OpenAI has launched a program called Guaranteed Capacity that allows business customers to reserve access to its AI services for one, two, or three years. Longer commitments come with larger discounts. The program is available until OpenAI's current allocation is sold out, after which the company plans to offer it again.
On its face, this looks like a straightforward commercial offering. Underneath, it tells you something important about where the AI market is heading.
The shortage is real. Token demand, which is the volume of text and queries processed by AI systems across the industry, jumped from roughly 6 million units per minute in October 2025 to around 15 billion by March 2026. Every major cloud provider reported the same thing in recent earnings calls: demand for AI computing is outpacing what they can supply. Google's CEO admitted the company left cloud revenue on the table because of capacity limits. Meta's CFO said they had consistently underestimated their own compute needs. OpenAI scaled back its video tool Sora earlier this year to redirect capacity to core services.
This is the backdrop for Guaranteed Capacity. OpenAI spent the last 18 months locking in its own supply, inking deals across chipmakers and data center operators that total well over a trillion dollars in commitments. It is now selling the demand side of that equation back to its customers.
For businesses already using OpenAI in production, the program is essentially a reservation system. You trade a multi-year financial commitment for the right not to be slowed down or cut off when your business needs the service most. Think of a retailer whose AI-powered customer service runs at high load during a sale weekend, or an insurance firm whose claims-processing tool needs to handle a surge after a weather event. In those moments, having reserved capacity matters.
OpenAI generated $13.1 billion in revenue in 2025, against a $10 billion target. Its 2030 revenue projection is $280 billion, split roughly between consumer and enterprise customers. Enterprise customers currently account for around 40% of OpenAI's revenue, and that share is expected to hit 50% by year's end. Guaranteed Capacity is a direct tool for growing and locking in that enterprise base before an IPO that could come as soon as this year.
There are two things to weigh before signing. First, this is a spend commitment, not a price freeze on specific models. The terms allow customers to draw down their reserved budget across OpenAI's product portfolio, which gives flexibility, but also means the exact service you need today may cost differently against that budget tomorrow. Second, signing a three-year deal with a single provider is a concentration risk. OpenAI is free to source its own computing from multiple vendors now; customers probably want the same flexibility with their AI providers.
The broader signal here is structural. AI computing is beginning to behave like a utility: scarce, essential, and worth contracting in advance. Companies that treat it as a pay-as-you-go commodity may find themselves behind those that planned ahead. That said, the right move for most businesses is not to rush into a long-term OpenAI commitment today, but to audit which AI workflows are genuinely business-critical and ask whether those are the ones that deserve protected access.