OpenAI now sells 3-year contracts to lock in scarce AI compute

OpenAI's launch of "Guaranteed Capacity" introduces long-term, multi-billion dollar contracts to the AI model market, a move that could secure hundreds of billions in future revenue and stabilize its business model ahead of a potential IPO this year. The program, announced May 19, marks the first time a major AI lab has offered multi-year subscriptions for computing power.

"Customers are increasingly asking us for certainty on capacity... we expect that the world will be capacity-constrained for some time," OpenAI CEO Sam Altman said in a post on X, calling the program a "big win-win."

The new offering allows customers to reserve access to OpenAI's models with one, two, or three-year commitments, receiving larger discounts for longer terms. This mirrors the long-term agreements (LTAs) used by upstream players like Nvidia, which has amassed $95.2 billion in procurement commitments to secure its own supply chain.

For OpenAI, valued at over $850 billion, these contracts provide predictable revenue streams crucial for a public listing. For customers, it’s a hedge against a projected compute shortage that J.P. Morgan estimates could reach a 444,000 wafer-per-month deficit for high-bandwidth memory between 2026 and 2030.

A Playbook Borrowed From the Supply Chain

OpenAI's strategy is a direct reflection of the pressures in the AI hardware market. The entire industry is grappling with scarcity, forcing companies to abandon just-in-time purchasing for long-term, high-value contracts. Nvidia's purchase commitments surged 89% in a single quarter, while Broadcom announced it has locked in the supply needed to hit a $100 billion AI chip revenue target, more than triple its current run rate. AMD's commitments have also more than doubled to over $21 billion.

This frantic race to secure supply is based on a shared belief that demand for AI will continue to outstrip production capacity for years. TSMC recently revised its forecast for the total semiconductor market to $1.5 trillion, up from a prior $1 trillion, signaling a massive expansion driven by AI. By offering guaranteed access, OpenAI is extending this logic from the hardware supply chain to the end user, effectively asking customers to pay a premium for certainty.

From Volatile Usage to Bankable Revenue

The shift to a contract-based model is a critical step in maturing OpenAI's financials for public markets. Predictable, recurring revenue from multi-year deals is valued more highly by investors than volatile, usage-based API calls. This could help justify the company's massive spending, with a reported target of $600 billion in total compute investment by 2030.

However, the model is not without risk. During a DRAM shortage in 2017, similar long-term contracts were signed, only to be abandoned when demand cooled and prices fell over 40 percent in a matter of months. Analysts note that the key difference this time lies in the contract terms. According to a statement from Sandisk's CEO on a recent earnings call, customers are now required to post billions of dollars in collateral, which is forfeited if they fail to meet their purchasing obligations. While OpenAI has not disclosed its specific contract terms, the success of its "Guaranteed Capacity" program will depend on whether its agreements carry similarly high costs for default.

This article is for informational purposes only and does not constitute investment advice.