Anthropic’s $1.4 Trillion IPO Target Rests on Claude Code Bills
Anthropic is targeting a $1.4 trillion IPO valuation as Claude Code-heavy customers drive bills into the millions, exposing the gap between managed AI budgets and uncontrolled token use.
By Olivia Hart
6 min read

What's News
- Anthropic is targeting a $1.4 trillion valuation in its IPO this week.
- Anthropic reported 250 business customers with $10 million or more in spending.
- Anthropic counted 11,900 business customers that spent at least $100,000 over the previous year.
- The top 5% of coding-tool users averaged about $1,500 per month, while business-account customers not using Claude Code averaged less than $75.
- Harness engineers used 265 billion AI tokens in a recent week.
By Olivia Hart
Anthropic said in its IPO prospectus that 250 business customers had spent $10 million or more as the company targeted a $1.4 trillion valuation for its initial public offering this week. Anthropic also counted 11,900 business customers that had spent at least $100,000 over the previous year, Business Insider reported.
Business-account customers that did not use Claude Code spent less than $75 a month on average, Anthropic said. The top 5% of coding-tool users averaged about $1,500 monthly, or roughly $18,000 a year at that rate.
Business Insider identified Claude Code as the product driving Anthropic’s pre-IPO growth. The outlet also reported that some engineering leaders still accumulated tens of thousands of dollars in Claude bills on individual projects months after the “token-maxxing” craze, while some engineers spent more on tokens than they made in salary.
Those figures establish heavy enterprise demand, but not its durability. Business Insider’s reporting does not disclose customer retention, gross margins, or revenue concentration among the largest spenders. Claude Code’s economics depend heavily on a small group of power users.
How much of Anthropic’s growth depends on Claude Code?
The prospectus data shows a sharply uneven spending pattern. Anthropic’s thousands of business customers and the popularity of Claude Code are funneling large sums into the company, according to Business Insider.
That revenue remains vulnerable to unusual usage patterns. The source does not show whether smaller customers will increase spending or whether the heaviest users will maintain their current budgets after the IPO.
How much does Grafana Labs gain from heavy model use?
Grafana Labs, an observability startup last valued at $9 billion, spends tens of millions a year on tokens, Chief Technology Officer Tom Wilkie told Business Insider. About two-thirds goes to Claude; the rest mostly goes to OpenAI’s Codex.
One team spent tens of thousands of dollars in one month to improve a Grafana service, producing a six-figure cut to the company’s Amazon Web Services bill. Wilkie supplied no gross spend or savings duration, so the experiment’s return remains unquantified.
Wilkie told Business Insider, “We're not worried about the spend.” He said he was not yet worried about controlling it. A non-engineer spent more than $30,000 routing customer feedback to Anthropic’s Fable model and got a cluster of successful changes into the product, Wilkie said.
Engineers have become more comfortable switching projects because the coding-language barrier feels lower, Wilkie said. After a data analyst left a model running and generated tens of thousands of dollars in spend, Grafana began granting access to its most powerful AI models in 24-hour increments.
Wilkie said the policy remained open-ended because tighter budgets and access controls could chill the next breakthrough. Grafana treats model access as an R&D investment, but its open-ended approach leaves cloud and vendor costs exposed to mistakes.
Why does Theory Ventures pay for premium models?
Bryan Bischof, Head of AI at Theory Ventures, leads six engineers who build custom software for investors, heads of talent and relationships, and intelligence leads at the San Francisco venture firm, Business Insider reported. He personally accumulated six figures, or at least $100,000, in token spend in one month, mostly on Claude.
Bischof said he could easily exceed $1 million if he spent more time running simulations. That figure marks his spending capacity rather than a disclosed bill, and his team also uses cheaper open-weight models.
Despite those lower-cost alternatives, Bischof often favors premium tools because engineers’ time is valuable. He said: “If a very intelligent model is going to dramatically reduce the cycle time, then that's absolutely worth it.”
Theory Ventures may also invest in its own always-running AI agents, similar to Meta’s Muse and OpenAI’s Dots, Bischof said. Such agents could extend current simulation work into continuous spending, and he disclosed no budget ceiling. Theory Ventures treats expensive models as labor-saving infrastructure, but its own agents could make the bill more persistent.
How does Menos AI turn Claude into a labor substitute?
Menos AI runs AI-powered research software for investors. Cofounder and Chief AI Officer Chris Yang told Business Insider that the engineering team uses Claude Code or Claude models through the coding tool Cursor.
The investor software consumes around 90% of Menos AI’s token costs, Yang said, because it routes user inquiries to different models depending on difficulty. Yang described himself as a hero user who spends $3,000 to $4,000 on Claude tokens in a month while running several AI agents at once.
That pace equals $36,000 to $48,000 a year, based on Yang’s monthly spending. He said he had little opportunity or willingness to try Codex because Claude’s coding capabilities had evolved so well.
Comparing the Claude Code expense with a human-resources hire, Yang said it remained much cheaper than human labor. The comparison may hold for Menos AI’s workload, but Business Insider supplied no headcount or revenue benchmark, and heavy Claude reliance leaves little room to test alternatives. Menos AI has a measurable productivity case for Claude, but concentration risk comes with the savings.
Can Harness control 265 billion tokens in a week?
Harness, a software-development startup worth $5.5 billion, has more than 500 engineers. Senior Director of Software Engineering Kiran Gangadharappa told Business Insider that his staff used 265 billion AI tokens in a recent week.
Engineers used Claude Code and Cursor most, Gangadharappa said. He also said AI was taking a growing share of manual code-writing work.
The largest token consumers are enormous one-off tasks such as rebuilding software architecture or conducting security research, Gangadharappa said. Using Sonnet, one of Anthropic’s cheaper models, helped reduce spending, and Harness created a soft $125-a-week budget for engineers, equal to about $6,500 a year.
Gangadharappa expects AI expenses to have reached a steady level and said agents never tire. At 265 billion tokens in one week, however, a soft budget does not establish a hard ceiling on companywide use. Harness has reduced routine spending with cheaper models and soft limits, but its weekly volume remains immense.
What will Anthropic’s IPO reveal?
In Business Insider’s interviews, leaders at four companies treated tokens as a variable labor input when speed or always-on agents could replace manual work. Their examples support Anthropic’s pre-IPO growth story, but the interviews do not independently measure output, savings or profitability.
Anthropic’s IPO this week will put the demand shown in its prospectus under market scrutiny, especially whether spending extends beyond unusually compute-intensive users. The $1.4 trillion valuation remains a target rather than a completed valuation, and the reporting leaves customer retention and revenue concentration unresolved.
Source: feeds.businessinsider.com
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,378 articles