Anthropic files to go public on NASDAQ as $ANT with $16.3B in revenue for first six months of 2026

Anthropic filed public paperwork for an initial public offering (IPO) on Monday. The filing shows that the company is growing very quickly. It also warns that the AI systems it wants to operate on their own could create legal duties, security problems, and costs for both its customers and the company. Insurance might not cover all these costs.

The company that makes Claude reported $16.3 billion in revenue in the first six months of 2026. That was 1,300% more than the $1.16 billion it earned during the same period a year earlier, according to its S-1 registration statement. Anthropic has asked to list its shares on Nasdaq under the symbol ANT. However, it has not yet stated how many shares it plans to sell or the price range.

For leaders at technology companies, the filing provides a detailed view of how Anthropic makes money and what it depends on. Its AI tools are becoming a bigger part of everyday business operations. The filing also explains Anthropic's plans for Claude. The company wants to move beyond answering questions and writing code. It aims to build millions of AI assistants that can work on tasks independently for weeks at a time.

The $80.7 billion loss needs to be understood

Anthropic reported a net loss of $80.7 billion for the first half of the year. But this does not mean the company spent $80.7 billion in cash to run Claude.

About $72.4 billion of that loss came from changes in the accounting value of convertible notes and other financial agreements used to raise money. These changes mainly reflect Anthropic's rising company value. Its cash-flow statement also lists another $8.72 billion in noncash charges related to other financial instruments.

The company's day-to-day business shows a different picture. Anthropic earned $16.3 billion in revenue but had about $17 billion in operating expenses. This left it with an operating loss of $693 million, down from $2.14 billion a year earlier.

Spending on computing power and infrastructure reached $8.95 billion, up from $2.33 billion. Staff costs totaled $3.15 billion. Sales, marketing, and partnerships cost another $3.17 billion.

When Anthropic leaves out stock-based pay and noncash charitable contributions, it reports adjusted operating income of $1.62 billion. This figure is not profit under standard accounting rules. The company also had positive operating cash flow of $3.93 billion. It says this improvement was partly due to when it received payments from customers and paid its suppliers.

More businesses using its services helps explain the rapid growth. Anthropic says that, as of September 30, more than 1,900 customers each brought in at least $1 million in revenue over the previous 12 months. About 250 of these customers each brought in at least $10 million.

Computing costs can last longer than customer commitments

One of the filing's most important warnings is that Anthropic may have to pay for resources for longer than its customers promise to keep paying it.

Sales based on how much customers use its services made up 80% of its revenue in the first half of the year. Much of this revenue is not backed by long-term customer agreements. But contracts for computing infrastructure can require Anthropic to make payments for years, even for computing capacity it does not use.

One equipment lease disclosed in the filing alone requires about $125.2 billion in rent over five years, not including interest. Anthropic warns that failing to meet some payment obligations could cause it to break other agreements as well and lose access to resources it needs.

The company also cannot be sure it will get all the computing capacity it needs. It says it has previously asked cloud providers for more computing resources than they were willing to supply. Its contracts also do not guarantee that the resources will arrive when it wants them.

For businesses buying AI services, these details show why it is important to consider guaranteed computing capacity, uninterrupted service, and the ability to move work between providers, as well as how well an AI model performs.

Millions of AI agents—and questions about who is responsible

Anthropic's long-term plan is to let Claude take on tasks with fewer limits, work independently for long periods, and eventually run millions of copies at the same time. The filing describes this future as a “country of geniuses in a data center.”

This plan raises a practical question for customers: What happens when an AI agent makes a serious mistake?

Anthropic says it is still unclear when an agent's actions could legally commit the organization using it to something. It is also uncertain whether limits on legal responsibility written into contracts will hold up. The company warns that human reviewers may not be able to keep up with how quickly autonomous systems work or how complicated their actions become.

Some agreements with outside companies that evaluate AI models could make Anthropic responsible for certain claims from third parties without setting a maximum limit on what it must pay. Anthropic warns that a serious enough AI incident could cause damage greater than the insurance industry can cover.

Government scrutiny is already taking place. Anthropic says the Federal Trade Commission sent it a formal request for information in October. The request covers consumer protection, privacy, data protection, and product safety and security. The filing does not prove that Anthropic did anything wrong or predict what will happen in the investigation.

For businesses considering AI services, the filing makes the trade-off clear. Anthropic wants its AI systems to take on more responsibility across entire organizations. But its own warnings show why customers must carefully decide who is responsible for what. They need to set rules about which systems AI agents can access, which decisions need human approval, who must pay for losses, and how work will continue if the AI model or its computing infrastructure stops working.