
OpenAI’s biggest problem right now may not be whether people want its newest model. It may be whether the company can afford to give everyone as much of it as they want.
The company is pausing new subscriptions to its $200-a-month ChatGPT Pro plan after what OpenAI staff described as unusually heavy demand for GPT-6 Astra. The update, first shared by OpenAI product lead Thibault Sottiaux and tracked by Techmeme, says existing Pro users will keep access, while other ChatGPT plans and the API remain available. In plain terms, OpenAI is trying to protect the experience for people already inside its most expensive consumer tier without shutting the door on every other product.
That matters because Astra was not launched as a small model refresh. OpenAI positioned it as a major step forward for coding, business work and difficult reasoning, and the company has already had to manage access carefully after a rocky rollout. The firm has also been talking more openly about safety and capacity limits, including in its recent note on pacing frontier model development.
The expensive side of popular AI
The Pro pause tells a bigger story about AI economics. For years, many users got used to treating powerful AI models like normal software; pay a subscription, type as much as possible and expect the service to be there. But frontier AI is closer to an energy and infrastructure business than a simple app business. Every heavy prompt touches scarce chips, power, data centre space and internal routing decisions.
This is why OpenAI’s move sits naturally beside the wider compute race. Google is committing billions to AI data centres in Europe, Oracle’s cloud infrastructure revenue is surging on AI demand and almost every serious AI lab is trying to lock down long-term compute. TechBooky recently looked at how Google’s $15B Finland AI bet shows that the AI race is also becoming an energy race.
For users, the immediate effect is simple. If you already have ChatGPT Pro, you are not being pushed out. If you wanted to newly subscribe mainly to get Astra, you may have to wait. For OpenAI, the message is more delicate. Demand is strong, but the company now has to show that its infrastructure can keep up with the expectations it has created.
The awkward part is that the better these models become, the more people use them for longer and more complex tasks. That is good for product adoption, but it is brutal for capacity planning. The next phase of the AI boom may therefore be less about who has the flashiest demo and more about who can keep the model available when millions of users actually try to work with it.






