Stewart Squared
Episode #50: Star Hubs, Server Farms, and the Strange New Geography of AI
- AI & Agents
- Money & Sovereignty
- Cloudflare
- AI training
- TPUs
- context window
- Claude
- bots
- Project Honeypot
Listen
Download MP3About this episode
In this episode, Stewart Alsop III talks with Stewart Alsop II about Cloudflare’s role in modern internet infrastructure, from its origins with Project Honeypot to its massive global network powering HTTPS, reverse proxies, DNS integration, and zero-trust systems. The conversation weaves through the evolution of enterprise networking since the Cisco-dominated 1990s, the growth of server farms and AI clusters, the history of dark fiber and undersea cables, and how Web 2.0, social media, crypto mining, and today’s generative AI have shaped bandwidth demand. They explore Cloudflare’s new pay-to-scrape policy, the business dynamics with Google, the rise of high-quality data labeling through companies like Surge AI, and the importance of metadata and privacy in a surveillance-heavy world.
Timestamps
00:00 Cloudflare origins, Project Honeypot, Google TPUs, context windows, Claude, bots paying to scrape
05:00 Early internet infrastructure, Cisco dominance, proprietary enterprise systems, rise of server farms
10:00 Server capacity limits, nanosecond communication, cooling and power issues, AI compute demand
15:00 AI metro areas map, superstar hubs in Silicon Valley, Texas data center project, NVIDIA role
20:00 Dark fiber history, optical components, trench building, undersea cables, global networking
25:00 Web 2.0 growth, social media real-time feeds, crypto mining inefficiency, scaling to AI
30:00 World’s largest data centers, Northern Virginia hub, CIA AWS air-gapped cloud, government secrecy
35:00 Cloudflare market share, AWS, Akamai, content delivery networks, token serving vs video streaming
40:00 Generative AI bandwidth demands, Google search shift, Cloudflare monetizing scraping
45:00 Surge AI and high-quality data labeling, Scale AI critique, metadata importance, privacy concerns
50:00 International capital networks, Middle East investment, Israel’s cybersecurity, Iron Dome, IP issues
Key Insights
- Cloudflare has evolved from its origins in Project Honeypot into a critical piece of internet infrastructure, now integrated into a significant portion of the world’s servers, providing HTTPS, DNS integration, zero-trust frameworks, reverse proxy services, and developer tools like Cloudflare Workers.
- The internet’s physical backbone shifted from proprietary enterprise systems dominated by Cisco in the 1990s to globally distributed server farms. This change was driven by demand for more bandwidth, the use of high-speed fiber connections, and the need to cool and power increasingly compute-heavy systems for applications like AI.
- The concept of “superstar” AI hubs—concentrated in places like Silicon Valley—highlights how certain regions dominate advanced computing due to proximity to key players such as NVIDIA, research talent, and data center infrastructure, with Texas emerging as a new mega-hub.
- The unused “dark fiber” laid during the telecom boom was later bought cheaply and repurposed, enabling the growth of Web 2.0, social media, and streaming. This terrestrial network, along with undersea cables, now underpins global connectivity for modern internet and AI workloads.
- Cloudflare’s new policy requiring payment for web scraping signals a shift in how infrastructure companies may monetize AI-related traffic, especially as large language models consume significant bandwidth to serve tokens in near real time—potentially rivaling video streaming in scale.
- Data quality is a growing competitive differentiator for AI training. Companies like Surge AI claim to outperform “body shop” models like Scale AI by emphasizing high-quality human-in-the-loop labeling, highlighting how metadata and accuracy directly influence model performance.
- The discussion touches on broader geopolitical and security contexts—such as air-gapped government networks, Middle Eastern sovereign wealth investments, Israel’s cybersecurity capabilities, and intellectual property debates—showing how technological infrastructure, policy, and global power dynamics intersect.
Episode transcript
Everyone is wrong about AI, but if you know the history of technology, you can be wrong and still be right.
I know the history of technology since I started with the personal computer in 1980. So listen in while we explore what is right about AI.
Welcome to the Stuart Squared podcast. Today we've got an exciting podcast planned out. Not really planned out because we can't really plan, we're just going to do improv. But Cloudflare did something really interesting, which I don't understand in relationship to AI and AI training. There's some sort of canary in the coal mine I imagine happening right now. And there's a relationship between Google and Cloudflare. Google is a giant in the field. They have all their TPUs who are now going and making sure that I can have a million tokens in my context window, whereas Claude only has a hundred thousand. All these sort and they're just like a gorilla. And there's something that happened with Cloudflare in relation to pay per. The bots now need to pay in order to scrape websites. But I want to understand the history. So I've got Cloudflare pulled up here. The history there's founded in 2009 by Matthew Prince, initially created to track email, spam and cyber threats through Project Honeypot. Okay, I got to go understand what Honeypot is publicly launched in Texas.
Exactly what it sounds like. It's like something that attracts the bees and then traps them.
Your first full transcript is free. After that, an email opens every transcript in the index — a list of readers we can write to, not a guest book.