The Information reported on Sunday that OpenAI is considering releasing externally software that would allow its AI to run on multiple chips.

The Information said that if OpenAI follows through, it could create cracks in Nvidia's monopoly.

According to the report, OpenAI compute and infrastructure chief Sachin Katti (사친 카티) said the company is developing software, an abstraction layer, that would let researchers and product teams work without worrying about which server their tasks run on. He called it an "agentic optimization capability" and said, "I want the whole world to be able to use it."

OpenAI has relied entirely on Nvidia chips for years. More recently, it has signed contracts to use chips from Amazon, Cerebras and AMD, and it is also developing its own AI chip.

Nvidia's CUDA software ecosystem is also cited as a core competitive strength behind its dominance of the AI market.

CUDA is a proprietary set of software tools optimized for Nvidia chips and has played a key role in making developers feel they have no choice but to use Nvidia chips.

The Information said Katti believes that lock-in would weaken if AI can generate optimized code on its own to run across different types of chips.

Katti said OpenAI has already secured samples of Nvidia's next-generation Vera Rubin chip and plans to use them for AI training later this year. He said, "The biggest bottleneck in scaling compute right now is power and the engineering capability to bring new hardware online, rather than chips."