Cloudflare's Act 4 bet: OpenAI pays for web page update signals

Cloudflare recorded its first revenue from “Act 4” products in the third quarter but disclosed no dollar figure; overall quarterly revenue stood at $782.5 million, up 39% year over year.

Co-founder and CEO Matthew Prince said on the earnings call that Act 4 covers products meant to create a new business model for the internet. The company’s results release did not break out those products, leaving their size unknown.

What is Cloudflare charging for?

Cloudflare’s July 8 research pilot used signals such as content freshness and page changes from participating websites. The company explained the OpenAI arrangement on its August earnings call: Cloudflare tells the crawler when a page has changed, so OpenAI can avoid repeatedly fetching unchanged content.

Prince said: “OpenAI is paying for that.” Cloudflare did not disclose the price or when payments began. In an Oct. 15 results posting, it said it was still working with OpenAI to test whether better discovery produces business outcomes.

Prince expects other AI companies to pay for the same efficiency, but that remains a forecast. He described Cloudflare’s work helping sites attract agent traffic as a new version of SEO, with owners potentially accommodating, metering or charging those agents.

A Goldman Sachs analyst asked when Act 4 would produce meaningful revenue. Prince said Cloudflare already recognized significant amounts from more efficient agent visits, but gave no figure, and the $782.5 million release did not isolate Act 4.

What does this change for publishers?

Customer demand is split. In response to a Wells Fargo analyst, Prince said some Cloudflare customers want as many agents as possible, while others see agents as a business threat and want them kept away. Cloudflare helps the first group attract and handle agent traffic; the second group blocks it.

Blocked agent traffic carries no charge from Cloudflare, so the customer does not pay for visits it rejects. Prince’s example: an agent might pull menus from 1,000 restaurant websites, the user chooses one, and the other 999 restaurants bear the cost without receiving the visit. A person, he said, might view only two or three pages.

Agent traffic at that scale would require more bandwidth and servers, Prince said, so it cannot remain a free ride. He previously forecast nonhuman traffic at up to 1,000 times human traffic within five years if trends continued; now he expects that sooner. That remains a forecast, not current traffic data.

The practical controls already exist. AI Crawl Control comes with every Cloudflare plan and lets users permit or deny individual AI crawlers. Cloudflare’s documentation lists Pay Per Crawl in private beta. But Prince mentioned no single Cloudflare service for optimizing a website for agent traffic, an omission that limits the promised playbook.

Could agent traffic become the bigger business?

Prince expects micropayments, with sites potentially charging for agent visits and charging less when an agent explains its purpose. He described a travel site that could admit a free Hawaii-planning agent for a family of four because the request creates advertising opportunities. He could not say when micropayments would become a major revenue source.

Looking 10 years ahead, Prince said: “I would imagine that this becomes the largest part of Cloudflare’s business.” That is a personal forecast, not disclosed demand.

Two payment products entered beta on Sept. 30. Monetization Gateway lets sellers charge AI agents per request, while Pay Per Use pays publishers who accept an AI company’s offer for reported uses.

The only detailed Act 4 payment went to Cloudflare for making OpenAI’s crawling more efficient. The call did not say whether websites in the pilot receive any share, so publishers have revenue potential but no disclosed payout formula. Cloudflare also did not disclose OpenAI’s payment size or start date.

What happens next?

Prince expects Cloudflare to discuss Act 4 more in coming quarters. Two tests matter: whether other AI companies pay for update hints, and whether future disclosures show site owners sharing in the revenue. Until then, publishers can allow or block crawlers, but they lack pricing and revenue data to budget agent traffic as dependable income.