OpenAI is reportedly widening its cloud computing strategy as its demand for AI infrastructure grows faster than Microsoft can support. According to sources who spoke to The Information, CEO Sam Altman and CFO Sarah Friar told employees about the shift after the company's recent $6.6 billion funding round.
The issue is not a break with Microsoft. It is a capacity problem. Friar reportedly told shareholders that Microsoft was not providing enough processing power fast enough, and OpenAI's contract allows the company to explore other data center options.
Why compute has become the bottleneck
For a company building frontier AI models, access to processing power is not just an operations question. It shapes how quickly models can be trained, improved, deployed, and scaled. The source article points to OpenAI's concern that Microsoft may not deliver servers quickly enough for OpenAI to stay ahead of Elon Musk's xAI.
That competitive pressure matters because Musk plans to release Grok 3, which he claims will be the most powerful AI model, by the end of the year. His AI company xAI is also building a massive server infrastructure in Memphis.
OpenAI's reported move toward Oracle should be read in that context. The company appears to be trying to reduce the risk that one cloud partner becomes the limiting factor for its AI roadmap. If models require more compute, and if competitors are building large server clusters of their own, delays in data center delivery can become strategic problems.
Oracle becomes a bigger part of the plan
OpenAI announced its first Oracle deal in June. According to The Information's sources, Microsoft was only marginally involved in that arrangement. Even so, the deal still contributes to Microsoft's Azure revenue because OpenAI runs Azure infrastructure on Oracle servers.
That structure is important. It suggests OpenAI can expand beyond Microsoft-owned data centers while still operating within parts of the Azure ecosystem. In practical terms, Oracle may provide the physical server capacity while Azure remains part of the infrastructure layer.
OpenAI is now reportedly negotiating with Oracle to lease an entire data center in Abilene, Texas. Sources cited by The Information said the facility could reach nearly a gigawatt of power by mid-2026, potentially housing hundreds of thousands of Nvidia AI chips. The site also has room to expand up to two gigawatts, if enough energy is available.
Those details show why data center power is becoming as central as chips. Nvidia AI chips are critical, but they need large facilities, sufficient electricity, and fast deployment schedules. A data center that can grow from nearly a gigawatt to two gigawatts would represent a major commitment to AI compute capacity, assuming the energy supply can support it.
Microsoft is still building for OpenAI
The reported Oracle push does not mean Microsoft is out of the picture. Microsoft aims to give OpenAI access to about 300,000 of Nvidia's latest GB200 graphics processors in data centers in Wisconsin and Atlanta by the end of next year.
Altman has asked Microsoft to speed up the Wisconsin project, which could open partially in the second half of 2025. That request points to the central tension in the partnership: Microsoft remains a major infrastructure provider, but OpenAI wants faster delivery than Microsoft may currently be able to provide.
The source article presents this as a timing problem as much as a supply problem. OpenAI needs servers and chips at a pace that matches its growth and competitive goals. Microsoft is working on large deployments, but OpenAI is reportedly seeking additional capacity from Oracle because waiting may carry its own risks.
Custom chips are part of the longer-term answer
OpenAI is also planning to use more of its own AI chips in the future. The goal is to meet growing computing demands and reduce costs. According to the source, the company is working with Broadcom and Marvell to design ASIC chips.
That effort would not remove the need for large data centers. Custom AI chips still require manufacturing capacity, infrastructure, and deployment. But it could give OpenAI more control over the hardware it depends on, especially if demand for Nvidia AI chips remains intense.
OpenAI has reportedly reserved capacity for TSMC's new A16 Angstrom process, with mass production set to start in the second half of 2026. That timeline indicates that custom silicon is a future lever, not an immediate replacement for cloud compute from Microsoft or Oracle.
What the reported shift means
The picture that emerges is clear: OpenAI is trying to secure enough AI compute from several directions at once. Microsoft remains deeply involved. Oracle is becoming more important. Nvidia AI chips remain central. Custom ASIC chips are being prepared for later.
The strategy reflects a simple constraint. Advanced AI development depends on access to massive processing power, and the companies that can secure servers, chips, data centers, and electricity fastest may gain an advantage. OpenAI's reported talks with Oracle show that compute is no longer a background infrastructure issue. It is now one of the main forces shaping the future of AI competition.