Anthropic’s S-1 shows how Claude is built to keep enterprise customers

Anthropic has filed its S-1 with the Securities and Exchange Commission, formally starting the paperwork toward an IPO. The filing’s biggest message isn’t about model benchmarks. It’s about how hard Anthropic plans to make leaving.

The Claude maker is building a web of connected products meant to keep enterprise customers locked in. Based on the numbers in the filing, it appears to be working.

The customer pyramid

As of September 30, Anthropic counted more than 11,900 customers spending at least $100,000 a year. More than 1,900 of those spend upwards of $1 million annually.

At the top sit about 250 customers spending $10 million or more each year. These aren’t pilot programs run from an innovation lab. A $10 million annual line item gets a CFO’s attention, and usually a long negotiation before it gets approved.

Anthropic made that case in the filing itself:

“The speed and depth of our enterprise customer adoption are powerful evidence of Claude’s position at the frontier of AI.”

The ecosystem strategy

Instead of selling one product, Anthropic sells several that work better together. Anthropic bundles Claude across APIs, developer tools, and enterprise chat products. A company might start by calling the API for one internal tool. Then its engineers pick up the developer tools, and its staff starts using the chat product. Each extra product makes switching to a rival a bigger project.

For an AI lab, the stakes are higher than usual. Foundation models can look interchangeable from the outside, and a rival release can shift attention within weeks. An ecosystem gives customers reasons to stay that have nothing to do with who topped the latest leaderboard.

Who’s behind the bet

Anthropic has raised approximately $12.7 billion to date. Amazon has committed up to $4 billion, and Google invested around $2 billion. Both companies also act as distribution channels, folding Claude models into their own cloud ecosystems.

That arrangement cuts both ways. Hyperscaler partners give Anthropic reach into enterprise accounts it might otherwise spend years courting. They also mean some customer relationships run through platforms Anthropic doesn’t own.

The competitive field is crowded. Competitors include OpenAI, Google, Microsoft, and Meta. Several of them also compete with or partner with Anthropic’s own investors.

What this means

For prospective public investors, the customer tiers are the headline metric. Revenue concentration in a few hundred very large accounts can be a strength or a vulnerability, depending on how sticky those accounts turn out to be.

The ecosystem strategy is Anthropic’s answer to that risk. If customers who spend $10 million a year are using several connected products, losing one of them becomes less likely. If they’re mainly buying raw API access, they could move spending to a cheaper or newer model with relatively little friction.

The filing also lands in the middle of heavy AI infrastructure spending by major tech companies. Large capex commitments from big technology players are a tailwind for model providers like Anthropic. That spending only pays off if enterprises keep buying AI services at scale, which is exactly what Anthropic’s customer counts are meant to demonstrate.

A successful Anthropic IPO may point to a strong appetite for AI-focused companies in public markets. A weak reception could do the opposite, making other private AI labs rethink their own timelines.