OpenAI is surfacing a 5 hour usage limit for Plus and Business Standard

The conversation around AI service limits reveals something fundamental about how computing infrastructure operates at scale. When a company offers unlimited or high-capacity access to a computational resource, they are making a choice about how to allocate finite hardware. GPUs, the specialized processors that run large language models, are expensive equipment that consume significant electricity. Every hour of model inference, whether you use the full capacity or not, represents a real economic cost. The question of how to price and ration access to this resource is not new. It echoes decisions made decades ago about cloud computing, premium internet speeds, and database query allowances. Companies must decide whether to charge by actual usage, offer tiered subscriptions with soft caps, hard limits, or unlimited tiers that occasionally throttle during peak demand. The history of AI service pricing has tracked alongside broader changes in how software companies monetize computational access. Early generative AI services were relatively unconstrained, partly because demand was uncertain and companies were still learning what customers actually needed. As adoption accelerated and inference costs became clearer, the economics forced a reckoning. The question became unavoidable: should a business model rely on average users consuming a moderate amount, with pricing that assumes that distribution? Or should it allow heavy users to access more, knowing some will exhaust capacity? The threshold at which a user becomes expensive to serve varies by the underlying model, the efficiency of the serving infrastructure, and the time of day when demand clusters. Usage caps serve multiple purposes beyond simple cost control. They influence user behavior. A five hour monthly limit shapes how people think about when to deploy a tool, whether to use it for exploration or only critical work, and what alternatives to evaluate. Caps also provide a predictability signal to infrastructure planning. If the company knows subscribers will not exceed a certain threshold, they can forecast hardware needs more precisely. Removing caps entirely sends a different message: we believe current demand can be absorbed, or we are willing to absorb it as a growth investment. Reintroducing caps reverses that signal. It typically means either demand exceeded infrastructure capacity, or the financial model revealed that unlimited consumption was unsustainable at the subscription price point. The pattern of removing and reinstating limits reflects a genuine tension in how AI service companies operate. Subscribers want predictability and access. Companies want sustainable unit economics. These goals sometimes align and sometimes conflict. When a cap reappears after removal, it usually indicates the company has gathered data about actual usage patterns and discovered those patterns differ from what was forecast. Maybe heavy users consume far more than expected. Maybe peak simultaneous u