Login to Continue Learning
Oxmiq Labs will focus on rebuilding the entire GPU ecosystem from scratch, prioritizing a “Software First” strategy that enables unmodified Python-based CUDA applications to run on non-NVIDIA hardware. The company’s goal is to produce AI GPUs from the ground up, using hundreds of patents and a novel chiplet architecture called OXQUILT.
The increasing demand for multimodal AI, where various types of data interact in real-time, has highlighted the need for flexible GPU designs capable of processing text, images, and video simultaneously. Oxmiq’s modular OXCORE architecture brings Scalar, Vector, and Tensor units together to support nano agents, Python acceleration, and CUDA/SIMD compatibility. This design will scale from tiny devices to large data centers.
Oxmiq’s flagship software, OXPython, allows NVIDIA CUDA applications to run on non-NVIDIA hardware without modification. It will be utilized first on Tenstorrent AI chips before expanding into a unified software ecosystem called OXCapsule. This system aims to hide the complexity of hardware and enable developers to deploy AI apps across multiple platforms.
Oxmiq’s business model is centered around licensing its IP, avoiding the costly tape-out process by not building and selling full-fledged GPUs. The company has already secured $20 million in seed funding from investors like MediaTek and Mediatek. With Raja Koduri leading this venture, it marks the first GPU startup in Silicon Valley in over 25 years.
Jim Keller of Tenstorrent expressed excitement about Oxmiq’s OXPython software stack, noting its ability to bring Python workloads for CUDA to AI platforms. Lawrence Loh of MediaTek praised Oxmiq’s bold vision and world-class team, emphasizing their potential to drive a new era of compute flexibility across devices—from mobile to automotive to AI on the edge.
Raja Koduri’s return to the GPU business with Oxmiq Labs is an exciting development in the tech industry, promising significant advancements in GPU design and software.
