OpenAI is reportedly considering making its internally developed cross-chip software optimization tools publicly available. If this move materializes, it would directly challenge the moat that NVIDIA has long built with its CUDA software ecosystem.

According to a report by The Information, OpenAI's senior vice president of computing and infrastructure, Sachin Katti, stated during a public discussion that the company is developing a software abstraction layer. This layer would allow researchers and product teams to run AI workloads without needing to concern themselves with the underlying hardware supplier.

When asked if this capability would be made available to the public, Katti explicitly stated that it is "on the table." He described it as an "agentic optimization capability," adding, "We want to give that to the world."

Analysts suggest this statement is significant. NVIDIA's market dominance has long relied on CUDA—its proprietary suite of compilers, libraries, and optimization tools, which is a core dependency for mainstream AI developers running software on NVIDIA chips. The public release of OpenAI's cross-platform tool could further erode CUDA's differentiating advantages and accelerate a more diversified competitive landscape in the AI computing market.

Accelerating Multi-Chip Strategy

Katti reportedly stated that the AI industry is moving towards a state of "high heterogeneity," where companies will simultaneously use AI chips from multiple vendors. This outlook reflects a profound shift in OpenAI's own strategy.

While OpenAI previously relied almost entirely on NVIDIA chips, it has recently signed agreements with Amazon, Cerebras, and AMD to incorporate their AI chip resources. The company is also developing its own custom AI chips.

Katti did not reveal in the discussion whether OpenAI would adopt custom chips from Google, as companies like Anthropic and Meta have done.

This trend is not unique to OpenAI. Other major AI players like Anthropic and Meta are similarly unwilling to depend on a single supplier for such a critical part of their operations, and no single supplier can meet their massive computational demands alone.

The Software Abstraction Layer: An AI Version of Google's Borg

The report indicates that Katti compared the software architecture OpenAI is building to Google's famous Borg computing management system. Borg is the key infrastructure that allows Google to scale products across heterogeneous hardware. "That's the path we're on for AI," he said.

More disruptively, Katti suggested that AI itself could become the tool to break CUDA's monopoly. "We expect to use AI to generate optimized kernels, which will truly support all these different chip options," he stated.

Anjney Midha, founder of Amp, noted in the same discussion that if developers like OpenAI publicly release such internal tools, enabling AI to run efficiently on chips from NVIDIA, Google, AMD, and others, it would pose a substantial challenge to NVIDIA.

In fact, CUDA's moat is already narrowing. Meta's PyTorch framework has long allowed developers to write AI code more easily for various chips, and some startups are selling tools that translate PyTorch code into low-level code that runs directly on the hardware.

Vera Rubin Deployment and Shifting Bottlenecks

Beyond software strategy, Katti also provided an update on OpenAI's deployment of NVIDIA's next-generation Vera Rubin chip system. He mentioned that OpenAI has received early samples of the chip and expects to put it to use for AI training by the end of this year.

Katti offered a positive assessment of NVIDIA's learning from issues encountered during the rollout of the Blackwell system. The initial Blackwell system caused headaches for several cloud service providers due to networking, firmware, and cabling complexities during large-scale deployment, but the new system has seen significant improvements. "NVIDIA definitely learned from a lot of growing pains," he said.

Katti did not specify which cloud service provider would host OpenAI's Vera Rubin cluster first, only noting there is "healthy competition" among them. OpenAI's current primary cloud providers include Microsoft, Oracle, and Amazon.

Notably, Katti identified the current major bottleneck for computing expansion as power supply and engineering capabilities, rather than the chips themselves.

"What's constraining us right now is more the power and the engineering, than anything else," he stated. This assessment has direct implications for how investors in AI infrastructure might allocate their resources.