Can cheaper AI finally make the business work?

Picture Credit: Tech Crunch

OpenAI has become the face of the artificial intelligence revolution, but behind its explosive growth lies an uncomfortable reality, the company is still losing staggering amounts of money. Despite annualized revenue climbing to around $25 billion, analysts estimate OpenAI could lose as much as $33 billion in 2026. Even Sam Altman has admitted that some of the company's highest-paying ChatGPT Pro subscribers cost more to serve than they generate in revenue. The reason isn't a lack of customers, it's the enormous cost of running AI at scale.

The biggest expense isn't training AI models, but inference, which is the computing power required every time a user asks ChatGPT a question. With hundreds of millions of users generating billions of prompts every week, each response consumes expensive GPU resources housed in massive data centers. Heavy users who rely on ChatGPT for coding, research, writing, and productivity can cost OpenAI more in computing than their monthly subscription fee brings in. In a fiercely competitive market, simply raising prices isn't an option, with rivals like Google Gemini and Anthropic continuing to cut costs and offer powerful alternatives.

Instead, OpenAI appears to be attacking the problem from the technology side. Over the past few months, the company has rolled out a series of major upgrades designed not only to improve performance but also to lower operating costs. Its latest AI models reportedly deliver comparable capabilities at a fraction of the computational expense, while Codex, its AI coding assistant, expands ChatGPT beyond conversations into software development and automation. At the same time, OpenAI is developing its own custom AI inference chip, codenamed Jalapeño, in partnership with Broadcom, a move aimed at reducing its dependence on Nvidia's costly hardware.

If successful, custom silicon could dramatically change OpenAI's economics. Early reports suggest the new chips could reduce the cost of generating AI responses by as much as 50%, potentially saving billions of dollars annually. By owning more of the AI technology stack, from models to hardware, the company hopes to escape the expensive reliance on third-party infrastructure that has weighed heavily on its finances since ChatGPT's launch.

The challenge, however, is that cheaper AI often leads to more AI usage. As costs fall, users tend to ask more questions, run more complex tasks, and increasingly rely on AI agents to complete work. Economists call this Jevons Paradox, greater efficiency driving higher overall consumption. For OpenAI, profitability may ultimately depend not only on making AI cheaper, but on staying ahead in a race where every improvement encourages the world to use even more intelligence than before.