Anthropic’s preparations for a potentially record-breaking IPO have revealed an unusual mix of financial metrics. The developer of Claude generates enormous revenue per employee, holds more than $130 billion in cash, and serves thousands of large enterprise customers. At the same time, the company has bound itself to commitments worth hundreds of billions of dollars for computing capacity, depends on a small group of its largest clients, and competes with the same tech giants that provide its cloud infrastructure.

The published financial documents provide the most detailed picture of Anthropic’s business to date. As Axios notes, they show both sides of the investment story: rapidly growing demand for AI and the risks associated with the colossal costs of developing it.

Arguments for a high valuation

Anthropic has already achieved significant success in the enterprise market. Nearly 2,000 customers generate at least $1 million in revenue each for the company, while 250 customers each account for more than $10 million. Half of the companies on the Fortune 500 list pay Anthropic at least $100,000 per year.

At the same time, these customers’ spending on Anthropic’s services still represents a small share of their budgets. In the third quarter of 2026, they spent less than 0.1% of their total expenditures on the company. This may indicate that the developer of Claude still has room to further grow sales to existing customers.

Technology analyst Ben Bajarin also points to the scale of the potential market: by his estimate, outside China there are about 300 million companies that spend nearly $1.5 trillion annually on enterprise software.

Another argument in Anthropic’s favor is revenue per employee. By the end of the third quarter of 2026, it had reached $6.6 million, only slightly below Nvidia’s figure, which is regarded as one of the most business-efficient players in the technology market.

As of June 30, Anthropic had more than $130 billion in cash and short-term investments. The documents also list more than $134 billion in cash and equivalent financial assets. Some investors view this as evidence of financial discipline: the company preserved a significant portion of the capital it raised despite its large-scale operations and investments.

However, high revenue per employee and a large cash reserve do not by themselves guarantee sustainable profitability in the future. To assess Anthropic, it is important to consider not only its current results, but also how much the company will have to spend on computing infrastructure.

$637 billion in commitments for computing capacity

Anthropic’s main financial risk is tied to the scale of its infrastructure commitments. According to the documents, the company has taken on commitments for computing capacity totaling about $637 billion. A significant portion of these agreements extends years into the future and includes minimum payments regardless of how much of the provided resources Anthropic actually uses.

At the same time, many customers of the Claude developer pay for model usage on a consumption basis and do not sign multi-year contracts. This creates a structural imbalance: the obligations to infrastructure providers are long-term, while future revenue from some customers is not guaranteed for a comparable period.

If demand for the models slows while computing costs remain high, it will become more difficult for the company to maintain its current financial performance. If demand continues to grow rapidly, however, the large-scale investments in infrastructure may prove justified.

Venture investor Tomasz Tunguz of Theory Ventures believes that a $2 trillion valuation for Anthropic would require the company to grow by 50%. He sees reasons for both optimism and caution.

On the one hand, Anthropic is growing quickly, demonstrating high margins and generating operating profit. On the other hand, its computing commitments are enormous, some contracts cannot be canceled, and customer loyalty is not guaranteed.

In Tunguz’s view, it is precisely the combination of these factors that makes valuing the company such a difficult task.

Largest clients and dependence on cloud platforms

Another source of risk is revenue concentration. Anthropic’s five largest customers accounted for about 25% of its revenue in the first nine months of 2026. Such dependence makes the company’s financial results sensitive to the decisions of a relatively small number of clients.

Anthropic also receives a significant share of its revenue through the cloud marketplaces of Amazon, Google, and Microsoft. In the first half of 2026, such sales accounted for nearly 50% of revenue.

All three companies are simultaneously Anthropic investors, suppliers of computing capacity, and competitors in the AI market. This model creates a complex system of interdependencies.

The cloud giants can profit from Anthropic in several ways at once: by providing infrastructure, by collecting payments for hosting and distributing models through their platforms, and by competing with it for enterprise customers. As a result, Anthropic’s growth can benefit them even in cases where they do not win direct competition for a particular customer.

For Anthropic itself, this dependence means that access to customers and computing resources largely runs through companies that have their own strategic interests in the AI market.

An unusual deal with SpaceX

The terms of Anthropic’s cooperation with Elon Musk’s SpaceX drew particular attention. According to the financial documents, Anthropic’s monthly payments to SpaceX may reach $3.1 billion, while total spending on computing capacity under the agreement through the end of 2029 may amount to $120.5 billion.

This is significantly higher than the figures that appeared in SpaceX’s documents during preparations for its own IPO: at that time, Anthropic’s monthly payments were estimated at $1.25 billion.

However, the agreement includes an important caveat: either party may terminate it by giving the other 90 days’ notice. Therefore, the stated amounts should not automatically be treated as the company’s unconditional future expenses.

The terms of the deal show how quickly the computing needs of advanced model developers are growing and what role large infrastructure providers play in their business.

Other questions about the financial reporting

Anthropic’s documents also raised questions about how the company records expenses in its financial reporting. Some observers noted that the cost-of-revenue category, which directly affects the calculation of gross profit, may include a relatively small share of total costs.

If a significant portion of expenses is classified differently, the published profitability metrics may not fully reflect the economics of the business. However, such questions alone do not prove that the company is misrepresenting its financial results: drawing conclusions requires taking into account the full reporting structure and accounting principles.

Another disclosure point concerns the use of Anthropic’s models in violation of its policies or terms of service. According to the company, such customers accounted for 5% to 7% of revenue in the first nine months of 2026. Some outside experts suggest that a significant portion of this revenue may be tied to distillation—large-scale queries to a model aimed at using its responses to train another system.

Distillation itself is a common AI development method, but its use may violate the terms of a specific service. In that case, an additional risk arises for the sustainability of this revenue: the company may restrict such use of its models.

The documents also state that CEO Dario Amodei received nearly $18 million in total compensation for 2025. Anthropic also disclosed multimillion-dollar spending on personal security for two senior executives.

Finally, the founders put in place an unusual succession mechanism: they may nominate candidates to receive their voting rights, but such appointments must be approved by the other co-founders.

Can demand justify the infrastructure spending?

The Anthropic story reflects the central dilemma of the entire advanced AI market. The company has shown that demand for powerful models can grow extremely quickly and that enterprise customers are willing to pay for access to them. But sustaining that growth requires computing capacity whose cost is measured in hundreds of billions of dollars.

The success of a potential IPO will depend not only on how attractive Anthropic’s current financial results appear to investors. They will also need to assess whether the company can maintain its technological edge, retain customers, and increase revenue quickly enough to justify its long-term infrastructure commitments.

Competition makes this task more difficult: other developers are also managing to create ever more powerful models, while cheaper alternatives may limit Anthropic’s ability to raise prices.

Thus, an investment in the company is simultaneously a bet on the continued expansion of the AI market and on enterprise demand remaining strong enough to cover the enormous cost of computing. The financial documents show how great Anthropic’s potential opportunities are—but also how expensive it may be for the company to realize them.