How Claude Code Made Vercel's Founder a Billionaire

Key insights

• Vercel's moat isn't the technology, it's a side effect: Next.js is open source and over-represented in the training data, so LLMs default to building Next.js apps that land on Vercel
• Claude Code punches far above its weight. Its users are 1 percent of Vercel's base but generate 15 percent of deployments, and around 70 percent of all AI-agent deploys on Vercel come from Claude Code
• The SaaS collapse that wiped billions off software stocks coincided with Vercel nearly tripling its valuation. Opposite outcomes from the same AI wave
• The forces enriching Vercel could also disrupt it. A Vercel investor admits it's easier to build a Vercel competitor now than it used to be

In Brief

Guillermo Rauch is the CEO of Vercel, a San Francisco company that hosts web apps and AI agents. In January 2026, shortly after Anthropic released Opus 4.5, he stood at Vercel's headquarters and gave his staff a short history of AI coding tools. GitHub Copilot in 2021 "could barely complete code." ChatGPT a year later was "a killer use case." Claude Sonnet 3.5 in 2024 could be "trusted with smaller bits of code." Then came Opus 4.5, and Rauch told them this was going to be a big moment for the world.

A month later, Opus 4.6 hit. It triggered what Forbes calls the so-called SaaS apocalypse, wiping billions off software stocks as investors worried those firms could be automated away. But the collapse that gutted SaaS was making Vercel rich. The company hosts web apps for brands like Under Armour, Stripe, and Sonos, and it also happens to be where Claude Code deploys by default. In September 2025, Vercel raised $300 million at a $9.3 billion valuation, nearly triple what it was a year earlier. Rauch, an Argentine immigrant, is now worth at least $2.1 billion.

The deeper story isn't that AI is eating software. It's that someone has to host the code AI generates, and Vercel has quietly become the default landing strip for vibe-coded apps. According to Forbes reporting, Vercel customers who use Claude are about 1 percent of users but generate almost 15 percent of all deployments. Around 70 percent of AI-agent deployments on Vercel come from Claude Code alone.

The core claim: someone has to host the code

Forbes frames Vercel's position as a classic "Pix and Shovel" story. That's a reference to the California Gold Rush of the 1840s, when the people who reliably got rich weren't the miners. They were the merchants selling picks, shovels, and supplies.

Rauch's own framing lands close to the same idea, as he tells Forbes:

"We've seen a tremendous acceleration on deployments. Fundamentally, we want to become the infrastructure layer of this new generation of software."

Translation: Vercel doesn't need to build the AI. It just needs to be where the AI's output lands. Every time a developer, or an AI agent, generates code and ships it to the web, Vercel collects hosting fees. It's a business that gets paid in throughput, not in model breakthroughs.

The numbers behind the boom

Vercel's traction with AI-generated code isn't a vague narrative. It shows up in the metrics.

Two numbers stand out. The first: Claude users are 1 percent of Vercel's base but account for 15 percent of deployments. That's roughly a fifteenfold overrepresentation. The second: when you zoom out to all AI-agent deployments on Vercel, Claude Code alone drives around 70 percent of them.

In other words: Vercel's growth curve rides on Anthropic's growth curve. If Claude Code keeps shipping, Vercel keeps growing. If Claude Code stalls, Vercel has a problem.

Why LLMs "love" Vercel

The mechanism behind the numbers is mundane. Next.js, the most popular React framework for building websites, is built and maintained by Vercel, and it's open source.

Open source matters because that's where large language models learn. When Anthropic, OpenAI, and Google train models on public code, they ingest billions of lines of Next.js. Models become very good at writing Next.js by default. When you then ask Claude Code to build you a web app, it tends to produce a Next.js project.

And the natural place to deploy a Next.js project is Vercel.

Dan Levine, an Accel partner and early Vercel investor, puts it bluntly:

"LLMs seem to love Vercel and we love them back."

The moat isn't technology. It's training-data distribution. Vercel didn't engineer this advantage by inventing a clever AI product. It inherited it by giving away a popular framework years before LLMs started writing code.

The pick-and-shovel play

This isn't the first time someone has gotten rich from a gold rush without ever picking up a pan. In 1849, most prospectors in California went home empty-handed. The people who reliably made money were the outfitters. Samuel Brannan, often called California's first millionaire, bought up every pickaxe and shovel in the San Francisco Bay Area and resold them to miners at a steep markup. The miners took the risk. The suppliers took the revenue.

Rauch's arc fits the pattern. He grew up in Lanús, just south of Buenos Aires, taught himself English as a kid so he could read open-source documentation, and became a core contributor to MooTools, an early JavaScript library, while still a teenager. Facebook tried to hire him, then backed out when they learned he was seventeen. He later moved to San Francisco, sold a file-sharing startup called Cloudup to Automattic, and founded Zeit: the hosting company that eventually became Vercel.

His customers today include the companies building on top of AI, not the AI labs themselves. Last month, Notion launched Notion Workers, a platform for building AI agents, hosted on Vercel. Notion CEO Ivan Zhao called Vercel "the best choice," noting it is "one of the fastest at adapting tools for both human developers and AI agents."

The risk: Vercel is also software

Here's the uncomfortable part. Vercel isn't outside the SaaS apocalypse. It is a software company. It faces the same AI that wiped out its customers' competitors. Cloudflare, Supabase, Netlify, Render, and Fly.io are all chasing the same market.

Dan Levine, the same investor cheerleading Vercel's growth, doesn't dodge it:

"It's easier to build a Vercel competitor than it was before."

The tools that let Vercel help its customers move fast also let someone else build a cheaper Vercel. If a smart team points Claude Code at "build me a Vercel clone with half the margin," they won't finish in a weekend, but the path is shorter than it used to be.

Rauch's defense is that most people don't want to vibe code infrastructure. Payments, data pipelines, and mission-critical systems still need human engineers being careful. He's probably right for now. The moat is less "we built something hard to copy" and more "people aren't ready to trust AI with the foundational layer yet."

The real shift

Zoom out and the pattern looks like this.

Before the AI coding boom:

1. Developers wrote software, sold it as SaaS, and collected subscriptions
2. Vercel made money hosting the occasional Next.js site for bigger brands
3. The action, and the margin, sat in the SaaS layer

After:

1. AI agents generate volumes of code that no human team could match
2. That code has to be hosted somewhere, and LLMs default to Next.js and therefore Vercel
3. The action has moved down the stack, into infrastructure

The companies that looked boring in 2023 (hosts, databases, frameworks) are where the margin sits in 2026. The SaaS layer is getting squeezed by AI on both sides: easier to build, easier to replace, and in some cases automated entirely by agents.

Vercel's bet is that this is a permanent shift, not a bubble. If Rauch is right, infrastructure is the most valuable real estate in software. If Levine is right that his own portfolio company is vulnerable, it's still a software company, and the next Vercel is sitting in a Claude Code window somewhere right now.

Glossary

Sources and resources