OpenAI sells guaranteed compute as AI capacity becomes scarce

OpenAI is turning scarce compute into a product. The company has launched Guaranteed Capacity, giving customers a way to secure long-term access to compute for their own AI systems through one-, two- or three-year commitments.

This is not just a technical product note. It is a signal that large AI providers are moving customers from flexible consumption toward capacity contracts. For executives, that changes the buying decision. AI can no longer be treated only as another SaaS subscription. Capacity, price, priority and exit rights are becoming board-level infrastructure questions.

CNBC reported the launch Tuesday evening. According to CNBC, OpenAI says customers can choose between one-, two- and three-year commitments, with discounts that increase with the length of the commitment. OpenAI CEO Sam Altman also wrote on X that customers are increasingly asking for certainty on capacity, and that the world may remain capacity-constrained for some time as models improve.

OpenAI will offer Guaranteed Capacity until its current allocation is sold out, with later rounds planned. Altman also said the company must keep enough capacity available for its own products, including ChatGPT and Codex. That is the key point: OpenAI is not only selling access to a model. It is selling a place in the queue for the infrastructure behind the model.

From model choice to capacity governance

For CIOs and CFOs, this is a different decision from picking the best model on a benchmark table. A company that commits to guaranteed AI capacity buys predictability. It also takes a position on one supplier's roadmap, pricing model and technical control surface.

That may be rational for banks, software companies, large analytics firms and organisations building AI into core processes with strict latency or availability demands. If an agent is changing code, answering customers, analysing transactions or supporting operations in real time, capacity is not decoration. It is part of production risk.

But the contract has to be treated as infrastructure, not as an AI subscription. The questions are concrete. What happens if capacity is not delivered? Which models and endpoints are covered? Can capacity move between ChatGPT Enterprise, the API, Codex and future agent products? What happens if EU regulatory requirements change? What rights does the customer have if OpenAI changes security terms, pricing or product priorities?

Scarcity is changing procurement

OpenAI has previously told investors it is targeting roughly $600 billion in total compute spend by 2030, according to CNBC. That figure explains why vendors want customers in more predictable contracts. Data centres, power, GPUs, networking and operations have to be financed before demand is fully proven.

The logic resembles cloud commitments, but with more uncertainty. In cloud, companies could often buy flexibility and optimise later. In frontier AI, the bottleneck is increasingly physical capacity and prioritised access. When a supplier says capacity may sell out, customers have to decide whether to pay for certainty or live closer to a spot market.

For leaders, this points to three practical moves.

First, AI capacity needs to be part of sourcing strategy. Not as a minor line item in the IT budget, but as a risk category alongside cloud, network and other critical suppliers.

Second, legal and technical governance must be joined up. A discount for a three-year commitment is not worth much if the organisation loses negotiating power, data control or the ability to move workloads.

Third, boards should ask for scenarios. What happens if one AI supplier becomes more expensive, slower or harder to use under regulation? Which workloads can run with another model provider? Which can run locally? Which should not be tied to a frontier model at all?

Not just an OpenAI story

This story does not stand alone. AI capacity is already becoming its own financial and strategic category. Earlier Tuesday, hogby.ai covered plans for futures contracts tied to AI compute. Guaranteed Capacity shows the other side of the same shift: suppliers want committed demand, while customers want guaranteed access.

Companies that only look at model names will enter the decision too late. The real competition is moving to capacity, contracts, data flow, security, agent governance and exit planning.

For organisations putting AI into operations, the lesson is blunt: do not wait until a critical agent lacks capacity before procurement, IT, security and finance agree on the rules.

Sources and media

• Primary source: CNBC, “OpenAI announces new Guaranteed Capacity offering for customers to secure compute”, published May 19, 2026 at 21:56 UTC: https://www.cnbc.com/2026/05/19/openai-announces-new-guaranteed-capacity-offering-for-customers-to-secure-compute.html
• Official product page: OpenAI Guaranteed Capacity: https://openai.com/business/guaranteed-capacity/
• Sam Altman/X: comment on customers asking for more certainty on capacity: https://x.com/sama/status/1986514377470845007?s=20
• CNBC article photo: Carlos Barria / Reuters, used as editorial reference. Not rehosted.
• Thumbnail: OpenAI Image 2 / hogby.ai, generated as an editorial illustration of guaranteed AI capacity and supplier risk.