OpenAI CEO Sam Altman announced plans to add tens of thousands of GPUs next week with the release of GPT-4.5. This model is described as the largest and most powerful to date.

The rollout will commence with the plus tier of the service. Altman stated that the model will be both “giant” and “expensive,” indicating OpenAI’s significant growth and current GPU limitations.

Altman’s statement reaffirms what many have already observed—OpenAI’s rapid expansion continues, but the sheer cost and technical burden cannot be ignored. A model of this size requires an enormous number of GPUs, something that has become a recurring concern for large-scale AI deployments. The mention of “giant” hints at both the model’s complexity and, more importantly, the increased resource demand that will come with it.

This comes at a time when demand for high-end chips is nowhere near slowing down. Nvidia remains the dominant supplier, and the industry has already been dealing with shortages that have, at times, constrained AI companies. The upcoming expansion suggests that OpenAI is securing hardware at a faster pace, which could be read as confirmation that supply issues are easing—at least for them. Whether this extends to others in the space is another matter entirely.

Then there is the question of operating costs. “Expensive” is not just a throwaway comment; these systems bring enormous electricity consumption and ongoing expenditure. If OpenAI is willing to publicly acknowledge these costs, it underlines the continued financial weight these models carry. Cloud providers supplying computing power to AI firms have been adjusting their pricing accordingly, which means these costs are unlikely to stabilise in the near term.

For those tracking where high-compute AI is headed, these developments bring both opportunities and fresh challenges. Pricing strategies may shift in response to the increased needs of OpenAI, and any supply chain disruptions would send ripple effects elsewhere. The launch taking place through the plus-tier subscription first is also revealing—OpenAI is prioritising those users, possibly as a way to better control demand or gather early insights before wider availability.

Even beyond OpenAI, the pressure to keep pace with these advancements will weigh on competitors. Scaling up infrastructure is not an overnight process, and any delay in acquiring more GPUs could mean a widening gap between market leaders and those trying to catch up. Some firms may respond by refining their own models to operate with fewer resources, but how that plays out will depend on whether they can maintain quality while doing so.

This is the environment that will shape decisions in the coming weeks. Every adjustment in compute availability, subscription access, and cost structure feeds into broader movements, all of which need to be watched carefully.