CEOInterviews.AI
Start App
Get an email when Matthew Prince gives a new interview
One email per new interview. Unsubscribe in one click.
Matthew Prince
Co-Founder, Co-Chairman & CEO, Cloudflare

Cloudflare CEO Says AI Agent Traffic Could Be 1,000X Human Traffic in Five Years

📅 Oct 07, 2026 CoinDesk 40 MIN 320 VIEWS 53 SEGMENTS · 3 SPEAKERS
Cloudflare CEO Matthew Prince sits down with CoinDesk's Jennifer Sanasie to explain why the internet as we know it is about to be rewritten. Prince reveals that AI agent traffic has already overtaken human traffic online, more than a year ahead of his own forecast, and he predicts it will be 1,000 times larger within five years. He breaks down why agents don't click on ads, why Cloudflare needs a payments system that can handle more than 100 times Visa's volume, and why today's blockchains are "three orders of magnitude" too slow to power it. He also explains why AI agents could actually bring...

What Matthew Prince said

Written from the verified transcript and checked against it. Every figure links to the moment it was said.

Matthew Prince, CEO of Cloudflare, said automated agent traffic surpassed human traffic in early 2026, far ahead of earlier projections, and he predicts it will reach 1,000 times human traffic within five years. He discussed the internet's shift from an ad-driven to a knowledge-based economy, arguing agents don't click ads and will need a micropayment system for accessing content. He estimated the average US consumer will spend about $1,200 per year on AI agents, a figure larger than Google's per-household ad revenue, and warned that developing-world users without that budget may face predatory ad-supported models. He highlighted challenges for small businesses in an agent-dominated internet, advocated for a future with 500,000 AI companies rather than five, and noted Cloudflare is exploring payment infrastructure to handle 10 million transactions per second on day one, potentially using crypto rails.

Key takeaways

  1. Agent traffic surpassed human traffic in early 2026 and could be 1,000 times higher in five years.
  2. Average US consumer will spend about $1,200 per year on AI agents, more than Google or Meta earn per household from ads.
  3. Cloudflare needs to handle 10 million transactions per second on day one for this system, scaling to 100 million.
  4. Prince predicts local media like his newspaper will earn more licensing to AI than from digital advertising this year.

Numbers and commitments

FigureWhat it refers toTypeAt
$1,200 Average annual US consumer spending on AI agents other 12:07
10 million transactions per second Required transaction processing volume on day one for micropayments commitment 8:38
$750 to $850 Google's ad revenue per US household per year metric 12:07
$250 to $350 Meta's ad revenue per US household per year metric 12:07
100 million transactions per second Scaled transaction target for micropayments commitment 28:31
80,000 transactions per second Visa's maximum transaction processing rate at peak metric 8:38
8,000 Circulation of Prince's hometown newspaper metric 37:07

Chapters

  1. 0:00Agent traffic surpasses human traffic
  2. 2:05Internet's evolving business model
  3. 17:36Small business threats in agent era
  4. 23:21Cybersecurity outlook and AI risks
  5. 28:18Crypto role in agent payments
  6. 32:48Privacy in an agent-driven internet
  7. 35:18Vision for knowledge-based economy

Questions asked in this interview

11
  1. 1:44And where do you feel the status quo is maybe getting it wrong?
  2. 4:14... timeline was a little off when you're looking at bot traffic, but did you think we would be sitting here today talking about what we're talking about, or did you think maybe the next iteration of the internet would have looked differently?
  3. 6:33And what needs to change from today?
  4. 11:36In this future, are AI agents maybe being incentivized to only consume the cheapest information out there, and therefore that information is the information that gets widespread, and the more high-quality information is not as discoverable?
  5. 18:29Do you think enough people care about those small businesses when we're talking about 70% of employment in the United States?
  6. 20:00I mean, do you think when you look at the technology layer, decentralization starts to address some of the issues that you've been talking about?
  7. 26:19You mentioned some of the stuff we're seeing in the headlines reads like sci-fi. Are you a sci-fi reader?
  8. 28:18Do you see stablecoins and crypto as the currency that agents are using?
  9. 30:23... to people who are building in this industry and so I'm just curious to get into your head as someone who's been building the internet for the last so many years, like where do some of the solutions presented by the industry fall short?
  10. 32:08In the future, in an internet that's being driven by both humans and AI agents, where does privacy come in and what are some of the challenges you think we're going to have to face?
  11. 35:03... a lot more recently and I want to give you the opportunity to just kind of talk to us a little bit about what is the message that you want to get out there, like what do you want people to know about Cloudflare and what you're building?
Matthew Prince 0:00 ↗
The automated traffic, the agent traffic, is already past human traffic online in early 2026, way ahead of what our schedule looked like. It was just like, wow, this is incredible. And if you extrapolate that out, with the caveat that I've been wrong in every prediction I've made so far, it looks like in five years, agentic traffic is going to be a thousand times as much as human traffic. Not because we think human traffic is going to decline — I think it'll hold pretty steady — but agents are just exploding, and I think that's going to put all kinds of different pressures. What is the future business model of the internet? But also, who's going to pay for all the infrastructure that these agents are going to inherently use?
Narrator 0:43 ↗
Crypto markets move fast. You need research that keeps up. That's The Stack — Grayscale's blog with sharp analysis from the team that's tracked digital assets since 2013. Head to grayscale.com/thestack.
Digital markets move fast. Move forward with clarity from S&P Global Ratings — risk analysis spanning stablecoins, tokenized funds, digital bonds, DeFi protocols, and more. Explore digital asset solutions with S&P Global Ratings at spglobal.com/ratings.
Interviewer 1:13 ↗
Matthew Prince, welcome to CoinDesk Spotlight.
Matthew Prince 1:15 ↗
Thank you for having me.
Interviewer 1:17 ↗
All right, I'm really excited about this conversation. You've been at this for a long time — almost two decades at Cloudflare.
Matthew Prince 1:24 ↗
That made me just feel really old.
Interviewer 1:25 ↗
No, I don't want you to feel really old.
Matthew Prince 1:28 ↗
Literally before we started here, we were talking about how I am going to like a geriatric dance party this weekend.
Interviewer 1:34 ↗
I just want to let the audience know. Yes, we were talking about that and it got me excited.
Matthew Prince 1:39 ↗
There you go. Well, I'll see you there.
Interviewer 1:41 ↗
Yeah, exactly. I'll see you at the dance party this week.
Matthew Prince 1:43 ↗
The geriatric dance party.
Interviewer 1:44 ↗
The geriatric dance party. All right. Talk to me. I know you've spoken a lot about the lessons you've learned over the years, but talk to me about what you're building towards. What does the next iteration of the internet look like to you? And where do you feel the status quo is maybe getting it wrong?
Matthew Prince 2:05 ↗
Yeah. I mean, I think what's really so interesting and fascinating about this point in time is that the internet felt like this very fixed thing. It's existed, it's had a really established business model largely driven by ads over the years, e-commerce as well, a little bit of subscriptions, but that's been what's powered the internet for the last 30 years. It's the internet that I grew up with, it's the internet that everyone is familiar with. And as AI is taking off, what's happening is we've got this new interface for the internet, which is the AI prompts and everything that they do. And because people are consuming information, because increasingly they're going to be conducting commerce through their agent, it's changing the business model of the internet. Agents don't click on ads. Agents can use one subscription and spread it across many, many different users. All of those things are changing exactly how it happens. And at the same time, the volumes are just crazy.
I was asked in November of 2025 at Web Summit, a big technology conference, when did I think that automated traffic would pass human traffic online. And you know, we looked at all the data, we pulled it and thought, oh, it looks like it's going to be the second half of 2027. A few months later at South by Southwest in March of 2026, I was asked the same question again, pulled the data again, and we were blown away that it had been moved up to the first half of 2027. And again, it just showed how much things were accelerating. And so when my team in May came to me and said, 'You won't believe it, but the automated traffic, the agent traffic is already past human traffic online in early 2026, way ahead of what our schedule looked like,' it was just like, 'Wow, this is incredible.' And if you extrapolate that out, with the caveat that I've been wrong in every prediction I've made so far, it looks like in five years, agentic traffic is going to be a thousand times as much as human traffic. Not because we think human traffic is going to decline. We think it'll hold pretty steady, but agents are just exploding. And I think that's going to put all kinds of different pressures. What is the future business model of the internet? But also, who's going to pay for all the infrastructure that these agents are going to inherently use?
Interviewer 4:14 ↗
I want to get into that in just a second, but if you think back to five years ago, did you think we were going to be on this trajectory? I know that the timeline was a little off when you're looking at bot traffic, but did you think we would be sitting here today talking about what we're talking about, or did you think maybe the next iteration of the internet would have looked differently?
Matthew Prince 4:30 ↗
No, I mean, five years ago, you know, it's 2021. We were in the midst of COVID and I think that what we were trying to do is just hold the internet together. Cloudflare felt like the one thing that was sort of holding everything together. And then a year later, as COVID was declining, I remember personally thinking, gosh, that was exciting. I felt like it was really, really important. Now we've got to get back to just sort of doing the regular stuff of signing up kind of enterprise customers. And I sort of looked at that and thought, you know, does that sound like fun or not? What are we going to do? And then it was shortly after that that ChatGPT gets released. And you could, as soon as that happened, see how that was going to change a lot of what the internet did and the implications for it. And again, even seeing that four years ago, I don't think you could have anticipated all the things that changed.
Here's another stat. So from about 2012 to 2025, the internet kind of plateaued and was actually declining. The number of web pages that were disappearing was faster than the number of new web pages that were getting created. And it was one of those things we watched somewhat concerningly that more and more of the internet was sort of going behind either walled gardens of places like Facebook or YouTube. And fewer people were actually creating things out there. That changed about mid-2025 because of all the new tools that are allowing just a much broader set of people to create. And if you look at the growth of the web today, it rivals what we saw sort of in the early 2000s where there's just an enormous amount of new interesting things that are coming online. And yeah, some of it's slop and some of it's garbage, but there's also a lot that's new. People that didn't previously have coding skills, didn't previously have the ability to create things, suddenly have the tools to do that. And they're using platforms like Cloudflare to actually host and run them. But then they're using the tools of companies like Lovable or Replit or Base44 in order to actually build these new sites. And it's amazing today that literally anyone can be a coder.
Interviewer 6:33 ↗
It is kind of crazy how much the internet has changed in the last five years, how much access people have now to information they didn't previously have before. If you think about that future, you're building towards one where bots are the predominant users of the internet, humans are still there. What does that look like? I imagine it looks pretty different. You know, right now we're really focused on user interfaces and user experiences when the users are all machines. What does that look like? And what needs to change from today?
Matthew Prince 7:04 ↗
Yeah. I mean, I think the answer is nobody quite knows. So I think in the camp of everything might change. Bots are really efficient at getting to things. They're not motivated as much by emotion. They have almost infinite time in terms of what they can scan. If I ask a bot today, we're sitting here in Manhattan, I say, where should I go to lunch? It might scan a thousand different menus and say, from everything I know, you should go to this particular place. And I think that's amazing as an end user. But it's putting a lot of load on all of those places that the 999 places I don't go to lunch have to serve their menus, but the one that's there. So I think we're going to have to figure out how to become more efficient, how to actually give bots the information they want. On the side of things might not change as much as you think. All of these bots are actually trained on how humans interact with the web. So they're looking at web pages and they're seeing that information. And so what they are really good at is the same thing that humans are really good at at some level, at navigating the web as if they're a human. So I think that there will be things that change, but my actual prediction would be less about the web and less about the internet changes than you might expect.
Interviewer 8:24 ↗
Let's talk about the economic model. Now I know that you've said that AI firms should pay for their agents accessing information around the web. Talk to me a little bit about what that could look like, what that new economic model looks like.
Matthew Prince 8:38 ↗
Well, let's start with just the basic problem, which is that we have a giant potential tragedy of the commons problem, which is that you have a whole bunch of these agents that are going to run around the internet and if not constrained by anything, they're going to access as much information as possible. So today, I might say, 'Where should I go to lunch?' And it might access a thousand menus. In the future, if there's no cost to them accessing everything, why wouldn't bots just access literally every piece of information on earth and say, ah, I found this obscure Chinese manuscript from 1318 and it suggests that if you've had a day like the one you had yesterday, you should definitely eat ramen, right? I mean, those are the sorts of things where you will make the haystack infinitely large in the search for a needle if there's no cost to doing that. And so something's got to impose some cost on these agents. And my sense is what that's going to be is some form of payment whenever an agent accesses a resource, just to cover the cost of actually serving that resource. And it could be a tiny amount, a thousandth of a penny every time you do that. But I think a little bit of what you're paying to the AI companies in order to have access to their services, that's going to go back to the places where they're getting that information in order to cover the cost of doing that. I think on top of that there will be organizations like media companies that will actually say, we want to impose a cost which is higher than that because we've got to pay the people who are the journalists that are doing the work, that are conducting the interviews, they're writing the stories, and some bit of that actually goes to it. And so what we're thinking about at Cloudflare is, you know, we see that there are a whole bunch of willing buyers in the market. The AI companies are hungry for more information, more data. They want access to more of that and they're willing to pay for it. There are a whole bunch of sellers in the market, all the different companies that say, listen, we need to have us cover our costs and we need to be able to actually create value for the media content that we're creating. And we sit in that middle, you know, where 80% of the large AI companies are already Cloudflare customers, you know, huge percentage of the web uses us, and we're bringing those two sides together and saying, how can we make a market around this in some interesting way? The challenge is the volumes. So if you look at the largest payments network in the world, it's Visa. Visa processes about 80,000 transactions per second at peak during the holidays. We think that the day that we turn this on, we'll have to process about 10 million transactions per second. Much smaller. Visa's average transaction size is like $25. This will be thousands, you know, a thousandth of a penny for everyone that's there. But still a financial transaction, and it still has to be in an enormous volume. And so we're searching for either companies that have already built that or internally whether we can build it ourselves. And honestly, we're sort of agnostic to which outcome comes, but we know that that technology has to be created in order to do that. And it is just way beyond the transaction volume anyone's ever seen before.
Interviewer 11:36 ↗
I'm curious how you're thinking about this. You know, as you were talking, I was imagining AI agents going out there searching for information. Information costs different amounts of money. You know, an investigative report at CoinDesk might cost more than an aggregated report at another publication. In this future, are AI agents maybe being incentivized to only consume the cheapest information out there, and therefore that information is the information that gets widespread, and the more high-quality information is not as discoverable?
Matthew Prince 12:07 ↗
You're hinting at what I think is going to be a real challenge for society going forward, which is if you look at the internet, the last 30 years of the internet, the miracle has been that advertising is actually a micropayment system that has discriminatory pricing based on who the user is. So to make it make sense, when an ad is served to me, it gets a higher rate, a higher CPM than when it's served to a kid in Sri Lanka. Higher for me because I have more disposable income. I occasionally buy really stupid things and so advertisers like to pitch me on doing that. The kid in Sri Lanka doesn't have a lot of disposable income, but the cost of serving them the page is still lower than the amount that you get for that ad. And so the internet has generally been open and it's generally made information open to as many people as possible. My hunch is that the average US consumer will probably spend about $1,200 on their agents every year. Where that money will come from, my hunch is it'll come from companies like Netflix and sort of the discretionary entertainment budget that most households have. That's a lot of money. If you compare it to Google, Google makes between $750 and $850 per US household per year. Meta makes between $250 and $350 US dollars per US household per year. And so $1,200, a ton of money to be competed over and the agents. What's then interesting is it's not clear that any one of them is going to be vastly superior than the others. I mean, in my case, I switched from ChatGPT to Claude to Grok just depending on who has the best model at any given time. And it's not clear that one is going to just run away with it. I think you're going to have very easy switching costs between, which means that what the models will compete on is actually who is the most reputable, who has the best data, who has access to that. And so I think for the Western world, the developed world, you're going to see that actually quite the opposite. People are going to actually pay their agents to go find the highest quality data even if it costs them something because again you've got a big pool of money that can go into that and that's how the AI companies are going to differentiate themselves. Where I'm worried is actually in the developing world. So what happens to the kid in Sri Lanka because he doesn't have or she doesn't have $1,200 to spend on the AI agent? So you're going to see much more in those spaces what will be ad-supported business models, but it's ads for the developing world which are often the most predatory, the most abusive. And I think if you think about an AI that has predatory and abusive ads, you could imagine just a huge parade of horrors. So I think we've got to figure out, like AI for wealthy parts of the world, I think, is going to be amazing and give you access to maybe some information you wouldn't have access to even before and help you make better decisions. But we may, the pendulum may swing back the other direction where the parts of the world that can't afford to pay for it get closed off from the biggest tools and AI may be something that really accelerates the haves over the have-nots.
Interviewer 15:07 ↗
Build on that a little bit more for me in the scenario you've just laid out. Give me an example of what it might look like for someone in the emerging world if we continue down this path.

24 more exchanges in this transcript

Sign in free to read the rest of this interview. No card required.

Sign in to read the full transcript

Cite this transcript

APA, MLA, BibTeX
APA

Prince, M. (2026, October 7). Cloudflare CEO Says AI Agent Traffic Could Be 1,000X Human Traffic in Five Years [Interview transcript]. CoinDesk. CEOInterviews.AI. https://ceointerviews.ai/interview/3000027/

MLA

Matthew Prince. "Cloudflare CEO Says AI Agent Traffic Could Be 1,000X Human Traffic in Five Years." CoinDesk, 7 Oct. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3000027/.

BibTeX
@misc{prince2026_3000027,
  author       = {Matthew Prince},
  title        = {Cloudflare CEO Says AI Agent Traffic Could Be 1,000X Human Traffic in Five Years},
  howpublished = {Interview transcript, CoinDesk. CEOInterviews.AI},
  year         = {2026},
  month        = {oct},
  url          = {https://ceointerviews.ai/interview/3000027/},
  note         = {Speaker-attributed transcript with timestamps}
}