Track · AI in Product4 min read

Anthropic's $1.4 Trillion IPO Runs on Token Bills From the Top 5%

Anthropic targets a $1.4 trillion valuation as top Claude Code users average $1,500 a month. Four engineering leaders explain token bills from thousands to tens of millions.

Meet 4 engineering leaders whose budgets explain how Anthropic has gotten so rich
Meet 4 engineering leaders whose budgets explain how Anthropic has gotten so richJakub Żerdzicki / unsplash

Context

  1. Anthropic targets a $1.4 trillion valuation in its IPO this week; top 5% of Claude Code users average ~$1,500 a month vs under $75 for non-users.

  2. Anthropic has 11,900 business customers that spent $100,000+ in the past year and 250 that spent $10 million or more.

  3. Grafana Labs spends tens of millions a year on tokens, about two-thirds on Claude.

  4. Harness's 500+ engineers used 265 billion tokens in a recent week under a soft $125-a-week budget.

  5. Theory Ventures' Bryan Bischof personally hit six figures of token spend in one month.

Anthropic is targeting a $1.4 trillion valuation in its initial public offering this week, and the prospectus shows exactly who is paying for it: business customers who didn't use Claude Code averaged under $75 a month, while the top 5% of Claude Code users averaged around $1,500 a month. The company counted 11,900 business customers that spent $100,000 or more over the past year, and 250 that spent at least $10 million.

Months after the "token-maxxing" craze peaked, engineering leaders say the bills haven't slowed. Some expect individual engineers to occasionally spend more on tokens — the units of AI output — than they earn in salary. Four of them walked Business Insider through the numbers, and their budgets explain how Claude Code became the juggernaut behind Anthropic's pre-IPO growth.

Grafana Labs: tens of millions a year, and no budget caps

Tom Wilkie, chief technology officer at Grafana Labs — the observability startup last valued at $9 billion — says token bills have jumped into the tens-of-millions-a-year range. About two-thirds goes to Claude, with the rest mostly to OpenAI's Codex.

"We're not worried about the spend," Wilkie said. "We're not worried about controlling it, just yet."

The spend pays for itself in specific cases. One team burned tens of thousands of dollars in tokens in a single month to improve the performance of one service — and cut a six-figure chunk from the company's Amazon Web Services bill. A non-engineer routed customer feedback through Anthropic's Fable model, spent more than $30,000 on tokens, and landed a cluster of successful product changes.

There have been accidents. A data analyst left a model running and racked up tens of thousands of dollars. Grafana's fix: grant access to the most powerful models in 24-hour increments — but otherwise "no questions asked." Wilkie's reasoning is blunt: "We really want to be careful with the chilling effect of budgets and access control to models because you just don't know where the next big piece of innovation is going to come from."

The takeaway: Grafana treats AI spend like infrastructure, not a cost center to ration.

Theory Ventures: six figures in a month, personally

Bryan Bischof, head of AI at Theory Ventures, leads six engineers building custom software for investors and intelligence leads at the San Francisco venture firm. He has personally hit six figures of token spend in one month, mostly on Claude, and says he could easily crack $1 million if he spent more time running simulations.

His team sometimes uses open-weight models, which cost far less than Anthropic's top-end options. But because engineer time is expensive, Bischof often prefers premium tools. "If a very intelligent model is going to dramatically reduce the cycle time, then that's absolutely worth it," he said. The firm is also considering investing in its own always-running AI agents, similar to Meta's Muse and OpenAI's Dots.

Menos AI: $4,000 a month beats a hire

Chris Yang, cofounder and chief AI officer of Menos AI, runs AI-powered research software for investors. His team uses Claude Code or Claude models through Cursor. "Because Claude has been evolving so, so well in terms of coding capabilities, I don't even have much opportunity or willingness to try out Codex," he said.

The investor-facing software consumes around 90% of the company's token costs, routing user inquiries to different models by difficulty. Yang calls himself one of the company's "hero users" — he spends $3,000 to $4,000 a month running several AI agents at once. "We do spend a lot of expenses on the Claude Code use, but when we compare it to a human resources hire, it's still a big benefit," he said. "It's still much cheaper than human labor."

Harness: 265 billion tokens a week, with a soft cap

Harness, a software development startup worth $5.5 billion with more than 500 engineers, used 265 billion AI tokens in a recent week, according to Kiran Gangadharappa, senior director of software engineering. Claude Code and Cursor are the most popular tools among his staff.

"We definitely see it improving and taking a bigger and bigger chunk of the manual work of writing code," he said. The biggest token-guzzlers are gigantic one-off tasks: rebuilding an existing software architecture or conducting security research. Harness cut spending by using Sonnet, one of Anthropic's cheaper models, and by setting a soft $125-a-week budget for engineers. He expects AI expenses have reached a "steady" level.

His closer works as a whole-industry summary: "Agents never get tired."

The pattern across all four is the same tradeoff: uncapped spend buys velocity and occasional six-figure savings, while caps risk missing the next big win — and as token bills become a permanent line item next to salaries, product teams will keep being asked to prove which one the money should go to.

via feeds.businessinsider.com (Source)

More from Daniel Okafor

Daniel Okafor

Show full bio

Staff writer covering technology and product news in product management at Roadmap File.

60 articles