Visionary GPU Expert, Raja Koduri, Unveils New Venture OXMIQ
OXMIQ Labs Revolutionizes AI and Graphics Computing
OXMIQ Labs Inc., a new GPU software and IP powerhouse founded by renowned GPU architect Raja Koduri, is on a mission to transform AI and graphics computing globally [1][3][4]. The company aims to re-architect GPUs for the multimodal AI era, where diverse data types—text, images, audio, video, and 3D environments—work together seamlessly.
At the heart of OXMIQ's mission is a unique combination of GPU hardware IP with a unified, software-first approach that emphasizes developer experience and compatibility across both OXMIQ and third-party AI platforms [1][2][5]. Key aspects of OXMIQ's technology focus include:
- Multimodal AI and Graphics Computing: OXMIQ's goal is to deliver GPU architectures optimized for the complex demands of multimodal AI workloads, seamlessly integrating multiple data types in real time [1][3][4].
- Modular GPU Core (OXCORE): The core GPU IP, called OxCore, is a scalable and modular design integrating scalar, vector, and tensor compute engines, supporting near-memory and in-memory computing for high efficiency [3][5]. It also includes support for nano-agents, native Python acceleration (OXPython), and CUDA/SIMD compatibility.
- Chiplet-Based SoC Building (OXQUILT): The OxQuilt architecture allows customized system-on-chip (SoC) creation via chiplets tailored to specific AI workloads—from edge inference to large data center training—allowing configurable compute, memory, and interconnect ratios [3][5].
- RISC-V and CUDA Compatibility: OxCore leverages the open RISC-V instruction set architecture and supports native Python acceleration (OXPython), enabling seamless execution of existing CUDA applications without modification, bridging compatibility especially for NVIDIA CUDA workloads on non-NVIDIA hardware [3][5].
- Unified Software Ecosystem (OXCapsule): Alongside hardware, OXMIQ offers a complete software ecosystem designed for ease of use and broad compatibility, focusing on a unified developer experience [3].
OXMIQ's software ecosystem, OXCapsule, simplifies deployment across diverse environments. A key component, OXPython, allows Python-based NVIDIA CUDATM AI applications to run on non-NVIDIA hardware without any code changes.
Strategic collaborations with Indian tech ecosystem players, such as Mihira Visual Labs (co-founded by Koduri), power cinematic AI platforms to enable Indian creators with AI-driven storytelling tools at a global scale [1][3].
The company has emerged from two years of stealth R&D with over 500 combined years of GPU and AI expertise, hundreds of patents, $100 billion in past product sales impact, and $20 million of seed funding from investors including MediaTek [1][2][3]. OXPython is launching first on Tenstorrent's platform, with Jim Keller, CEO of Tenstorrent, expressing excitement about partnering with OXMIQ on their OXPython software stack.
As the Indian government pushes to make India a semiconductor manufacturing and AI innovation hub through the Digital India and India Semiconductor Mission, OXMIQ's licensing-first and software-first model and GPU IP offer opportunities for Indian chipmakers, AI startups, and research institutes to access next-gen GPU IP without heavy capital investment. Lawrence Loh, SVP of MediaTek, believes OXMIQ's GPU IP and software innovations will drive a new era of compute flexibility across devices.
- OXMIQ Labs, with its focus on the unified, software-first approach, aims to revolutionize finance and investing by creating AI tools that utilize data-and-cloud-computing and artificial-intelligence technologies to analyze large datasets and make more accurate predictions in the stock market.
- Leveraging the open RISC-V instruction set architecture and native Python acceleration (OXPython), OXMIQ's technology could potentially be applied to various industries, including investing, by allowing the deployment of Python-based AI models on non-NVIDIA hardware, thereby reducing costs and increasing accessibility to advanced AI technologies.
