TLDR: Former Intel chief architect Raja Koduri has launched Oxmiq Labs, a new startup focused on developing and licensing RISC-V-based GPU intellectual property. The company, which secured $20 million in seed funding, aims to disrupt the GPU market with an asset-light model, targeting AI and multimodal workloads, and offering CUDA compatibility.
Raja Koduri, a veteran of the semiconductor industry and former chief architect at Intel, has officially launched Oxmiq Labs, a new GPU startup poised to make significant waves in the computing landscape. The company, which has emerged from stealth mode, is dedicated to developing and licensing GPU hardware and software intellectual property (IP) based on the open-source RISC-V International architecture.
Oxmiq Labs has successfully secured US$20 million in seed funding from a diverse group of investors, including the prominent Taiwanese semiconductor company MediaTek. This initial capital will fuel the company’s ambitious plans to innovate in the GPU space.
Distinguishing itself with an ‘asset-light’ business model, Oxmiq Labs will primarily focus on IP licensing and software platform development, rather than engaging in the capital-intensive mass production of chips. This strategic approach aims to circumvent the enormous financial barriers and manufacturing risks typically associated with full chip production in the semiconductor industry, where cutting-edge chip design alone can exceed $500 million.
The startup’s core technological offerings include OxCore, a modular GPU IP core built on the RISC-V instruction set architecture. OxCore integrates scalar, vector, and tensor compute units, designed to support advanced computing paradigms such as near-memory and in-memory computing. Complementing this, Oxmiq is also developing OxQuilt, a chiplet-based system-on-chip (SoC) platform. OxQuilt enables customers to create highly customizable configurations of compute, memory, and interconnect modules, tailoring solutions for diverse applications ranging from edge AI accelerators to large-scale training SoCs.
A key strategic move by Oxmiq Labs is its direct challenge to Nvidia’s dominant software ecosystem. The company has developed a tool that allows software written for Nvidia’s widely used CUDA platform to run on Oxmiq’s hardware without requiring any code modification or recompilation. This compatibility is crucial for attracting developers and breaking down barriers to adoption in a market heavily influenced by existing software infrastructures.
Oxmiq Labs is positioning itself as potentially the first new GPU startup in Silicon Valley in over two decades to adopt such an asset-light model. Koduri, whose extensive career includes senior roles at ATI Technologies, AMD, and Apple, in addition to Intel, acknowledges the complexity of the endeavor, stating, ‘GPUs are not easy.’
Also Read:
- AI Safety Debates Intensify Between Anthropic and Nvidia; New Startup Targets CUDA Workloads
- Neoclouds Reshape AI Investment Landscape with Specialized Computing Solutions
The company’s primary focus is on next-generation graphics and, significantly, multimodal AI workloads. Oxmiq’s GPU IP is designed for scalability, capable of supporting applications from single-core use in robotics to thousands of cores for demanding cloud data centers, with a clear emphasis on AI computing. This direct targeting of AI applications underscores the company’s relevance in the rapidly evolving artificial intelligence landscape.
