Oxmiq Labs raises $35M Series A for licensable GPU IP, eyes Arm-like architecture
GPU startup Oxmiq Labs has closed a $35 million Series A round from investors including Taiwan's MediaTek and Pegatron Venture Capital, bringing its total funding to $60 million. The company, founded by veteran GPU architect Raja Koduri (former Intel chief architect and AMD executive), is building licensable GPU IP and software to lower the cost of AI chip design and deployment, rather than shipping proprietary chips.
Oxmiq's core offering is a modular GPU architecture with two main components: OXCORE, a scalable GPU core with scalar, vector, and tensor compute engines that can be customized for specific workloads, and OXQUILT, a chiplet-based architecture that lets customers configure optimal ratios of compute, memory, and interconnect for their needs. The company also packages a unified software stack—OXCapsule and OXPython—that enables Python CUDA applications to run on non-NVIDIA hardware without code changes. OXCORE scales from single-core edge devices to thousands of cores in data centers.
The strategy mirrors Arm's IP licensing model: instead of the $500M+ capital expenditure required for custom chip fabrication, semiconductor companies and data center operators license OXMIQ's designs and adapt them to their own roadmaps. Koduri told Reuters the company wants to "be Arm for the next generation." Partners like Tenstorrent (which will launch OXPython support) and MediaTek position Oxmiq to reach emerging markets like India, where capital efficiency is critical for AI infrastructure buildout.
For architects and infrastructure teams, Oxmiq breaks NVIDIA's CUDA lock-in by enabling portable AI workloads across heterogeneous hardware. The Series A lets the team finish its first IP products and scale hiring; OXCORE and OXQUILT are slated for H1 2026 licensable release, while OXCapsule and OXPython launch via partner integrations by late 2025. Koduri's track record spanning S3, ATI (acquired by AMD for $5.4B), Apple, and Intel signals credibility for delivering production-grade silicon IP.
Sources
- Primary source
- eetimes.com
“GPU IP and AI software that lets semiconductor companies and data centers build custom AI compute without proprietary lock-in”
- oxmiq.ai
“OXMIQ develops licensable compute IP and adaptive AI infrastructure software”
- datacenterdynamics.com
“Oxmiq's hardware offering consists of a RISC-V-based GPU IP core, called OxCore, that supports near-memory and in-memory compute capabilities”
- finance.yahoo.com
“OXMIQ will also get into the custom chip market, where Broadcom, Marvell and MediaTek compete”