How AI Companies Make Money: Subscriptions, APIs and Enterprise Deals Explained
- Standfirst (first paragraph, distinct from meta): "AI companies make money in three main ways: monthly subscriptions from people, per-token API bills from developers, and per-seat enterprise contracts from organisations.
AI companies make money in three main ways: monthly subscriptions from people, per-token API bills from developers, and per-seat enterprise contracts from organisations.
Key points
- Consumer subscriptions are priced per user per month: Claude Pro is $20 billed monthly, Google AI Pro is $19.99 a month, and every major chatbot keeps a free tier underneath.
- APIs are a separate business: developers pay per million tokens, from $0.10 input on OpenAI’s gpt-6-luna to $10 input on gpt-6-astra.
- Enterprise buyers pay for control: Anthropic’s Enterprise plan is $20 per seat per month plus usage at API rates, and Google’s Gemini Enterprise editions start at $21 per seat per month.
Three streams pay for the same underlying models
One model can earn money from a casual subscriber, a busy developer and a large employer at the same time.
If you have ever wondered how a chatbot that answers your questions for free can possibly be a business, you are asking the right question. Running these models is not free for the companies, even when it is free for you. Read the official pricing pages side by side, though, and the answer is simpler than the hype suggests: OpenAI, Anthropic and Google all sell the same technology in three wrappers, each suiting a different customer.
The first wrapper is the subscription: a fixed monthly fee for a chatbot app. The second is the API, where developers rent the model by the token and build it into their own products. The third is the enterprise deal: per-seat payment, sometimes with usage on top, in return for security and administration an individual never asks about. Our earlier guide to AI subscriptions and pricing covers the consumer plans in depth; this article looks at how the three streams fit together.
Subscriptions turn heavy users into predictable monthly income
Paid plans are priced per user per month, and what rises with price is usage, not a different chatbot.
OpenAI’s FAQ puts the structure plainly: the free version of ChatGPT is available to everyone, and paid plans — Go, Plus, Business and Enterprise — are priced per user per month, with monthly plans for Go, Plus and Business and annual plans for Business and Enterprise. The free tier includes unlimited text chats subject to abuse guardrails, but limits uploads, image creation, voice chats, deep research, memory and Codex access. Each paid step mainly buys back room: more uploads, more images, longer memory, expanded deep research and, on Plus, projects, scheduled tasks and custom GPTs.
Anthropic publishes the same ladder with prices visible. Claude is free with limited usage on five-hour and weekly windows. Pro costs $17 per month with the annual discount, $200 billed up front, or $20 billed monthly, adding more usage, Claude Code and Projects. Max costs $100 a month for five times Pro usage; Ultra costs $200 for ten times Pro’s weekly usage, with no five-hour limit for verified accounts. Google’s plans rhyme: Free is $0 with 15 GB of storage, AI Plus is $4.99 a month for twice the free limits and 400 GB, and AI Pro is $19.99 for four times the free limits, Gemini in Gmail and Docs, and 5 TB. See also ChatGPT free versus Plus.
The API rents the same models to developers by the token
Instead of a monthly seat, software pays for exactly the text it sends in and gets back.
The second stream is quieter, and for many products you use every day it is the important one. When an app summarises your emails or answers support questions, it is usually not someone typing into a chatbot. The app calls the model through an API and pays for the tokens it processes, priced per million tokens on the official rate cards.
OpenAI’s API pricing page shows how wide that market is: standard rates run from gpt-6-luna at $0.10 per million input tokens and $0.50 per million output tokens, up to gpt-6-astra at $10 input and $50 output. Anthropic and Google publish the same per-token tables for Claude and Gemini. That spread is the business in miniature: simple work pays pennies, while the strongest reasoning costs a hundred times more for the same million words. Our cluster guide to AI API pricing, tokens and tiers walks through those rate cards.
Enterprise deals sell control, security and accountability
What a large buyer purchases on the official plan pages is governance around the model, not a smarter chatbot alone.
The third stream is the least visible and, seat for seat, the most valuable. Anthropic states its Enterprise price openly: $20 per seat per month plus usage billed at API rates, billed annually, including SCIM, audit logs, custom data retention and role-based access. Anthropic also says Enterprise and API pricing can be customised with volume and committed-spend discounts through its sales team, and that it does not publish those discount rates.
Google sells Gemini Enterprise in editions. The Business edition, for teams of up to 300 seats with no IT setup needed, starts at $21 per seat per month. Standard and Plus start at $30 per seat, with unlimited seats and advanced security features such as VPC-SC and customer-managed encryption keys. Google stresses the promises that relax a cautious IT department: you own your data, prompts and outputs are not used to train its models, and customer data is not sold or used for advertising. The Gemini API page shows the same pattern one level down: free-tier content is used to improve Google’s products, paid-tier content is not.
Free tiers exist because they feed the other three streams
A free plan costs the company something on every chat, so it has to earn its keep as an introduction.
Every company above gives a capable model away, at real computing cost. Google’s API page describes the free tier as a way to start building free of charge with generous limits, then scale up with prepaid and pay-as-you-go pricing for production applications.
In our view, the free tier is a shop window with a till just inside the door. It lets a curious person form a habit before anyone asks for a card, and a developer test an idea before committing a budget. Both then reach a moment — a limit hit, a feature locked — when upgrading feels like their own idea rather than a sales pitch. That is analysis, not a company quote, but it fits every pricing page below: generous at the bottom, metered above it.
What this means
The money comes from the people who hit limits, build products or need guarantees — not from the casual free chat.
Most of us meet AI as a friendly free box on a screen, so it is easy to assume the business must be mysterious. Reading the pricing pages is almost a relief: there is a person paying $20 a month because the free limit kept interrupting their work, a small team paying per seat for one bill and an off switch when someone leaves, and a developer watching a per-token meter, because every customer question costs a fraction of a cent and those fractions decide whether the app survives.
That picture should change how you judge your own spending, too. If you chat a few times a week, the free tiers are genuinely built for you, and paying mostly buys headroom you may never use. If the limit messages keep landing in the middle of your working day, a subscription is not a luxury; it is the company charging rent on the exact frustration you already feel. And if you are building something for other people, your real cost is the API meter, and the model you choose per task matters more than the brand on the door.
In our view, the healthiest way to read AI pricing news is to ask one question: which stream is this about? A cheaper consumer plan is a fight for subscribers. A token price cut is a fight for developers, and it quietly lowers the cost of every app built on that model. A new enterprise edition is a fight for buyers who care about audit logs and data promises more than clever answers. Each stream steadies the others: subscriptions are predictable, API usage grows with other people’s success, and enterprise contracts are sticky. You just need to know which till your own money goes into, and what, exactly, it buys you there.
Bottom line
Subscriptions rent convenience, APIs rent raw model time, and enterprise deals rent trust and control.
AI companies make money the unglamorous way: per person, per token and per seat. The free tiers recruit, the paid tiers monetise, and the pricing pages below show every step in the companies’ own words.
Related Guides
Go deeper on API pricing and the consumer plans behind the first stream.
- AI API Pricing Explained: Tokens, Tiers and Why the Same Model Costs Different Amounts
- AI Subscriptions and Pricing Explained
- ChatGPT Free vs Plus: What You Actually Get on Each Plan
Sources
The official pricing pages behind every price and plan detail in this article.
