Anthropic IPO filing reveals massive computing costs and high revenue per employee

A high-margin business model

Anthropic's recent filing provides a detailed look at the company's financial health as it prepares for a potential public offering. One of the most striking figures is the company's revenue per employee, which reached $6.6 million by the end of the third quarter. This level of efficiency is notable because it is only slightly lower than Nvidia, a company often cited as a gold standard for operational efficiency in the tech sector.

The company has also demonstrated significant financial discipline in how it manages its capital. As of June 30, Anthropic held more than $134 billion in cash and short-term investments. This amount is only about $10 billion less than the total amount the company has raised to date, suggesting that it has been careful with its spending despite the high costs of AI development.

The filing also highlights a strong customer base that validates the demand for the company's products. Anthropic has nearly 2,000 customers who generate at least $1 million in revenue, and 250 of those customers generate over $10 million. Furthermore, half of the Fortune 500 companies pay the firm $100,000 annually, with those clients spending less than 0.1% of their total third-quarter expenses on Anthropic's services.

The weight of computing costs

While the revenue figures are impressive, the filing also reveals the staggering costs required to power Anthropic's models. The company owes nearly $650 billion for computing, which includes non-cancellable contracts and minimum payments. These obligations mean the company must pay for massive amounts of computing power regardless of how much capacity its customers actually use.

This creates a unique risk profile for the company's future valuation. Investors must decide if customers will continue to spend at a rate that keeps pace with these massive infrastructure commitments. If demand slows down or if cheaper alternatives emerge, the company could be left with enormous bills for computing power that it cannot easily cancel or reduce.

The scale of these costs is further illustrated by Anthropic's expanding relationship with SpaceX. The company's payments to SpaceX for computing are projected to reach up to $120.5 billion through 2029. This is a significant increase from the $1.25 billion a month reported at the time of SpaceX's own IPO filing, though either party can cancel the deal with 90 days of notice.

Dependencies and marketplace dynamics

The filing exposes some vulnerabilities regarding where Anthropic's revenue originates and who it relies on for distribution. Nearly 50% of the company's revenue in the first half of the year came through cloud marketplaces operated by Amazon, Google, and Microsoft. These three companies are unique because they act as investors, computing suppliers, and competitors to Anthropic all at once.

Because these hyperscalers collect revenue through both platform fees and the provision of compute, they are positioned as major winners in the AI ecosystem. This dynamic suggests that while Anthropic is a major player, it is deeply integrated into the infrastructure of the largest tech giants in the world. This creates a complex web of incentives and potential conflicts of interest as the market matures.

Concentration risk is another factor noted in the public documents. Anthropic's five largest customers accounted for approximately 25% of its revenue during the first nine months of 2026. Additionally, many of the company's customers pay on a usage basis rather than making long-term commitments, which could make revenue streams less predictable over time.

Policy and internal governance

Beyond the balance sheet, the filing includes details on how the company handles policy violations and internal leadership. Anthropic disclosed that between 5% and 7% of its revenue in the first nine months of 2026 came from customers using its models in violation of its terms of service. Experts suggest a significant portion of this may involve distillation, which is the practice of querying a model many times to train other systems.

The company also provided a look at its executive structure and compensation. CEO Dario Amodei received nearly $18 million in total compensation in 2025, and the filing noted millions spent on personal security for the company's two top executives. These details offer a glimpse into the high-stakes environment and the personal protections required for leaders of prominent AI firms.

Finally, the prospectus outlines an unusual succession plan for the company's leadership. The co-founders have the ability to nominate successors for their voting power, provided those nominations are approved by the other co-founders. This structure ensures that the original leadership maintains a significant level of control over the company's direction as it moves toward becoming a public entity.