Recently, OpenAI has been seeking new cloud computing solutions, seemingly no longer pinning its hopes on Microsoft.
According to internal sources, OpenAI's CEO Sam Altman and CFO Sarah Friar have informed employees about this change, primarily due to the recent completion of a $6.6 billion funding round. Friar told shareholders that Microsoft could not provide the necessary processing power quickly enough, prompting OpenAI to explore other data center options. Under the contract with Microsoft, they have the right to do so.
Altman is concerned that if Microsoft cannot deliver servers quickly, OpenAI will struggle to maintain its competitive edge, especially under the pressure from Elon Musk's xAI. Musk plans to launch his Grok3 by the end of the year, claiming it will be the most powerful AI model. Meanwhile, xAI is building massive server infrastructure in Memphis.
In this context, OpenAI's collaboration with Oracle is deepening. As early as June, OpenAI entered into a preliminary partnership with Oracle, in which Microsoft's involvement was very limited. Nevertheless, this agreement still brought benefits to Microsoft's Azure, as OpenAI was effectively running Azure infrastructure on Oracle's servers.
Currently, OpenAI is in negotiations with Oracle to lease an entire data center in Abilene, Texas. By mid-2026, the facility's power could approach 1 gigawatt, capable of housing hundreds of thousands of NVIDIA's AI chips, and potentially expandable to 2 gigawatts with sufficient energy.
Meanwhile, Microsoft is committed to providing OpenAI with approximately 300,000 of the latest NVIDIA GB200 graphics processors by the end of next year, distributed across data centers in Wisconsin and Atlanta. Altman hopes Microsoft will expedite the progress of the Wisconsin project, which is expected to be partially operational by the second half of 2025.
Additionally, OpenAI plans to use more of its own AI chips in the future to meet the growing computational demand and reduce costs. To achieve this goal, the company is collaborating with Broadcom and Marvell to design application-specific integrated circuit (ASIC) chips. It is reported that OpenAI has reserved capacity for TSMC's new A16 Angus process, with mass production expected to begin in the second half of 2026.
Key Points:
- 🚀 OpenAI is starting to look for new cloud computing solutions due to Microsoft's inability to provide the necessary processing power quickly.
- 🏢 OpenAI's collaboration with Oracle is deepening, with plans to lease a large data center in Texas.
- 💻 OpenAI will use more of its own designed AI chips to meet increasing computational needs and reduce costs.