Y Combinator CEO Garry Tan puts his AI token bill at $100,000 to $200,000 a year
In brief
- Garry Tan told Isaacson he spends about $100,000 to $200,000 a year on AI tokens.
- Tan compared the cost to roughly a senior engineer's salary, Crypto Briefing reported.
- "Tokenmaxxing" is Tan's name for the heavy-usage approach; he shared 23 Claude Code skills as gstack.
- Garry's List took Tan about five days and roughly $200 in tokens to build, he said.
- Crypto Briefing flagged Tan's reliance on a single model provider, Anthropic's Claude Code.
What Tan calls "tokenmaxxing"
Tan has a name for it.
He calls it tokenmaxxing, and the idea is that pouring resources into AI usage lets a small team (or even one person) produce output that would normally take a much larger organization. Crypto Briefing reported that he made it the focus of an August Startup School keynote.
His number moves around. According to Crypto Briefing's summary of his 2026 interviews, Tan has put his annual spending on AI workflows anywhere from $50,000 to $200,000, and his latest estimate is at the top of that range.
In a conversation with Isaacson, Tan said he spends “about maybe $100[,000]-$200,000 a year on tokens.”
He uses agents for a wide range of work projects and coding tasks. He's joked that he's suffered from "cyber psychosis," and he's said he sometimes barely sleeps because working with his agents excites him that much.
gstack and Garry's List
Tan shared 23 Claude Code skills, bundled as gstack, on GitHub. They're role-based skills meant to simulate an engineering team running inside Claude Code, and Tan has described the release as an effort to teach others what agents can do. Crypto Briefing put the repository at between 100,000 and 136,000 GitHub stars.
His most concrete example of the payoff is Garry's List, a refreshed blogging platform. Tan said he used Claude Code agents to build it in about five days for approximately $200 in tokens.
That's $200, set against an annual token bill he pegs at up to $200,000.
He's also been on the podcast circuit in 2026 (including a16z and YC shows), where he's argued that AI-driven workflows can produce hundreds of thousands of lines of code.
The single-provider question
Crypto Briefing pointed to a catch. Tan's public toolkit is built around Anthropic's Claude Code, so in the outlet's words, "a lot of eggs sit in one basket." It wrote that a company whose productivity depends on heavy AI usage is exposed to the pricing, availability and policies of its model provider.
Frequently asked questions
What is tokenmaxxing?
Tokenmaxxing is Garry Tan's name for pouring resources into AI usage. The idea is that heavy AI usage lets a small team, or even a single person, produce output that would normally take a much larger organization. Crypto Briefing reported that Tan made it the focus of an August Startup School keynote.
What are the AI tokens Garry Tan is paying for?
They're usage credits for AI models, the units developers are billed in when they run large language models through tools like Anthropic's Claude Code. Tan told Isaacson he spends about $100,000 to $200,000 a year on them, which he put at roughly a senior engineer's salary.
What is gstack?
gstack is a set of 23 Claude Code skills that Tan shared on GitHub. It offers role-based skills meant to simulate an engineering team running inside Claude Code. Tan described the release as an effort to teach others what agents can do, and Crypto Briefing put its star count at between 100,000 and 136,000.
