Oracle allocates “$40 billion for Nvidia”’s latest AI chips, which it will deploy in a large-scale data center project for OpenAI in Texas
With an investment of $40 billion in chips, the company will build the first American “Stargate” initiative, an ambitious AI infrastructure project that is expected to cost $500 billion in the long term.
400,000 Superchips
The Abilene data center will be one of the largest in the world. Oracle is said to purchase 400,000 Nvidia GB200 chips and subsequently lease the computing power to OpenAI. Construction started in June 2024, and completion is scheduled for mid-2026. Oracle is leasing the entire site for 15 years.
The GB200 is Nvidia’s most powerful AI chip to date and is designed for training and running large-scale AI models. The chips will be placed in eight buildings on a site financed with $15 billion in loans and equity from Crusoe and investor Blue Owl Capital. JPMorgan provided $9.6 billion of the debt financing.
OpenAI Wants to be less Dependent on Microsoft
Until now, OpenAI’s AI infrastructure ran almost entirely on Microsoft Azure. But the explosive demand for computing power is said to have exceeded Microsoft’s capacity. The two parties ended their exclusivity agreement earlier this year and are still negotiating licensing rights for OpenAI models.
Stargate is intended to give OpenAI more autonomy. The project aims to raise between $100 billion and $500 billion for AI data centers worldwide in the long term. In the US, Amazon and Elon Musk are working on similar projects, according to the Financial Times. OpenAI is also expanding across borders, with a second 5-gigawatt Stargate campus coming to Abu Dhabi.