The founder and CEO of Oxmiq Labs, Raja Koduri, recently sat down to talk with Salah Nasri on his Semiconductor Leadership Podcast while dropping a few nuggets of knowledge. The AI-focused startup has been working behind the scenes to develop licensable GPU software and hardware IP to “meet the demands of next-generation gaming, graphics, and multimodal AI” and emerged from its ‘stealth’ state in August 2025. Koduri previously served as a GPU architect at AMD, Intel, and Apple.

In the interview, Koduri discusses the importance of a close connection between hardware and software architecture, pointing towards Nvidia as a prime example of how to do it correctly. Many people tend to separate the two when they think about building a new product, but Koduri says they should be “tightly tied at the hip”.

Hardware and software architecture are “one and the same thing”

“People who make mistakes entering into the AI space and who think that hardware and software are two separate things, and two separate investments or two separate architectures end up going down the wrong path.

The primary reason why Nvidia is so successful is because the software architecture and hardware architecture are one and the same thing. So the hardware reflects the software architecture. We are one of the few people – with our background and history – that understand it.”

Raja Koduri, Founder and CEO of Oxmiq Labs

The Oxmiq Labs founder went on to say that the company openly avoids splitting development into a “hardware team” and “software team,” even though they are tempted to do so. Unifying these components is a clear focus of Oxmiq Labs.

Koduri also talks about an overwhelming reliance on Nvidia’s CUDA platform, even jokingly calling it the “CUDA virus”. Attempting to make powerful local AI for non-Nvidia hardware, Koduri says the team found it “impossible […] even some installers wouldn’t work on non-Nvidia hardware”. This led to rethinking the GPU stack and the development of one of Oxmiq’s flagship components, OXPython, to enable Python-based Nvidia CUDA AI applications to run on non-Nvidia hardware without code modification or recompilation.

Later in the interview, Koduri admits he’d “probably say CUDA” when it comes to choosing the ‘greatest GPU software’ ever developed, though points out that GPU software is not monolithic and relies on “many layers of the stack and many APIs”. Again, this goes back to the close connection between hardware and software, as CUDA is “co-architectured with hardware”.

“I probably say CUDA because it is the abstraction that is co-architected with hardware. So that is the most shiny example that is out there of hardware software co-architecture, and that’s actually the reason why it’s successful.”

Raja Koduri, Founder and CEO of Oxmiq Labs

Oxmiq labs was originally founded in 2023, though it wasn’t in the public eye. After two years of “intensive IP development,” we first got to learn what the company was all about in its “From Atoms to Agents” blog post launch.

About the Author

Jack Goodall

At PC Guide, Jack is mostly responsible for reporting on hardware deals. He also specializes in monitors, TVs, and headsets and can be found putting his findings together in a review or best-of guide.