OpenAI and Oracle recently made headlines with a surprising five-year, $300 billion agreement, a deal that significantly boosted Oracle’s stock and underscored its continued, if often overlooked, prominence in the AI infrastructure landscape. Despite its legacy status, Oracle’s deep-seated capabilities in providing extreme-scale and high-performance core infrastructure proved instrumental in securing this massive partnership, drawing comparisons to its past collaborations with hyperscalers and its role in powering TikTok’s substantial U.S. operations.

For OpenAI, this agreement is highly revealing, even with a lack of specific details. It highlights the startup’s enormous appetite for compute resources, indicating a strategic move to build what Gartner’s Vice President Chirag Dekate describes as “one of the most comprehensive global AI supercomputing foundations for extreme scale, inference scaling where appropriate.” This diversification across multiple cloud providers is crucial for OpenAI, mitigating risks and granting a significant scaling advantage over its competitors, setting a potential benchmark for a robust model ecosystem.

However, the landmark deal also brings to light critical unanswered questions, particularly concerning payment and power. OpenAI has committed to an annual expenditure of approximately $60 billion for compute from Oracle, alongside a $10 billion investment in custom AI chips with Broadcom. This comes as the company’s annual recurring revenue reached $10 billion, up from $5.5 billion the previous year, yet it continues to burn through billions of dollars in cash annually. This financial dynamic raises questions about the long-term sustainability and funding mechanisms for such extensive infrastructure commitments.

Equally pressing is the challenge of powering this immense compute capacity. The energy demands of data centers are skyrocketing, with projections suggesting they could consume 14% of all electricity in the U.S. by 2040. While industry observers anticipate a near-term boost for natural gas, alternatives like solar, batteries, and nuclear power are increasingly being explored by tech companies to ensure a reliable and sustainable energy supply. Compute has always been a bottleneck for AI development, leading investors to acquire thousands of Nvidia chips to guarantee access for their startups; yet, without adequate power, these resources are effectively useless.

Unlike some of its tech giants like Google, Meta, or Amazon, OpenAI has remained relatively quiet on directly investing in energy infrastructure. Nevertheless, CEO Sam Altman has strategically backed several energy startups, including Oklo, Helion, and Exowatt, aligning with OpenAI’s future power requirements. This indirect approach allows OpenAI to remain “asset light,” offloading the complexities and capital expenditure of physical infrastructure to partners like Oracle, who possess extensive experience in this domain. This strategy is likely to please investors, helping to maintain OpenAI’s valuation closer to other software-centric AI startups rather than traditional tech firms burdened by costly infrastructure assets, even as it navigates the immense energy needs of a 4.5-gigawatt compute deal.