ChatGPT Subscribers Nearly Tripled to 15.5 Million in 2024
The Takeaway
• Five percent of weekly ChatGPT users pay for a subscription
• Business usage of OpenAI’s models increased sevenfold in 2024
• Its CEO said shunning open-source software put the company on the ‘wrong side of history’
Paid subscribers to ChatGPT nearly tripled to 15.5 million last year from 5.8 million a year earlier, OpenAI recently told some shareholders, despite competition from chatbots made by Google, Anthropic and Meta Platforms.
Based on what OpenAI charges for the chatbot subscriptions, the increase means ChatGPT was likely generating at least $4 billion in annualized revenue around the end of last year, or $333 million per month, just two years after its launch.
Separately, usage of OpenAI’s application programming interface—which gives companies such as Salesforce and T-Mobile access to its AI models—increased seven times, the company told the shareholders.
The growth could help explain why the valuation of OpenAI in an equity financing now under discussion is $260 billion, up 73% from the value at which OpenAI paid to buy employees’ shares in a recent tender offer. SoftBank is discussing being a lead investor in a $40 billion equity financing at the new valuation.
Meanwhile, OpenAI is negotiating to reduce Microsoft’s 20% cut of its revenue as they rework the terms of their contract, according to two people who have spoken to OpenAI leaders about it. Such a change could further boost OpenAI’s value.
The revenue sharing is part of a deal in which Microsoft funded the startup and provided it with servers to run its technology. In return, OpenAI has gotten a 20% cut of the revenue Microsoft generates from selling OpenAI models to its cloud customers.
OpenAI’s total 2024 revenue could not be learned, but the company last fall projected revenue of $4 billion, mostly from subscriptions to ChatGPT, and a loss of $5 billion--excluding stock based compensation--in part due to high computing costs. For the second half of that year, it set a goal of generating at least $3.5 billion in annualized revenue, or $290 million per month, from the chatbot by the end of the year, and it beat that by 14%. OpenAI’s API business generates the rest of the company’s revenue.
Five Percent Conversion
ChatGPT subscriptions start at $20 a month, and at least 1 million subscribers were part of plans that charge $25 per month or higher as of September, based on earlier private disclosures by OpenAI.
Far more people use ChatGPT without paying anything. Less than 5% of ChatGPT’s weekly active users were paying subscribers as of the end of 2024, the company told some shareholders. During 2024, OpenAI’s weekly active users more than tripled to about 350 million from roughly 110 million, OpenAI told the shareholders.
But OpenAI is counting on growth from higher-priced subscriptions for versions of its chatbot powered by more-advanced models. Seven weeks ago, it launched a $200-a-month Pro tier of the chatbot, or 10 times more than regular subscriptions, and it is already generating at least $300 million in annualized revenue from that business. ChatGPT Pro appeals to software engineers and other researchers who use it for their work.
The company has considered even higher prices of $2,000 per subscriber per month to use even more-advanced models. And it’s developing an AI coding product that aims to handle the work of a senior software engineer, for which it could charge a premium.
‘Wrong Side of History’
OpenAI’s API business pales in comparison with what ChatGPT generates, and it faces a lot more competition from Anthropic as well as open-source options like Llama and, more recently, DeepSeek, which could squeeze prices in the API market. That’s because the DeepSeek reasoning model, R1, is considered comparable in quality to OpenAI’s o1 but it’s much cheaper to operate. It’s already attracting business users including ZoomInfo and Perplexity.
However, OpenAI seems to be making an effort to release cheaper reasoning models; its new o3-mini AI is better at coding and math than o1 but even cheaper to access through an API than GPT-4o, OpenAI’s most popular nonreasoning model.
On Friday, OpenAI CEO Sam Altman said on Reddit that the company had been on the “wrong side of history” by not releasing more free, open-source models and that it was discussing doing so. It isn’t the company’s “current highest priority,” he said.
In 2024, OpenAI’s API grew rapidly in terms of how much content, known as tokens, it was analyzing for customers or generating in its answers to their queries. (A token can be a word or part of a word.) OpenAI was processing and generating around 1.4 billion tokens per minute through its API at the end of the year, up from about 200 million at the beginning of the year, the company told some shareholders.
It’s tough to estimate OpenAI’s current revenue from the API business based on those figures, though last fall the company estimated it would generate about $1 billion from the business.
As a proxy, customers of Braintrust, which says it helps more than 10,000 companies including Instacart, Zapier and Airtable to evaluate AI models, used GPT-4o about 70% of the time they used an OpenAI model and GPT-4o mini about 27% of the time, with a small amount of usage of other OpenAI models like o1. Based on those proportions and the prices OpenAI lists for GPT-4o and GPT-4o mini, OpenAI’s API business could be generating $270 million in revenue per month, or $3.2 billion on an annual basis.
That figure would put OpenAI far ahead of its projection last fall that it would generate $2 billion from API sales in 2025.