Distributors monetize position rather than product. The full framework maps where companies sit in the AI stack and which layer captures durable value.

Read the FDE Framework →

The Bottom Line

It has disclosed no buyer that has agreed to pay. Two closed-beta products, no disclosed buyer counterparties, and no commercial figures mean this remains a well-positioned architecture rather than a proven model. The position, however, is not in question.


91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

Every figure above is Cloudflare’s own, measured on its own network and published on its blog on 30 September 2026. This publication has verified none of it independently. Cloudflare’s network is a very large sample of Internet traffic, but it is not the whole Internet, and claims phrased as facts about Internet traffic generally should be read with that in mind. Two distinct metrics appear above and should not be combined.

The 63 million, 115 million and 150 million figures are total HTTP requests per second across Cloudflare. The growth of more than 1,700 per cent is a separate measure of daily requests from AI agents. Nothing above attributes the per-second totals to agents. The decline of as much as 40 per cent in human traffic is given by Cloudflare as an upper bound for named categories — software, IT and services, and financial services — not as an average.

The conversion figure of more than 50 per cent is a single customer’s reported experience. Nothing above states or implies any history, prior status or date for the HTTP 402 status code. It is reported only that Cloudflare returns a 402 using the open x402 protocol. Both products are in beta, and the Monetization Gateway is a closed beta limited to eligible United States customers.

Not established and therefore absent: any published pricing, any list of participating publishers or buyers, any transaction volume, any revenue, what Cloudflare takes, and whether any AI company has agreed to pay. Nothing above predicts adoption or outcome, and nothing here is investment advice.

Sources: blog.cloudflare.com · openrouter.ai · fourweekmba.com · blog.cloudflare.com · blog.cloudflare.com

Scroll to Top

Discover more from FourWeekMBA

Subscribe now to keep reading and get access to the full archive.

Continue reading

FourWeekMBA