OpenAI hits $25 billion in revenue with an IPO possibly on the way

Two years ago, a $25 billion annualized revenue figure for an AI lab would have sounded like a projection from an overly optimistic pitch deck. OpenAI has now crossed that number for real. The company, which launched ChatGPT to the public in late 2022, has grown its revenue at a pace that most software businesses take a decade or more to approach. It is also, according to people familiar with the matter, beginning to explore what a public listing in late 2026 might look like.

How OpenAI got to $25 billion

The bulk of OpenAI's revenue comes from two sources: consumer subscriptions to ChatGPT Plus and its higher-tier plans, and API access sold to businesses building products on top of GPT-4o and its successor models. ChatGPT Plus costs $20 per month. The Pro tier, aimed at heavy users, runs $200 per month. Enterprise contracts, which are priced separately and often in the millions annually for large deployments, have grown significantly as companies move AI from pilot programs into actual production systems.

OpenAI ended 2024 with roughly $3.7 billion in full-year revenue. Getting from there to $25 billion annualized in roughly 15 months is a steep climb, and it required both a rapid expansion of paying users and a series of enterprise deals that the company has not fully disclosed publicly. The $25 billion figure refers to annualized revenue, meaning the current monthly or quarterly run rate multiplied out, rather than total revenue collected over a full fiscal year.

Anthropic is not far behind

Anthropic, the San Francisco lab founded in 2021 by former OpenAI employees including Dario and Daniela Amodei, is approaching $19 billion in annualized revenue. That number puts it well ahead of where most AI companies sit, and well behind OpenAI, which has the consumer brand recognition that Anthropic largely lacks. Anthropic's Claude models have found a strong foothold in enterprise software, legal tech, and developer tooling, where the preference tends to be for models that follow instructions carefully rather than ones that generate the most vivid responses.

Amazon has committed up to $4 billion to Anthropic, and Google has invested separately. That backing has given Anthropic the runway to keep training large models without needing to chase consumer revenue as aggressively as OpenAI does. Whether that strategy holds as the gap between the two companies widens in raw revenue terms is a question the Amodeis will have to answer publicly at some point.

What an OpenAI IPO would actually involve

OpenAI's corporate structure has been one of the more unusual arrangements in recent tech history. It was originally set up as a nonprofit with a capped-profit subsidiary, which meant that investors could earn returns up to a certain multiple before excess profits flowed back to the nonprofit. Sam Altman, the CEO, held no equity under that original structure. In late 2024, the company began restructuring toward a more conventional for-profit model, which is a prerequisite for going public.

A public offering in late 2026 would require OpenAI to file a registration statement with the SEC, likely an S-1, at least several months in advance. That document would, for the first time, make the company's full financials public, including its cost of revenue, which is understood to be extremely high given the compute required to train and serve its models. OpenAI's most recent private valuation was $157 billion, set during a funding round in late 2024. A public market valuation could be higher or lower depending on how investors price the company's spending levels against its revenue trajectory.

What the revenue numbers say about the AI market

The combined revenue of OpenAI and Anthropic, at roughly $44 billion annualized, is a concrete measure of how much money companies and consumers are now spending on AI model access. That figure does not include revenue from Google's Gemini products, Meta's AI offerings, or the dozens of smaller labs. It also does not include the hardware side of the market, where Nvidia's data center revenue hit $47.5 billion in a single quarter in early 2025.

The speed of growth has also changed the competition. OpenAI and Anthropic are both releasing new model versions every few months, and the pricing for API access has dropped substantially over the past two years. GPT-3.5 cost $0.002 per 1,000 tokens when it launched. Comparable capability today costs a fraction of that. Lower prices have pulled in more API users, which drives revenue volume even as per-unit margins compress. It is a pattern that looks more like cloud infrastructure pricing than traditional software.

If OpenAI does go public in late 2026, it will likely be one of the most watched tech IPOs since Meta listed in 2012. The S-1, whenever it arrives, will finally answer the question that private market investors have been circling around for three years: whether the revenue growth is outrunning the compute costs by enough to make the business structurally profitable at scale.