OpenAI has introduced a Guaranteed Capacity program that lets customers reserve compute for one, two, or three years, with discounts that scale up with the length of the commitment. The launch, announced on Tuesday, marks OpenAI’s first formal attempt to sell cloud-style reserved infrastructure directly to enterprise buyers.
The structure will feel immediately familiar to anyone who has negotiated AWS Reserved Instances or Google Committed Use Discounts. Customers trade pricing certainty for term commitments; the vendor gains revenue predictability and demand signal for capacity planning. OpenAI is applying that same logic to AI inference, a market that did not exist at scale five years ago but that now sits at the center of every enterprise IT conversation.
CEO Sam Altman framed the launch in supply terms on X, saying customers want certainty on capacity and that he expects the world to remain capacity-constrained for some time as models get better. He described the program as a win for both sides and said OpenAI will offer Guaranteed Capacity until its current allocation sells out, then replenish and offer it again.
The announcement comes as OpenAI is targeting roughly $600 billion in total compute spend by 2030, a figure CNBC reported earlier this year. The company is valued above $850 billion by private investors and is preparing for a potential IPO, possibly this year. Locking in enterprise revenue streams now strengthens the unit-economics story it will need to tell public-market investors.
The GPU supply tightness Altman references is real and measurable. Frontier inference throughput has not kept pace with demand growth from agentic workloads, which consume far more tokens per task than simple chat completions. Enterprise teams running coding agents or document-processing pipelines are already hitting rate limits or absorbing volatile spot-market pricing. A multi-year reservation at a fixed discount gives procurement teams a number they can put into a budget and defend.
The competitive framing matters here. Both Anthropic and Google offer enterprise API agreements, but neither has publicly announced a tiered reserved-capacity product with the explicit 1/2/3-year structure OpenAI is advertising. If OpenAI sells out its first allocation quickly, it serves as a demand signal that will pressure competitors to follow. If it sells slowly, that tells a different story about whether enterprises are ready to make multi-year bets on a single model provider.
Altman also said OpenAI will preserve enough capacity for its own consumer and developer products, including ChatGPT and the Codex coding assistant. That caveat is worth watching: a company selling capacity commitments to external buyers while also trying to scale its own product line is managing two competing claims on the same infrastructure.
OpenAI has not disclosed the specific discount tiers, the volume thresholds required to access the program, or how much total capacity is available in the first allocation. Those details will determine whether this is a product for large enterprises spending seven figures annually or something accessible to mid-market teams.
Enterprise procurement teams currently evaluating AI infrastructure contracts should ask their OpenAI account managers for the Guaranteed Capacity terms before signing any spot-rate agreements, because the discount structure likely changes the total-cost comparison against Azure OpenAI Service, AWS Bedrock, or Google Vertex commitments they may already be modeling.
CNBC reported the launch of OpenAI’s Guaranteed Capacity program on May 19, 2026, in a story by Ashley Capoot.
