What Anthropic's IPO filing reveals about AI vendor risk

Anthropic filed its IPO paperwork with the SEC on October 26, and the S-1 lays out numbers that matter to anyone paying for Claude, not just investors. Revenue for the first half of 2026 hit $16.3 billion, roughly four times what the company made in all of 2025, and 80 percent of that revenue comes from metered API and model usage rather than flat subscriptions. Five customers account for about a quarter of it. If your business touches Claude through n8n, Zapier, a vendor's app, or the API directly, this filing is the clearest public look yet at how your bill gets set and how much pull you have as a customer.

A prospectus like this is written for investors, but it forces disclosure a product page never would: real revenue, real customer counts, real concentration numbers. Most vendors you build automations on, your form plugin, your CRM, your workflow tool, never file one. Anthropic just did, and a few of those numbers are worth your attention before your next pricing email arrives.

What the filing actually shows

Net revenue retention came in around 900 percent through the third quarter. That means accounts that signed up a year ago are now spending roughly nine times as much. The filing also shows more than 1,900 customers paying $1 million or more a year, and over 11,900 paying $100,000 or more in the trailing twelve months. The model that produces those numbers is simple: land an account on usage-based pricing, then let usage grow on its own as the integration gets used more. A business running agents through n8n or a vendor's Claude integration is exactly the kind of account that model is built around, because workflow volume tends to climb once something works and nobody is watching the meter.

The AI vendor risk hiding in a usage-based bill

Once a company answers to public shareholders, the pressure to grow revenue and protect margin gets louder, and pricing, rate limits, and model retirement schedules are the levers available to do that. The filing also notes that a substantial share of revenue depends on a limited number of customers and distribution channels. With roughly 250 customers contributing $10 million or more each, a small account's usage is a rounding error to the vendor. That is normal for any SaaS company at this scale, but it means the roadmap, support priority, and the fine print on pricing changes will track what the largest accounts need, not what a five-person shop needs.

Three things to check in your own Claude usage

1. 1

   Pull your real monthly spend

   Check your Claude or API billing dashboard for the trailing 90 days, not a single invoice. Usage-based pricing means a slow month hides what a busy one actually costs.

2. 2

   Map which workflows call the API directly

   List every n8n node, Zap, or script that calls Claude and note which model each one is pinned to. You cannot manage a cost you cannot locate.

3. 3

   Set a spend alert, not a guess

   Most API dashboards let you set a usage threshold alert. Turn one on so a runaway loop or a traffic spike shows up as a notification instead of a surprise invoice.

If you're running Claude through an n8n flow, a Zapier app, or a tool you built on the API, take a plain look at what you're actually spending and what happens if that usage-based line moves. Send over your setup and I'll help you map where the cost sits and whether it's time to change how you're metering it.