No Priors Ep. 126 | With Cloudfare CEO Matthew Prince

Detailed Summary

Introduction (0:00 - 0:37)

• Matthew Prince, co-founder and CEO of Cloudflare, is introduced as leading a company that underpins a significant portion of the internet.
• The discussion will cover the economics of AI for publishers, the shift from chat to search, the evolving web architecture, current funding issues, and Cloudflare's stance on Google.

Cloudflare’s Role in Securing the Internet (0:37 - 2:08)

• Cloudflare is a $66 billion company with $1.8 billion in trailing revenue, often mistaken as primarily a CDN.
• Prince clarifies that Cloudflare started as a security company, aiming to put a firewall in the cloud and make the internet faster, more reliable, secure, efficient, and private.
• The company's initial goal was to achieve parity in speed, but they ended up making things significantly faster.

The Road to Cloudflare’s Dominance (2:08 - 3:20)

• Cloudflare launched in September 2010, nearing its 15th anniversary.
• Their dominance stems from focusing on customer needs and solving problems, often acting as "customer zero" for their own solutions.
• This journey led them to build a free service, scale to millions of customers, develop robust security, and even create their own registrar after a hacking attempt.

The Internet’s Shift from Search to AI (3:20 - 6:34)

• The dominant value creation model of the past 30 years was search, driving content creation for sales, ads, or ego/fame.
• The web is now shifting to an AI interface, with trends showing decreased Google usage and Google itself morphing into an AI company.
• This shift means consuming derivatives rather than original content, leading to a significant drop in clicks for content creators.
• A Pew Research study indicates that AI overviews reduce link clicks, and it's 10 times harder to get a click from Google than a decade ago.
• For OpenAI, it's 750 times harder, and for Anthropic, it's 30,000 times harder to get clicks compared to the "Google of old."
• Prince worries that if AI doesn't send traffic, content creators will stop creating, starving both the web and AI companies that rely on this content as fuel.

Role of Agents and Content on the New Web (6:34 - 9:44)

• The rise of agents that perform actions on users' behalf will further reduce direct web interaction.
• Commerce agents and AI are likely to be beneficial for businesses, but content presents a different challenge.
• The current problem is the assumption of free content, with creators often looking to copyright law for solutions.
• Copyright law favors derivatives, making it difficult for creators to claim infringement against AI models that transform content.
• Prince suggests a new business model where content creators are compensated, noting that 80% of major AI companies (Cloudflare customers) agree they should pay for content.
• The main concern for AI companies is a level playing field, ensuring all players pay for content, not just some.

Reshaping the Content Market Online (9:44 - 13:05)

• To create a market for content, scarcity must be introduced, as current content is often given away for free.
• Cloudflare spent a year working with publishers (print, video, audio, music, film) who expressed concern about AI taking value without giving back.
• On July 1st, Cloudflare launched "Content Independence Day," blocking AI training by default for all their customers, including Google.
• They are working with the IETF and other standards organizations to define how crawlers announce their intentions, allowing fine-grained permissions (e.g., humans get free content, robots pay).
• This approach extends the concept of robots.txt but with more granular control and enforcement.
• Cloudflare is adept at identifying and stopping AI companies that try to bypass these blocks, even if they claim to follow rules.

De-emphasizing Traffic as a Proxy for Value (13:05 - 18:04)

• Prince argues that Google is partly responsible for current web issues by making "traffic" the primary proxy for content value.
• This led to content designed to stimulate emotional responses (e.g., Huffington Post, Buzzfeed) rather than advance human knowledge.
• AI models, representing the sum of human knowledge, have "holes in the cheese" (gaps in information) that are highly valuable.
• A new market could reward creators for filling these knowledge gaps, similar to how Spotify identifies unmet musical demand and compensates artists.
• This is different from data labeling; it's about identifying and rewarding the creation of genuinely new or missing information.

Will We Run Out of Quality Human-Generated Content? (18:04 - 20:01)

• Prince believes there will always be human-generated content from new experiments, discoveries, and research.
• He worries that without proper compensation for independent creators, the web could revert to a siloed model where a few large AI companies employ their own teams of journalists and researchers.
• This would undermine the web's historical role as a distributed creator of value and knowledge.

Scaling the Value of Content in the AI Age (20:01 - 22:32)

• The amount paid to content creators needs to scale with the AI companies' business models.
• Flat-fee deals are naive; compensation should be a percentage of subscription fees or ad revenue, allowing creators to share in the upside.
• A market with scarcity will determine the true value of content.
• Advertising and subscriptions will still exist, but content creators should also be compensated for their work being ingested by AI systems.

Cloudflare’s Approach to Inference (22:32 - 24:55)

• Cloudflare has been investing in edge computing for inference, partnering with Nvidia in 2020 to place GPUs at the network's edge, though the market wasn't ready then.
• Four years later, the same initiative has taken off, as the market is now receptive.
• While much inference will happen on devices, larger or more resource-intensive models will run at the network edge.
• Cloudflare is investing in protocols like MCP to secure and facilitate agent-to-agent communication, including payment infrastructure.
• Their solutions for content payment could naturally extend to agent-to-agent payments.
• Cloudflare's fundamental role as a network provider remains crucial as AI and agentic communication grow.

How Cloudflare Responds to Market Demand (24:55 - 26:04)

• Cloudflare's strategy is driven by market demand and customer needs, including those of the 80% of AI companies that use their services.
• They pride themselves on executing and innovating to stay current with trends, positioning themselves at the center of many internet transactions.

Open vs. Closed Models (26:04 - 27:21)

• Cloudflare supports both closed and open models running on their network, but Prince personally favors open models due to Cloudflare's open-source philosophy.
• He is skeptical of arguments that open-source models will lead to catastrophic outcomes, suggesting that regulation should focus on the physical machines that can cause harm (e.g., printing pathogens) rather than the AI models themselves.

Path to the New Marketplace for Content (27:21 - 30:58)

• The immediate next step is to ensure Google plays by the same rules as other AI companies.
• Google's historical dominance has led them to believe they have a right to free content, but their actions (taking content while reducing traffic) are unsustainable.
• Content creators are increasingly considering blocking Google, a previously unthinkable action.
• Other AI companies fear Google's privileged access to content.
• Google needs to differentiate between search indexing and using content for AI overviews or models like Gemini, requiring different compensation models.
• Regulatory and legislative pressure on Google is likely to level the playing field.
• Initially, large content producers (e.g., Condé Nast, New York Times) will make direct deals with large AI model companies.
• Cloudflare can facilitate deals for smaller content providers or for large providers dealing with many AI startups, potentially by negotiating basic rates or managing a capital pool (like Spotify) or micro-payments.
• The first step is forcing Google to adhere to the same rules, which will accelerate the development of the content marketplace.

Advice for Content Creators (30:58 - 32:47)

• First, regain control of your content to create scarcity; Cloudflare and other companies are making this easier.
• Second, recognize that the web's business model is changing, and you can do something about it.
• Third, actively participate in conversations with AI companies to establish fair value exchange.
• Google is already starting pilot projects to pay news providers, indicating a shift in their stance.
• Without compensation or even the "ego hit" of being read, content creation will decline.

Exploring the Timeline for Running Models Locally (32:47 - 40:07)

• Many models are already running locally on devices like phones, especially for applications requiring immediate responses (e.g., driverless cars).
• The biggest limiting factor for local inference is power efficiency.
• Apple has demonstrated power-efficient GPUs/TPUs, contrasting with Nvidia's historical focus on raw power over efficiency.
• Prince recalls Intel's similar resistance to power efficiency in 2011 when Cloudflare needed "cores per watt" for their edge data centers.
• He anticipates a "speedrun" of 30 years of CPU efficiency gains in GPUs over the next 5-10 years.
• Model compression and efficient pruning are key areas for making AI models more efficient.
• The DeepSeek project, despite its origin, showed significant innovation in making training and inference more efficient.
• Cloudflare focuses on making models run efficiently, not on building frontier models.
• The question is who will be the "VMware of AI," creating virtualization and efficiency layers for GPUs, as current GPU allocation is often inefficient and expensive.

The Future of Agentic Infrastructure (40:07 - 44:52)

• Cloudflare expects to compete in the agentic infrastructure space, which will likely involve multiple providers working together.
• Standards for agent communication (e.g., MCP, temporal, LangGraph) are crucial.
• Efficient data collection and payment for data will be essential for training agents.
• The shift towards micro-payments for content could leverage blockchain and cryptocurrency technologies, which were previously viewed skeptically for this purpose.
• Agentic permissioning and identity, potentially using ZK proofs on blockchain, will be vital for controlling how agents, bots, and browsers access and use data.
• Browsers that feed data back to LLMs might face more restricted access.
• The building blocks for this new web architecture have been developing for years, and the convergence of technological maturity and business model needs is now making the future shape of the internet clearer.

Conclusion (44:52 - 45:13)

• The hosts thank Matthew Prince for his insights.
• The discussion concludes with a look towards the emerging shape of the internet, driven by AI, agentic infrastructure, and new content monetization models.

Notable Quotes

"Cloudflare is what the network should have been, what the internet should have been had we known in the '60s, '70s, '80s, how important it was going to be. So how can we make it faster, more reliable, more secure, more efficient, more private, and and that fundamentally is what we're working on every day at at Cloudflare." — Matthew Prince

"If the value creation model of the web has been all about how do I get traffic, the new interface of the web isn't going to send you traffic. And if that's the case, if content creators can't get value from selling a thing or a subscription, selling ads, or just the ego that they get knowing someone is reading their stuff, I worry that people aren't going to create content." — Matthew Prince

"I mean this is this is going to take us down a little bit of a tangent, but I think a lot of the things that are wrong with the world today are ultimately Google's fault." — Matthew Prince

"The first step of creating scarcity is what you have to do in order to figure out what the market is." — Matthew Prince

"I think the first step for content creators is recognize that you have a that the business model of the web is changing. Second, recognize there is something you can do about it, right? You can actually create this scarcity. And then third, actually participate, start to go out and say based on the data, hey, you keep trying to call crawl my stuff. Like let's figure out a way that we can have some fair exchange of value for that." — Matthew Prince