Two Customers Account for a Fifth of Anthropic's Revenue, a Concentration Risk IPO Buyers Should Know

Anthropic, one of the world's fastest-growing AI companies, plans to go public by the end of the year. Reports say it is aiming for a $2 trillion valuation. If it reaches that value, it would be the biggest IPO (initial public offering) in history and one of the world's most valuable companies.

However, recent leaks from its confidential S-1 filing, a document companies submit when they plan to go public, reveal some serious warning signs. Reports say Anthropic made $16.3 billion in revenue in the first half of 2026 but lost a huge $80.7 billion overall. Of that loss, $72 billion came from accounting charges that did not involve actual cash payments. Even so, the numbers raise serious questions about Anthropic's claim that it reached "adjusted" profitability in the third quarter. This means it reported making a profit after leaving out certain costs.

Also, about one-fifth of Anthropic's revenue in the first half of 2026 came from just two customers. Anthropic did not name these companies in its S-1 filing, but several industry reports have identified them as SpaceX (SPCX +7.17%) and Microsoft (MSFT -0.02%). Let's look at why these two companies depend on Anthropic and whether relying so heavily on just two customers could be a major long-term risk.

Why are SpaceX and Microsoft relying on Anthropic?

In August, SpaceX's AI business acquired Cursor, a start-up that makes AI coding tools, for $60 billion. Cursor's code editor is built around AI and lets users work with Anthropic's Claude AI models.

Microsoft uses Claude in GitHub Copilot, an AI assistant that helps people write computer code. GitHub first developed Copilot with OpenAI. As GitHub added more AI models to its platform, millions of developers began sending their requests through Claude.

Anthropic earns money whenever these platforms use Claude through its API, or application programming interface. An API lets software communicate with another service. As more developers use Claude through Cursor and GitHub Copilot to help them write code faster, Anthropic's revenue should continue to grow.

Is this a major risk for Anthropic?

Claude's coding models, especially Sonnet and Opus, are widely seen as some of the best AI tools for writing complex code. Users of Cursor and GitHub Copilot can also choose other AI models, including OpenAI's GPT, Alphabet's Google Gemini, and SpaceX/xAI's Grok. Even with these other choices, most programmers still use Claude.

However, according to Anthropic's S-1 filing, its biggest customers pay to use its models based on how much they use them, without signing long-term contracts. This means they can switch to other services without waiting for a long contract to end. If Anthropic's competitors catch up and offer similar coding models at lower prices, Anthropic's revenue could fall sharply. If competition forces the company to lower its prices, its net losses will probably grow.

Anthropic will likely attract a lot of attention when it finally goes public. But, as with SpaceX, investors may become less excited once they start to notice its major weaknesses.