OpenAI spent its first years as a research lab chasing a scientific goal. It now runs something closer to a utility. ChatGPT reached 800 million weekly active users by October 2025, roughly double the 400 million it reported in February of that year, and OpenAI says it crossed 900 million by early 2026. A product that most people first tried as a novelty has become a default tool for writing, coding, and search-adjacent lookups. That shift, from lab to load-bearing service, is the real story, and it explains both the company's leverage and its exposure.

The pattern rhymes with earlier platform waves. AWS turned servers into a metered API. Stripe turned payments into a few lines of code. OpenAI is doing the same thing with frontier models: wrapping expensive, hard-to-build capability in an interface any developer can call. The difference is the cost base. Serving a search query is cheap. Serving a reasoning model is not.

How OpenAI Became Default Infrastructure

Two businesses sit under one brand. The consumer side is ChatGPT, sold as free, Plus, and enterprise tiers. The developer side is the API, where companies pay per token to embed OpenAI models inside their own apps. The API is what makes OpenAI infrastructure rather than a destination site. Customer support tools, coding assistants, and internal copilots increasingly route their intelligence through a handful of model providers, and OpenAI is the most widely adopted of them.

Pricing is usage-based, split between input tokens (what you send) and output tokens (what the model returns), with no flat access fee. That structure is deliberate. It lets a solo developer start for cents and lets an enterprise scale to millions of calls without renegotiating a contract. It also means OpenAI's revenue rises with usage rather than seat count, which is attractive when a product is spreading fast and awkward when the underlying compute is expensive.

The strategic value is lock-in through habit. Once a company writes its prompts, evaluation suites, and fine-tunes against a specific model family, switching carries real cost. That is the same moat cloud providers built, and it is why every serious model lab now ships an API first and a chatbot second.

The Microsoft Relationship, Rewritten

No account of OpenAI works without Microsoft. The two struck a deal that made Azure OpenAI's primary cloud and gave Microsoft privileged access to the models it then wove into Copilot across Windows, Office, and GitHub. For years that partnership was described as exclusive and tightly coupled.

That changed in late 2025. OpenAI completed a for-profit restructuring into a public benefit corporation controlled by its nonprofit, and under the revised terms Microsoft took roughly a 27% stake in the new entity, with IP access extended through 2032. Two clauses matter more than the headline percentage. Microsoft's cloud exclusivity ended, freeing OpenAI to buy compute elsewhere. And OpenAI committed to purchase a further $250 billion in Azure services, a figure that tells you how capital-intensive this business has become.

Read together, those terms describe a maturing relationship. OpenAI gained room to diversify suppliers and raise capital. Microsoft locked in a large ownership position and a very large customer commitment. Neither side can walk away cleanly, which is usually a sign a partnership has stopped being a bet and started being a dependency.

The Commercial Pressure Ahead

Here the numbers turn sobering. OpenAI reported roughly $13 billion in revenue for 2025, a remarkable figure for a company that barely sold anything three years earlier. But reported losses dwarf that revenue, driven by the cost of training and, increasingly, of inference at scale. OpenAI has told investors its compute bill could approach $600 billion by 2030, a scale of commitment that reframes the whole business. Serving hundreds of millions of weekly users is not a fixed cost you amortize once. Every conversation consumes GPUs.

That is the central tension. The consumer product's success is also its heaviest liability. Free ChatGPT users generate goodwill and data but little direct revenue, while each query costs money to answer. A spending commitment that large only pencils out if usage monetizes far better or model-serving costs fall sharply. Both may happen. Neither is guaranteed.

Competition adds a second squeeze. Anthropic, Google, Meta, and a growing set of open-weight models keep narrowing the quality gap. When rivals ship a comparable model, per-token prices fall, which is good for developers and hard for anyone trying to earn back a training run. OpenAI's answer has been to move up the stack, toward agents, reasoning models, and tightly integrated products where switching is harder and value is easier to charge for.

The honest read is that OpenAI has won the adoption war and not yet the economics war. It built the most-used AI product on earth and turned frontier capability into something you can rent by the token. Whether that becomes a durable, profitable utility or an expensive land grab depends on questions still unresolved: how fast inference costs drop, how much enterprises will pay for reliability, and whether the Microsoft alliance stays a strength as both companies chase the same customers.

Frequently Asked Questions

How many people use ChatGPT?

OpenAI reported 800 million weekly active users in October 2025, up from about 400 million in February of that year, and said the figure passed 900 million in early 2026. Weekly active users count people who used the product in a given week, so the number reflects habitual use rather than one-time sign-ups.

What is the difference between ChatGPT and the OpenAI API?

ChatGPT is the consumer-facing app, sold in free and paid tiers. The OpenAI API is the developer product: companies pay per token to call OpenAI's models inside their own software. The API is what lets OpenAI function as infrastructure that other products are built on top of.

Does Microsoft own OpenAI?

No. After OpenAI's 2025 restructuring, Microsoft holds roughly a 27% stake in the OpenAI Group public benefit corporation, which remains controlled by OpenAI's nonprofit. Microsoft is a major shareholder and cloud partner with model access through 2032, but it does not control the company.

Is OpenAI profitable?

Not as of the latest reported figures. OpenAI generated around $13 billion in revenue in 2025 but ran large losses, driven by the cost of training and serving its models. Its path to profitability depends on monetizing usage more effectively and on the cost of running models falling over time.

Share this article: