OXMIQ Raises $35M to Scale Arm-Style GPU for AI Chips

OXMIQ Labs Inc., a unified GPU and AI architecture company founded by Raja Koduri, closed a $35 million Series A financing. Total capital raised stands at $60 million. The company will use the funds to scale OxCore, its licensable GPU architecture. Semiconductor companies and AI system builders use OxCore to build custom AI silicon without a full chip program.

Fundomo and Samsung Catalyst Fund co-led the round. MediaTek, AM Intelligence Labs, Pegatron Venture Capital, CDIB-TEN, Darwin Ventures, and Morgan Creek Digital also participated. The investor list includes a contract manufacturer and a mobile chip giant, plus several venture firms. Each has a strategic interest in diversifying the AI chip supply chain.

Samsung Catalyst Fund is the venture arm of Samsung, the world's largest memory chip maker. MediaTek is the leading mobile chip designer for Android devices. Pegatron is a major contract manufacturer for Apple and other electronics companies. Their involvement suggests large semiconductor houses are exploring alternatives to the current single-supplier dynamic.

The AI chip market is fragmenting. Nvidia's CUDA ecosystem still commands the bulk of training and inference workloads. A growing number of hyperscalers and device makers are designing custom silicon to cut costs and improve power efficiency. A licensable GPU architecture occupies a middle ground. It gives customers a head start on hardware design, avoiding years of R&D, while retaining control over customization.

OXMIQ's model resembles Arm's approach in mobile. Arm licenses processor designs to chipmakers who integrate their own accelerators and memory controllers. OXMIQ aims to do the same for GPU compute in AI workloads. Koduri previously led GPU architecture at Intel and held senior roles at AMD and Apple. He founded OXMIQ in 2025 with a focus on open, licensable GPU architecture. The company is based in Campbell, California.

MediaTek has been expanding its AI-capable chips for edge devices. The company could integrate OXMIQ's IP into future system-on-chips, giving it a differentiated offering against Qualcomm and other rivals.

Licensable GPU architectures have a history of mixed results. Imagination Technologies licensed PowerVR cores to Apple and others for years. Arm's Mali GPUs are widely used in mobile. The difference now is the scale of AI demand. Training and inference workloads require massive parallel compute, and the market is large enough to support multiple architectures. OXMIQ's pitch is that customers can get close to Nvidia-level performance with less time and money, and with full control over integration.

The $60 million cumulative figure is modest relative to the hundreds of millions Nvidia spends annually on R&D. The licensable model changes the unit economics. Each OXMIQ customer pays upfront license fees per project and royalties per chip sold. If OXMIQ signs a handful of serious customers, revenue could compound without the capital intensity of building its own fabs.

For traders tracking the AI hardware space, OXMIQ's progress is a potential long-term pressure point on Nvidia's margins. If licensable GPU architectures gain adoption, the pricing power Nvidia enjoys today could erode. The read-through for the broader semiconductor sector is that the differentiation edge is moving from raw compute to customization and ecosystem integration. Companies that can offer both, like NVIDIA with its CUDA lock-in, have the most to lose in a fragmentation scenario.

The round closed in July 2026. OXMIQ said it will use the funding to scale OxCore and support customer engagements.