Brief IA
›

Anthropic Aims for $1.4 Trillion Valuation in the Stock Market

Code & Dev·Tom Levy·

Anthropic Aims for $1.4 Trillion Valuation in the Stock Market

Anthropic Aims for $1.4 Trillion Valuation in the Stock Market
Key Takeaways
1Anthropic aims for a valuation of $1.4 trillion for its IPO this week.
2Harness has established a flexible budget of $125 per week for its engineers.
3Anthropic reports having 250 commercial clients who have spent at least $10 million.
Why it matters : Claude Code has generated billions of dollars in revenue for Anthropic and contributed to its growth ahead of the IPO.

Anthropic aims for a valuation of $1.4 trillion for its upcoming IPO this week. Among its clients, technical leaders report spending tens of thousands of dollars on certain projects using Claude. Their practices range from flexible weekly budgets to temporary access to the most powerful models.

Harness sets a budget of $125

Valued at $5.5 billion, the software development startup Harness employs over 500 engineers. Among the tools they favor are Claude Code and Cursor. Kiran Gangadharappa, senior director of software engineering, reports that staff consumed 265 billion tokens in a recent week. According to him, one-off large-scale operations, such as redesigning existing architecture or conducting security research, are the most resource-intensive.

The director attributes a reduction in costs to the choice of Sonnet, one of the cheaper models from Anthropic, as well as a flexible budget of $125 per week for engineers. He now anticipates a stabilization of AI spending.

Grafana Labs has chosen a different restriction after a data analyst accidentally ran a model and incurred a bill of several tens of thousands of dollars. Its CTO, Tom Wilkie, indicates that the most powerful models are now accessible for 24-hour periods, without justification required. He aims to prevent budgets or access restrictions from discouraging initiatives whose origins cannot be predicted.

From token bills to expected savings

Specializing in observability and recently valued at $9 billion, Grafana Labs spends tens of millions of dollars annually on tokens, according to Tom Wilkie. About two-thirds go to Claude; the remainder primarily goes to Codex from OpenAI. The CTO claims he is not alarmed, at this stage, by the amounts spent or their management. He cites a team that consumed several tens of thousands of dollars in a month to improve the performance of a service, which he believes resulted in a six-figure savings on the Amazon Web Services bill.

Bryan Bischof, head of AI at Theory Ventures, a San Francisco venture capital firm, oversees six engineers developing custom software for investors, talent managers, and intelligence executives. He reports having spent a six-figure sum in a month, primarily on Claude. He estimates he could easily exceed $1 million if he dedicated more time to simulations. His team also uses open-weight models, which are much less expensive than Anthropic's high-end offerings, but he believes the higher expenditure is justified when it significantly reduces turnaround times.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Chris Yang, co-founder and head of AI at Menos AI, which develops research software for investors, reasons by comparison with human costs. He describes himself as a particularly intensive user and states he spends between $3,000 and $4,000 monthly on Claude tokens, with several agents active simultaneously. Despite these amounts, he considers Claude Code to be much less expensive than human labor. Some executives interviewed even suggest that the token consumption of certain engineers sometimes exceeds their salary.

From code to customer feedback, the uses of Claude

These companies have adopted AI for programming and show a clear preference for Claude. At Grafana Labs, Tom Wilkie reports that a non-engineering employee submitted customer feedback to Anthropic's Fable model, incurring a cost exceeding $30,000. He attributes successful product modifications to this approach. He also observes that engineers are more willing to switch from one project to another, finding the differences between languages less daunting.

At Menos AI, developers use Claude Code or access Claude models through the programming tool Cursor, explains Chris Yang. The advancements of Claude in coding give him, he says, few opportunities or desire to try Codex. However, about 90% of the company's token costs come from its investment software: this directs user requests to different models based on their difficulty.

At Harness, Kiran Gangadharappa notes that tools are improving and are taking on an increasing proportion of the code writing that was previously done manually. Theory Ventures, for its part, may invest in continuously operating agents, comparable to Muse from Meta or Dots from OpenAI, indicates Bryan Bischof.

Anthropic has 250 clients with massive spending

Anthropic claims to have 250 commercial clients who have spent at least $10 million, as well as 11,900 who dedicated $100,000 to its services in the previous year. Claude Code has generated billions of dollars and contributed to its growth ahead of the IPO, receiving particularly enthusiastic reception among tech startups.

Anthropic's IPO prospectus distinguishes usage patterns: clients without Claude Code on their business account spent an average of less than $75 monthly. At the other end, the top 5% of users of the programming tool reached an average of about $1,500 per month.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.