OpenAI is increasingly positioning its business model around the principle that investments and revenues must grow in line with the value generated by artificial intelligence.
In a blog post, CFO Sarah Friar describes how AI within OpenAI has shifted from an experimental tool to a form of infrastructure. Costs and revenues are directly linked to actual use and added productivity.
According to OpenAI, ChatGPT quickly became an integral part of both personal and professional workflows. What started as a tool to support studying, writing, and planning is now used by users on a structural basis in software development, marketing, financial analysis, and management decision-making. According to the company, this broader use forms the basis for an economic model that differs from traditional software licensing.
Instead of fixed prices per user, OpenAI has opted for a combination of subscriptions and usage-based rates. Paid ChatGPT models are available for consumers and teams. Developers and companies, on the other hand, pay via APIs according to the amount of computing power the models use. According to OpenAI, this means that costs are aligned with the value that organizations actually derive from intelligence, similar to cloud infrastructure.
Advertisements as support
The company also increasingly sees ChatGPT as a platform that helps users make decisions, not just obtain information. This opens up opportunities for new forms of commerce and advertising, provided they are relevant to specific decision-making moments and are demonstrably relevant. Advertisements are presented as a supporting element in the decision-making process, not as a generic source of income.
Additional context comes from a report by Neowin, which writes that OpenAI is underpinning this strategy with solid revenue and capacity growth. According to that publication, OpenAI’s revenue tripled last year, reaching more than $20 billion in annual recurring revenue. That growth is said to run parallel to a strong expansion of available computing power since the introduction of ChatGPT.
Over a three-year period, total computing capacity is said to have grown from approximately 0.2 gigawatts in 2023 to approximately 1.9 gigawatts in 2025, an increase of approximately 9.5 times. Friar argues that faster availability of computing power in previous years could have further accelerated adoption and monetization, positioning infrastructure as a direct driver of growth rather than a cost center.
The same report also describes a strategic shift in OpenAI’s infrastructure choices. Whereas in 2023 and 2024 the company was completely dependent on Microsoft, it now works with multiple infrastructure partners. This diversification should make it easier to plan, finance, and scale capacity as demand for AI services continues to grow.
More cost-efficient infrastructure
In addition, OpenAI differentiates its use of hardware based on performance and efficiency. Training of new frontier models takes place on the most advanced systems, while large-scale inference workloads are moved to more cost-efficient infrastructure. This approach is in line with the broader goal of using computing power as purposefully as possible.
Together, the OpenAI blog and the additional figures paint a picture of a company that has consciously opted for a capital-intensive strategy, in which investments in computing, model development, product adoption, and revenue reinforce each other. OpenAI wants to show that scaling up does not lead to unfocused spending, but is intended to position artificial intelligence as a structural infrastructure layer for the global economy.
Read also: OpenAI’s business model isn’t working as bankruptcy looms