Cloudflare is described as an internet infrastructure company focused on security, performance, reliability, and privacy, rather than just a CDN.
The company was launched in September 2010 and has grown to a $66 billion market cap with around $1.8 billion in trailing revenue.
Cloudflare's approach evolved by solving issues as "customer zero," growing a suite of security and efficiency products in response to real-world challenges like attacks and scaling.
The dominant value creation model for the web over the past 30 years was driven by search, supporting commerce, ads, and creators seeking fame.
The web is now shifting toward AI-driven interfaces, resulting in fewer clicks to original content sources due to AI overviews and answer boxes.
For example, content gets 10 times fewer clicks from Google than it did a decade ago; with OpenAI it’s 750 times harder, and with Anthropic 30,000 times harder to get a click.
This decline jeopardizes incentives for content creation, threatening both the web’s vitality and the content supply for AI models.
Business Models and Compensation for Content Creators 06:35
Agent-based commerce (where AI agents act on behalf of users) may be easier to monetize, but compensating content creators remains challenging.
Copyright law offers limited protection; derivative AI outputs are often considered "fair use," making legal solutions unreliable for creators.
Cloudflare observes that most major AI companies agree in principle that they should pay for content, but struggle with ensuring a level playing field, particularly if competitors can access content for free.
Cloudflare believes a functioning content economy requires artificial scarcity since web content has traditionally been available for free.
On July 1st, Cloudflare began blocking AI training crawlers across all customer sites by default, applying the same rules to Google and others, to help create scarcity.
The aim is to use standards (with organizations like IETF) to allow content creators to specify that humans can access content for free, but AI bots must pay.
Advancing from the simple robots.txt, finer-grained permissions and enforcement are needed—Cloudflare is developing tools to detect and block unauthorized scraping.
Some AI companies bend or break crawling rules, and Cloudflare intends to publicly call out these behaviors.
Rethinking Content Value and Filling Knowledge Gaps 13:12
Search-era content rewards maximizing clicks and engagement, leading to repetitive and sensational content.
An ideal AI-driven content market would reward originality and filling knowledge gaps, not just generating traffic.
Companies could identify unmet content needs (as Spotify does by revealing search queries with no results to musicians), incentivizing creation in less-covered areas.
Compensation Models and the Need for Market Evolution 18:06
There is ongoing debate about the magnitude and method of compensating creators, as prior deals with AI companies often don't scale with long-term model value.
A scalable compensation model should link payments to AI-generated revenue or subscription fees, allowing content creators to share in upside as AI usage grows.
Most present deals are between large publishers and major AI companies; a true content marketplace could allow smaller creators and startups to participate via standardized rates or pooled capital distribution.
Evolving Architecture and Inference Efficiency 22:33
Cloudflare anticipates a shift from serving web pages to serving model inference results, with increasing importance of edge computing.
The company has invested in placing GPUs at the edge for AI inference, predicting that some models will always be too large for user devices and will need to run on the network edge.
Efficiency, particularly in terms of power use for inference, is a major focus. Cloudflare seeks to optimize "tensor units per watt" and believes significant progress in model and compute efficiency is still forthcoming.
Unlike hyperscalers, Cloudflare incentivizes efficient use because it charges customers by work done, not for reserved capacity.
Cloudflare supports both closed and open-source AI models but has a preference for open source and collaborates with companies like Meta.
The company is skeptical of fears that open-source models will lead to catastrophic outcomes, arguing security risks should be managed at the physical level (e.g., restricting what machines can do) rather than by banning open models.
The emergence of standards for AI agent communication and marketplaces will be crucial, with Cloudflare positioned at the network chokepoint for enforcement and innovation.
For a functional marketplace, Google must lose its privileged access and pay for content when using it for AI overviews or answer boxes.
Regulatory and legislative pressure may push Google to comply; once this happens, broader market mechanisms for content compensation can emerge.
Deals will initially focus on transactions between major publishers and AI companies, but there’s a need for platforms (potentially run by Cloudflare) to handle many-to-many payments and micropayments for both large and long-tail creators.
Content creators should implement mechanisms to control access to their content, creating scarcity rather than giving it away for free.
Engage AI companies and seek fair value exchange as Google and others begin to pilot payments for news providers.
Recognize that traditional web business models (ads, subscriptions, traffic) are fading; active participation in shaping new compensation models is needed.
Future of AI Inference, Devices, and Infrastructure 32:48
Inference is increasingly moving to user devices where feasible, especially for latency-sensitive applications (e.g., autonomous vehicles).
Power efficiency and model size are key technical challenges for broader on-device AI deployment.
Progress in model pruning, compression, and smarter allocation of GPU resources is expected to accelerate, with the eventual goal of running advanced models (like current ChatGPT versions) entirely on consumer devices.
Blockchain and micropayments could play a vital role if future web economics depend on per-use compensation, though current blockchains may not yet handle required volumes.
Agentic permissioning and digital identity systems (possibly with zero-knowledge proofs) will be important to regulate AI agents and browsers accessing content, with cryptographic guarantees providing compliance and privacy.