skip to content
Meir Dick

My Cloudflare Thesis: The 100,000x Code Explosion

/ 5 min read

View more blogs with the tag thesis , View more blogs with the tag investing , View more blogs with the tag ai
Table of Contents

Here’s my thesis on Cloudflare (NET). It’s a convex bet — the kind where if I’m right, the upside is asymmetric.

Peter Thiel talks about the difference between 0 to 1 (creating something genuinely new) and 1 to n (copying what works). Most people are thinking about AI as a 1-to-n story: developers get more productive, we write more code, things scale linearly.

That’s not what’s happening. This is a 0-to-1 moment — a phase change in what generates code.

The Internet Traffic Lesson: Why Projections Failed

In 1992, the internet had about 10 million users — roughly 0.2% of the global population. Traffic was about 4,000 GB per month globally.

If you were projecting forward, the math seemed straightforward: internet penetration would grow from 0.2% to maybe 60-70% of the world. That’s a 300-400x increase in users. So traffic should grow by roughly 300-400x, right?

That would put 2025 traffic at around 1.5 petabytes per month.

The actual number? 409,000,000 petabytes per month (409 exabytes).

That’s not 400x. That’s 100 million times the 1992 level. The projections were off by a factor of 250,000x.

Here’s the data (from Cisco’s historical records):

YearTraffic (per month)Users
19924,444 GB~10M
1995150 TB~16M
200075 PB~360M
201442 EB~3B
2025409 EB~5.5B

What happened?

The projections assumed traffic would scale with human users doing human things — sending emails, browsing websites, downloading files.

They couldn’t conceive of:

  1. Machine-to-machine traffic — APIs calling APIs, servers syncing, IoT telemetry
  2. Video everywhere — streaming, conferencing, surveillance (a single 4K stream = thousands of emails)
  3. Always-on connectivity — from dial-up hours to 24/7 persistent connections
  4. Multiple devices per person — phones, tablets, laptops, smart TVs, watches
  5. Cloud everything — continuous backup, sync, updates, monitoring

The people projecting weren’t wrong about user adoption. They were wrong about the nature of what would generate traffic. Most traffic today isn’t humans browsing. It’s computers talking to computers.

The Same Pattern Is Emerging With Code

Right now, people are projecting code volume growth based on developer productivity gains. “AI will help developers write 2x more code.” Maybe 5x. Maybe 10x.

They’re thinking too small. They’re making the same mistake.

The numbers that matter:

  • GitHub Copilot: 27% of code at launch (2022) → 46% of all code written by active users (2025)
  • Java developers: 61% of their code is now AI-generated
  • Developers complete tasks 55% faster with AI assistance
  • I wrote 15,000 lines of code in 24 hours — one person, using AI agents

But here’s what the projections miss:

Just like internet traffic exploded because machines started generating traffic (not just humans), code volume will explode because machines will start generating code at machine scale.

We’re not going from “developers write X lines” to “developers write 5X lines.”

We’re going from “humans write code” to “AI agents write code continuously, autonomously, at scale.”

Every AI agent running. Every automated pipeline. Every system that can spin up, test, iterate, and deploy without human intervention.

The amount of code being written isn’t going to 10x. It’s going to 100,000x.

Here’s why the timeline is compressed: AI adoption is 4x faster than internet growth.

The internet followed Moore’s Law — doubling roughly every 2 years. That meant 100,000x growth took about 34 years (17 doublings × 2 years).

AI capabilities are doubling every 6 months. Same 17 doublings, 4x the speed.

Growth RateDoubling PeriodTime to 100,000x
Internet (Moore’s Law)2 years~34 years
AI Adoption6 months~8.5 years

By 2035, the volume of functional code produced will dwarf everything written in human history combined. Most of that code will be written by machines, for machines, deployed automatically.

We can’t even conceive yet how much code is going to be written.

Code Needs Infrastructure

Here’s the part everyone knows but few internalize: code in a repository is useless.

For code to do anything, it has to:

  • Deploy to infrastructure
  • Run on servers
  • Be served to users and systems
  • Be tested in live environments
  • Interact with the internet

The LLM writes the code. But the code has to live and run somewhere.

If code volume 100,000x’s, infrastructure demand 100,000x’s.

Cloudflare’s Position: The Connectivity Cloud

In their 2023 Founders’ Letter, Matthew Prince and Michelle Zatlyn articulated exactly where Cloudflare is positioning:

We are a Connectivity Cloud… Our job isn’t to be the final destination for your data, but to help it move and flow.

They contrast this with “Captivity Clouds” (the hyperscalers who want to lock your data in). Cloudflare’s vision: be the layer where everything connects — at the edge.

Their AI thesis:

The biggest opportunity in AI is inference.

Inference happens continuously — every query, every request. And it needs to happen at the edge, not in centralized data centers:

  1. On-device = Too small (limited storage, battery, expensive)
  2. Centralized cloud = Too far (latency, privacy, cost)
  3. Edge (Cloudflare) = “Not too small. Not too far. The perfect step in between.”

They launched Workers AI, Vectorize, and AI Gateway. By end of 2023: GPUs in 100+ cities, 40+ countries.

We estimate that across the AI startup ecosystem, Cloudflare is the most commonly used cloud provider.

The Bet

Connect the dots:

  1. Internet traffic didn’t 400x like user growth suggested — it 100,000,000x’d because machines started generating traffic
  2. Code volume won’t 10x like productivity gains suggest — it will 100,000x because machines will generate code
  3. That code needs to run somewhere — infrastructure demand scales with code volume
  4. Cloudflare is positioned at the edge — where AI code deploys, runs, and serves
  5. They’re already the AI startup infrastructure layer — Workers, R2, D1, Pages

The big cloud providers (AWS, Azure, GCP) will benefit. That’s the obvious play.

But Cloudflare is positioned for something more specific: to be the edge layer where the code explosion actually runs.

The Math

If code volume 100,000x’s — not over decades, but over the next 8-10 years — and Cloudflare captures a meaningful share of where that code lives…

The business doesn’t grow linearly. It grows with the explosion.

That’s a convex bet. The kind I like to make.

Not financial advice. A thesis about where the world is going. I could be wrong. But if I’m right, the upside is asymmetric — exactly the kind of bet worth making.

Sources: