He worked as an architect at AMD and Apple in the past, and successfully promoted the development of many display applications. He later left Intel in March 2023.Raja koduri, announced earlier on his personal "X" pageA startup called Oxmiq Labs, will combine GPUs designed with RISC-V architecture with software design solutions to challenge the CUDA computing platform ecosystem established by NVIDIA.
Oxmiq Labs will focus on GPU hardware and software technology development and operate through a licensing model. Its biggest highlight is its software layer solution, called OXPython, which allows developers to run Python-based CUDA workloads on non-NVIDIA GPU hardware without modifying any code, thereby enhancing the openness and application flexibility of the computing platform.
GPU design based on RISC-V architecture
Oxmiq Labs' hardware core, OxCore, utilizes the RISC-V instruction set architecture and integrates scalar, vector, and tensor arithmetic units to support both near-memory and embedded memory computing. Combined with the OxQuilt chiplet SoC architecture tool, customers can quickly assemble multi-core compute modules, memory modules, and interconnect modules based on their needs, flexibly creating customized chip solutions for diverse scenarios, from edge AI inference to large-scale AI training.
Emphasis on software is the key
Compared to hardware, Oxmiq Labs emphasizes the value of software stacking. Its core platform, OXCapsule, abstracts underlying hardware differences with a unified runtime and scheduling layer, encapsulating application services into independently operable "heterogeneous containers." This allows developers to deploy tasks on different CPUs, GPUs, or AI accelerators without having to deal with complex configurations.
Among them, the OXPython software layer solution can translate Python applications based on the CUDA computing platform into the Oxmiq execution environment. The first wave of supported platforms will be those invested by Amazon founder Jeff Bezos and Samsung.Chip startup TenstorrentTenstorrent CEO Jim Keller stated that Oxmiq Labs' technology will significantly enhance the developer ecosystem's portability and align with the development trend of open AI hardware and software computing resource stacks.
Not a consumer-grade GPU, may be cooperating with MediaTek
Although Raja Koduri has led multiple graphics card architectures in the past, Oxmiq Labs has not developed consumer-grade GPUs, nor does it include built-in ray tracing or image output capabilities. Instead, it operates as an IP licensing business model, requiring licensees to complete their own GPUs.
Oxmiq Labs also adopts an "asset-light" strategy, avoiding expensive chip mass production and manufacturing processes and focusing on IP and software licensing. It has already secured $2000 million in seed funding from investors including MediaTek and has already generated its first software licensing revenue. MediaTek's investment in Oxmiq Labs suggests that MediaTek may leverage Oxmiq Labs' software technology to enhance its computing capabilities across mobile devices, automotive electronics, and edge AI.
Raja Koduri believes that Oxmiq Labs is a rare GPU startup team in Silicon Valley in the past 25 years. Through a development strategy that separates software and hardware, it has expanded the CUDA computing platform ecosystem, which was previously dominated by NVIDIA GPUs, to a wider range of heterogeneous computing platforms, providing developers with new cross-hardware options.
On the other hand, Raja Koduri also expects that promoting the development of more flexible computing platforms through Oxmiq Labs will help enhance India's position in the global AI and semiconductor market competition.
Been waiting for this day…
Oxmiq Labs Inc., the all-new GPU software and IP startup emerges from stealth. We assembled a world-class team of GPU and AI architects with over 500 years of combined experiences. OXMIQ™'s licensable IP rearchitects the GPU from the ground up.…
— Raja Koduri (@RajaXg) August 5, 2025
