OpenAI offers 1–3 year capacity

OpenAI has started selling reserved AI compute in multi-year contracts, moving a core part of its business closer to infrastructure procurement than ordinary software licensing. The new program, called Guaranteed Capacity, lets eligible customers commit to one-, two- or three-year terms for access to OpenAI compute across supported cloud providers and model families. OpenAI says the product is aimed at critical workloads, production systems and customer-facing agents. The company says discounts increase with the size and length of the annual commitment. ### What exactly is OpenAI selling here? OpenAI says Guaranteed Capacity gives customers “certainty of access to compute” rather than a promise tied to one fixed model snapshot. The company says customers can draw down their commitment across its portfolio of products, and use guaranteed spend allocations across supported cloud providers and model families. That makes the offer broader than a standard API reservation for a single endpoint. (openai.com) The sales form published by OpenAI shows the target buyer: companies running API production environments, internal workflows, Codex, ChatGPT or other customer-facing systems. The form asks prospective customers about endpoint and data-residency needs in regions including the United States, European Union, United Kingdom, Middle East, Japan, Korea, India, ANZ and ASEAN, as well as estimated monthly usage in dollars. (openai.com) ### How is this different from OpenAI’s older enterprise capacity products? OpenAI already offered Scale Tier, an enterprise API product that lets customers buy token throughput upfront for a minimum of 30 days on specific model snapshots, with published latency and uptime service levels. Scale Tier is narrower: it is tied to token units per minute and is designed around predictable performance for a chosen model. (openai.com) Guaranteed Capacity is framed differently. OpenAI says it covers long-term access to compute for “products, agents, and customer workflows that matter most,” and allows customers to spend against the commitment across its broader product portfolio. In practice, that gives large buyers a way to plan longer than the 30-day minimum attached to Scale Tier. ### Why is OpenAI offering multi-year reservations now? (openai.com) Sam Altman said demand from customers is the immediate reason. “Customers are increasingly asking us for certainty on capacity,” Altman wrote on X, according to CNBC. He added that “as models get better, we expect that the world will be capacity-constrained for some time,” and said the program would help OpenAI plan ahead. (openai.com) OpenAI has also been expanding its infrastructure build-out. In an April 29 post, the company said its Stargate effort had already surpassed its January 2025 goal of securing 10 gigawatts of AI infrastructure in the United States by 2029, with more than 3 gigawatts added in the prior 90 days. OpenAI said then that “compute is the critical input” for training models, serving them reliably and lowering costs over time. (cnbc.com) ### What does this mean for enterprise buyers? Large companies buying AI at scale often care less about list price than about whether capacity will be available when a product launches or usage spikes. OpenAI is pitching the new program as a planning tool for forecast demand, product expansion and multi-year AI adoption plans. The company says the offering is for organizations that want to “scale confidently” without worrying whether infrastructure can keep up. (openai.com) CNBC reported that OpenAI will offer Guaranteed Capacity until it sells out of its current allocation, with plans to offer it again later. Altman also said OpenAI would leave enough capacity for products such as ChatGPT and Codex. Those details suggest the company is rationing a defined pool rather than opening unlimited reservations. (openai.com) ### How does this fit into OpenAI’s broader business model? CNBC reported that OpenAI has told investors it is targeting roughly $600 billion in total compute spend by 2030. The outlet also said the company had signed several multi-billion-dollar compute deals late last year, raising questions among investors about how it would finance a large infrastructure build-out. (cnbc.com) OpenAI’s own infrastructure post described compute as the center of its “AI flywheel,” saying more capacity enables better models, more usage and more revenue that can be reinvested in infrastructure. Guaranteed Capacity adds a commercial layer to that strategy by giving OpenAI forward demand commitments from enterprise customers. That last point is an inference from OpenAI’s published terms and Altman’s comments about planning ahead. (cnbc.com) OpenAI is currently taking inquiries through its Guaranteed Capacity request form, where prospective customers are asked to specify workload type, geography, deployment needs and estimated monthly usage. The company’s public page says customers can choose one-, two- or three-year commitments, with discounts that increase based on annual commitment. (openai.com 1) (openai.com 2)