AI is expensive. Oracle will provide $30 billion worth of OpenAI servers per year

At the end of June, Oracle filed its financial statements, which stated that the company had signed a new $30 billion per year contract. For the sake of understanding, Oracle provided a total of $24.5 billion in cloud services in fiscal year 2025. This deal will make Oracle founder and CEO Larry Ellison the second richest person in the world. But until recently, there was no official information on who became the customer of these facilities. And now it has officially become known that the new facilities will serve the needs of OpenAI.

What is stipulated by the agreement

• 4.5 GW of capacity is about a quarter of all US data centres, which is equivalent to the energy of two Hoover Dam hydroelectric power plants.
• The capacity will be used to expand data centres in Texas (Abilene), Michigan, Wisconsin, Wyoming, New Mexico, Georgia, Ohio and Pennsylvania.
• Oracle will build the Stargate central node in Abilene, Texas (1.2→2 GW), purchase about 400,000 GB200 GPUs, and prepare to spend more than $25-40 billion in 2026-2028 on infrastructure.

Why OpenAI is important

This is a strategic step within the Stargate project, a joint venture of OpenAI, Oracle, SoftBank, and MGX with a total investment of up to $500 billion in global data centres. The goal is to provide the AI company with its own infrastructure alongside Microsoft's Azure, CoreWeave, and Google Cloud.

Currently, OpenAI does not have such resources to build physical infrastructure. Its annual revenue reached $10 billion in June. This is $5.5 billion more than last year, so by the time the facilities are commissioned in 2028, OpenAI will likely be able to afford it. Ultimately, this is an investment in the future, as no one doubts that artificial intelligence will penetrate further into our daily lives.