OpenAI Business Model Explained: APIs, ChatGPT, and Enterprise

Update, November 2026: OpenAI told investors its annualised revenue was about $50 billion at the end of September, below the $68 billion figure widely reported because the larger number included partners’ gross revenue. It is valued at $852 billion after a $122 billion round in March, has reportedly filed confidentially for an IPO, and is reportedly eyeing 2027. More in our reality check.

The business in four lines

Each line has different margins, lock-in and cost to serve. Consumer subscriptions are simple but support heavy use at a flat price. The API is usage-based and flexible. Enterprise deals bring larger contracts and more support obligations.

The numbers behind it

According to CNBC’s report of an investor presentation first covered by the Financial Times:

• OpenAI reached roughly $50 billion of annualised revenue at the end of September.
• The widely reported $68 billion figure from late September included partners’ gross revenue.
• Run-rate grew 77% in the third quarter, and the enterprise business grew 107%.
• OpenAI is valued at $852 billion after a $122 billion round in March, has reportedly filed confidentially for an IPO and is reportedly eyeing 2027.
• It is in early talks about another round of around $30 billion.

Market reaction was sharp. On the day, Nvidia fell about 3%, Oracle nearly 6% and CoreWeave nearly 8%.

Why the number moved

Run-rate means multiplying a recent period by twelve, and it has no standard definition. Anthropic includes cloud partners’ sales in its run-rate, while OpenAI leaves them out. Neither is dishonest, but they are not comparable until adjusted. Public filings report audited revenue, which will settle the question. Compare with Anthropic’s business model.

What it means for the economics

OpenAI has committed to very large compute purchases and relies on revenue growth to cover them. That makes it sensitive to growth slowing and to the cost of capital. The wider discussion is in AI circular financing and the bear case. As a customer, you do not need to resolve the debate, but you should treat your dependency on any single lab as a risk.

What to evaluate as a customer

1. Quality on your tasks. Build a test set from real work and score outputs.
2. Cost per successful task. Include retries, caching and the cost of failures. See token costs.
3. Limits and reliability. Rate limits, uptime history, support response.
4. Data terms. Retention, training use, regional options and enterprise controls.
5. Model lifecycle. How long models stay available and how deprecations are handled.
6. Contract flexibility. Commitments, minimums and price protection.

Build in portability

• Put a thin routing layer between your product and any model API.
• Keep prompts, tool schemas and evaluations model-agnostic where possible.
• Test an alternative provider on the same evaluation set every quarter.
• Consider an open-weight option for sensitive or high-volume tasks.
• Avoid features that only work on one vendor unless the value is clearly worth the lock-in.

Using the consumer products at work

Teams often adopt ChatGPT informally before IT approves it. Set a policy: which plan is allowed, what data may be pasted in, and how outputs are checked. Business and Enterprise plans typically offer admin controls and different data handling from personal accounts, so confirm the terms for your plan.

Reading the next headlines

Watch for:

• A public filing with audited revenue and margins.
• How partner revenue is accounted for.
• The share of revenue from a few large customers.
• Commitments to chips and data centres, and whether they are cancellable.

An IPO would turn many of these into reported facts, as Anthropic’s S-1 did for its numbers. See the AI reality check for the comparison.

Questions for your OpenAI contract

Ask about rate limits at peak, model retirement notice, data retention and training terms, price protection on commitments and whether you buy directly or through a partner. Keep the answers with your cost-per-task data.

Your next step

Add a short vendor file for each AI provider: what you use, monthly cost, alternatives tested and contract terms. Review it at every renewal and whenever a provider announces a pricing or model change.

Bottom line

OpenAI is a fast-growing business built on subscriptions, an API and enterprise deals, with enormous compute commitments behind it. The October revenue restatement was a reminder to ask how numbers are counted. Evaluate it on your own tasks and cost per outcome, and architect your product so that no single vendor decision can stop you.