GPU IP startup Oxmiq, founded by former Intel and AMD graphics chief Raja Koduri, raised a $35M Series A to accelerate product development and customer integration of its licensable GPU architecture and software stack. The round was co-led by Samsung Catalyst Fund and other strategic investors, expanding from a $20M seed round disclosed in August 2025.
Oxmiq delivers two core products: OxCore, a modular, RISC-V-based GPU IP core supporting scalar, vector, and tensor compute, and OxQuilt, a chiplet-based system builder that lets customers configure optimal ratios of compute, memory, and interconnect. The company also ships OxCapsule, a unified software ecosystem, and OXPython—a key differentiator that lets Python CUDA applications run unmodified on non-NVIDIA hardware, launching first on Tenstorrent's Wormhole and Blackhole AI platforms.
The licensing-first model avoids the heavy capex of traditional GPU chip startups; Oxmiq has already recorded software revenue. Oxmiq scales from single-core edge AI applications to thousands of cores for zettascale data centers and incorporates novel features including nano-agents in silicon using RISC-V cores, near-memory and in-memory compute, and light transport—addressing heterogeneous AI workloads beyond fixed-function accelerators.
For architects, Oxmiq's approach matters: it promises to reduce vendor lock-in and accelerate time-to-market for custom AI silicon without rebuilding from scratch. CUDA compatibility broadens adoption, while the chiplet-first design aligns with the industry shift toward modular, power-efficient AI compute. The $35M validates demand for open GPU IP in a market increasingly frustrated by NVIDIA's dominance.