What OpenAI actually does
OpenAI makes ChatGPT, the most widely used AI assistant, and the GPT models behind it. It earns money from people who subscribe to ChatGPT, from companies that buy it for their staff, from developers who pay for the models by usage, and since 2026 from advertising. Running all of it takes one of the largest computing build-outs ever attempted.
How OpenAI makes money
OpenAI started as a consumer business and is becoming an enterprise one. ChatGPT gave it the largest user base in AI; companies, developers and advertisers are where that audience is increasingly turned into revenue.
There are four ways money comes in. Individuals pay monthly for ChatGPT plans, from the low-cost Go tier to the $200 Pro plan. Companies buy ChatGPT Business and Enterprise seats and larger agent platforms. Developers and software companies pay per token for the models through the API. And free users now see ads, which OpenAI began testing in February 2026.
The balance shifted in 2026. OpenAI entered the year with roughly 60% of revenue from consumers, but its chief financial officer, Sarah Friar, told shareholders in August 2026 that enterprise revenue had overtaken the consumer business. Codex, its coding agent, and company-wide ChatGPT deployments drove the change.
On the latest figures in our snapshot, OpenAI's annualized revenue run-rate is about $70B (End Sep 2026). The full revenue history is on our financials page.
| Channel | Who pays | How it is billed | For example |
|---|---|---|---|
| Consumer subscriptions | Individuals | Monthly plans: Go, Plus and Pro | More than 50 million consumer subscribers by February 2026 |
| ChatGPT for business | Companies, schools and governments | Per seat per month, or annual enterprise contracts | More than 9 million paying business users by February 2026 |
| API and agent platform | Developers, software companies and enterprises | Per million tokens, priced by model; enterprise platform contracts | OpenAI Frontier, its platform for running teams of AI agents, also sold through AWS |
| Advertising | Advertisers | Ads shown to Free and Go users | Passed a $1 billion annualized run-rate by the end of August 2026 |
The products
ChatGPT is the product most people know, but OpenAI also sells its models to developers, a coding agent, an agent platform for companies, and a growing set of consumer services built on top.
The GPT models
Everything runs on OpenAI's GPT family. GPT-6, released on September 3, 2026, is its newest generation, aimed at computer use, browsing, software engineering, science and professional work. Its most capable version went first to approved business testers and arrived in paid ChatGPT plans and the API in the following days, with extra guardrails for cybersecurity uses.
OpenAI sells several sizes at different prices, and like its rivals it cuts the price of a given level of capability over time. Cheaper models in 2026 put pricing pressure on competitors including Anthropic, but also lower what OpenAI earns per task, so its growth depends on usage rising faster than prices fall.
ChatGPT
ChatGPT launched in November 2022 and reached 100 million users in about two months, then the fastest growth of any consumer app. By 2026 it had more than a billion weekly users. Plans run from free to Go, introduced worldwide in January 2026, through Plus and Pro, with Business, Enterprise and Edu plans for organizations.
The app has grown into a platform. Users can shop and check out inside ChatGPT, use apps from other companies within it, and browse the web with Atlas, OpenAI's browser, launched in October 2025. OpenAI also bought io, the hardware start-up founded by former Apple designer Jony Ive, in 2025 to build its own devices.
Codex and agents
Codex is OpenAI's coding agent, which writes, tests and fixes software on a developer's behalf. Together with OpenAI's other agent features it reached 20 million weekly active users by mid-August 2026, and it competes directly with Anthropic's Claude Code, GitHub Copilot and Cursor.
For companies, OpenAI Frontier is a platform for building and managing teams of AI agents that carry out work across a business's own systems. Amazon Web Services is its exclusive outside cloud distributor under the partnership the two companies signed in February 2026.
The API
The API lets developers build OpenAI's models into their own products, paying per token. It was OpenAI's first commercial product in 2020, long before ChatGPT, and it is how thousands of other companies' AI features work behind the scenes.
Until April 2026, OpenAI's models were sold through Microsoft's Azure as the only major cloud outside OpenAI itself. A renegotiated agreement ended that exclusivity, and OpenAI now sells its models on other clouds too, starting with AWS.
Advertising
OpenAI began testing ads in ChatGPT in the United States on February 9, 2026, for logged-in adult users on the Free and Go plans; paid plans have no ads. It says ads do not influence ChatGPT's answers and that conversations are not shared with advertisers. By the end of August 2026 the ad business had passed a $1 billion annualized run-rate and expanded to India and parts of Asia.
Advertising is how OpenAI plans to earn money from the large majority of ChatGPT users who never pay, and it puts the company in competition with Google and Meta for ad budgets. It also drew criticism, including a Super Bowl campaign from Anthropic mocking ads in AI chatbots.
Who pays for OpenAI
OpenAI has by far the largest audience in AI: more than a billion weekly ChatGPT users in 2026, more than a million organizations by December 2025, and millions of developers. Only a small share of those users pay, which is why turning the audience into revenue is the central business question.
Consumers were the original engine. More than 50 million people paid for ChatGPT by February 2026, most of them on the $20 Plus plan. Subscriptions remain a very large business, but growth there has slowed relative to enterprise, and the cheaper Go plan and advertising are meant to reach users who won't pay $20 a month.
Companies are now the larger and faster-growing part. More than 9 million business users paid for ChatGPT by February 2026, and organizations from banks to governments have rolled it out widely. Enterprise sales rely on OpenAI's own sales force, Microsoft's reach through Azure and Copilot, and, since 2026, Amazon's.
Developers form the third group. Thousands of software companies build on OpenAI's models through the API, though many now spread their usage across OpenAI, Anthropic and Google to get the best price and performance for each task.
Compute: the biggest bet
OpenAI has signed up for more computing capacity than any other AI company. Its multi-year commitments add up to well over a trillion dollars in headline value, far more than its revenue today.
The commitments span almost every supplier. Microsoft remains the primary cloud partner, with a $250 billion Azure commitment, but OpenAI has added Oracle, Amazon, CoreWeave and Cerebras, chips from Nvidia and AMD, and its own custom accelerators designed with Broadcom. Stargate, launched in January 2025 with SoftBank, Oracle and MGX, aims to build up to 10 gigawatts of US data centers, starting in Abilene, Texas.
Some of the deals turn suppliers into shareholders, or the reverse. Nvidia invested $30 billion in OpenAI in 2026, replacing an earlier plan for up to $100 billion tied to data-center capacity. AMD gave OpenAI a warrant for up to 160 million AMD shares, about a tenth of the company, as part of a 6-gigawatt chip deal. Critics call these arrangements circular, because the money a supplier invests comes back to it as purchases.
| Partner | Announced | Hardware | What it covers |
|---|---|---|---|
| Stargate (SoftBank, Oracle, MGX) | January 2025 | Nvidia GPUs | Up to $500B and 10 GW of US data centers over four years, starting in Abilene, Texas |
| CoreWeave | March 2025 | Nvidia GPUs | $11.9B over five years, expanded twice in 2025 to about $22.4B; OpenAI also took a stake in CoreWeave |
| Oracle | September 2025 | Nvidia GPUs | About $300B of cloud capacity over roughly five years from 2027 |
| AMD | October 2025 | AMD Instinct | 6 GW, the first 1 GW from the second half of 2026; AMD granted OpenAI a warrant for up to 160M shares |
| Broadcom | October 2025 | Custom OpenAI accelerators | 10 GW of chips designed by OpenAI, deployed from late 2026 |
| Microsoft Azure | October 2025 | Nvidia GPUs | An extra $250B of Azure services; Microsoft gave up its right of first refusal on OpenAI's compute |
| Amazon Web Services | November 2025 | Nvidia GPUs, later Trainium | $38B over seven years, expanded by $100B over eight years in February 2026, including about 2 GW of Trainium |
| Cerebras | January 2026 | Cerebras wafer-scale systems | 750 MW through 2028, reported at more than $10B |
The scale is the point and the risk. OpenAI argues that capacity is what limits its growth, and every deal secures more of it. But the commitments run far ahead of current revenue, OpenAI expects to keep burning cash for years, and some suppliers, Oracle in particular, have tied their own growth plans to OpenAI's ability to pay. For what these costs mean for losses and cash, see our financials page.
How OpenAI is governed
OpenAI began in 2015 as a nonprofit research lab. After years of legal and political fights over its structure, it is now a public benefit corporation controlled by that original nonprofit.
In 2019 the nonprofit created a capped-profit company to raise money, which is how Microsoft came to invest. On October 28, 2025 OpenAI completed a recapitalization: the for-profit became OpenAI Group PBC, the nonprofit was renamed the OpenAI Foundation and kept about 26% of the equity plus the right to appoint the company's whole board, and Microsoft ended up with about 27%. Employees and other investors hold the rest.
The Microsoft relationship was redrawn twice. The October 2025 deal extended Microsoft's rights to OpenAI's technology through 2032 and committed OpenAI to $250 billion more of Azure. A further agreement in April 2026 ended Microsoft's exclusivity, so OpenAI can sell on any cloud. It also ended Microsoft's revenue-share payments to OpenAI, while OpenAI keeps paying Microsoft a share of its revenue through 2030, now subject to a cap.
Elon Musk, a co-founder and early donor, sued to unwind the for-profit shift. A jury rejected his main claims in May 2026, finding that he had waited too long to sue. OpenAI is led by chief executive Sam Altman, who was briefly fired and reinstated by the old nonprofit board in November 2023; Bret Taylor chairs the board.
For who owns the company, see the investors page; for where the listing stands, see the IPO page.
Risks to the business
OpenAI's reach is unmatched, but its business carries large and specific risks. These are the ones that bear most directly on how it makes money.
Compute commitments
OpenAI has committed to far more spending than it earns, years ahead. If revenue growth disappoints, as reports in April 2026 said it had against internal targets, the gap has to be funded by new capital.
Cash burn
OpenAI expects to lose money for several more years while it builds capacity. That makes it dependent on continued access to very large funding rounds, debt and, eventually, public markets.
Competition
Anthropic has grown faster with businesses and in coding, Google bundles Gemini into products billions already use, and open-weight models undercut on price. ChatGPT's lead with consumers is large but not guaranteed.
Monetising free users
Most ChatGPT users pay nothing. Advertising is the plan to change that, but it risks user trust and pits OpenAI against Google and Meta, the two companies best at selling ads.
Partners that compete
Microsoft is its largest shareholder, its main cloud and also a rival selling its own AI. Amazon backs OpenAI and Anthropic at once. Suppliers holding OpenAI equity, and OpenAI holding AMD warrants, tie its fortunes to theirs.
Legal and regulatory
OpenAI faces copyright suits from publishers and authors, scrutiny of its structure and safety practices, and questions from regulators in the US and Europe. Altman himself cited AI safety concerns in September 2026 when ruling out a 2026 IPO.
The public S-1 will be the first document to set these out in OpenAI's own words, next to audited numbers. See the IPO page for where that stands.