Anthropic and Claude Business Model Explained for Founders

Update, November 2026: Anthropic publicly filed its S-1 on 26 October to list on Nasdaq as ANT. Per Fortune’s summary: $16.3 billion of revenue in the first half of 2026, a $974 million operating profit in Q2, and $498 billion of noncancelable compute commitments. More in our reality check.

How Anthropic sells Claude

Anthropic’s products reach customers through several channels.

The API is the heart of it for builders, and the plans and coding tools are where much of the recent growth has come from.

What the S-1 shows

On 26 October 2026, Anthropic publicly filed its S-1 to list on Nasdaq as ANT. The headline figures, as reported by Yahoo Finance and Fortune (we have not read the filing itself):

• Revenue: $16.3 billion in the first half of 2026, against $4.6 billion for all of 2025. Q2 was $11.6 billion, up from $4.7 billion in Q1.
• Profit: $974 million of operating income in Q2, the first operating profit shown. Q1 had a $1.7 billion operating loss.
• Net loss: $80.7 billion in H1, mostly non-cash fair-value changes on convertible notes.
• Customers: enterprise customers paying over $100,000 a year rose to 11,900 from 950 a year earlier.
• Compute: eight contracts worth $657 billion, $498 billion of them noncancelable.
• Cash: $131.7 billion in cash and short-term investments, with the IPO expected to add about $100 billion.

The tension in the numbers

Revenue is growing very fast and the operating line turned positive. But compute commitments run at roughly $70 billion a year on Fortune’s reading, and Q2 revenue annualises to about $46 billion, our arithmetic. Management is betting that revenue will keep outrunning a fixed cost base. The S-1 itself warns that new compute can compress margins before revenue arrives.

We cover the wider picture in the AI reality check and the sceptic’s view in Ed Zitron’s five messages.

Revenue definitions

Anthropic includes sales through cloud partners in its run-rate, while OpenAI leaves them out, according to Investing.com’s summary of the Financial Times report. Compare the two labs’ headline numbers only after adjusting for that difference. A listed company reports audited revenue, which ends much of the ambiguity.

What it means for a customer

Where Claude fits well. Teams often choose it for long documents, careful writing, tool use in agent workflows and coding. Claude Code has made it a common choice for engineering teams.

Where to be careful.

• Concentration: if one model powers your product, a pricing or policy change hits you directly.
• Cost per task: compare successful outcomes, not list prices. See token costs.
• Change of ownership structure: a public company answers to quarterly targets, which can influence pricing, rate limits and roadmap.
• Security: agents with tool access need strict permissions. See AI agent security.

A simple evaluation checklist

1. Build a test set of 50 to 100 real tasks from your product.
2. Run Claude and at least one alternative and score quality.
3. Calculate cost per successful task, including retries.
4. Test long-context behaviour, tool use and refusal patterns on your data.
5. Check enterprise terms: data retention, training use, regional availability and support.
6. Repeat quarterly, because models change quickly.

Building resilience

• Route all model calls through your own interface.
• Keep prompts and evaluations portable.
• Use a second provider for critical flows, or an open-weight option for sensitive work.
• Watch the vendor’s disclosures, since filings are more reliable than slides.

For the rival’s model, see OpenAI’s business model.

Questions for your Claude contract

Before you commit volume, check:

• Which models are available in your region, and how long are they supported?
• How are rate limits set for your account, and how quickly can they be raised?
• What are the data retention and training terms for your plan?
• Is there price protection on committed usage?
• Do you buy directly or through a cloud marketplace, and how do terms differ?

Record the answers beside your cost-per-task numbers so a later review has context.

Your next step

Add one competitor to your evaluation set this week and run your fifty hardest real tasks through both. Record quality, cost per successful task and latency. Keep the results, because they are your leverage in any pricing conversation.

Bottom line

Anthropic has built a fast-growing business around the Claude API, subscriptions and coding tools, and its S-1 now shows audited growth and an operating profit alongside very large compute commitments. Use Claude where it is strong, measure cost per successful task, and keep your architecture portable so that one vendor’s decisions do not become your problem.