Oxmiq Labs, an AI chipmaking startup founded by renowned GPU architect Raja Koduri, secured $35 million in Series A funding to accelerate development of its proprietary GPU architecture and reduce the high costs associated with custom AI silicon design. With a vision to become the 'Arm Holdings' of AI chip IP, Oxmiq targets semiconductor firms and AI system builders globally, offering a scalable and flexible solution that integrates GPU, CPU, and tensor engines.
- Oxmiq integrates GPU, CPU, and tensor processing into one licensable core to simplify AI chip design.
- Series A funding led by Fundomo and Samsung Catalyst Fund boosts scaling and product readiness.
- OxQuilt chiplet fabric and software stack enable flexible configuration and compatibility with mainstream AI frameworks.
Market signal
Oxmiq Labs' recent $35 million Series A funding round highlights growing market interest in solutions that lower barriers to custom AI chip development. The startup’s approach of bundling GPU, CPU, and tensor engine IP into a unified licensable core addresses a critical pain point where multi-component chip design and long development cycles make custom silicon prohibitively expensive and slow to bring to market.
Participation from strategic investors including Samsung Catalyst Fund, MediaTek, and Fundomo signals endorsement from both semiconductor manufacturing and AI ecosystem stakeholders. Oxmiq aims to tap into the expanding demand among cloud providers, tech companies, and AI system designers seeking more tailor-made compute solutions, potentially disrupting existing licensing models dominated by firms like Arm Holdings.
Operator impact
For semiconductor companies and AI hardware designers, Oxmiq’s technology promises a streamlined path to develop custom AI chips. By collapsing CPU, GPU, and tensor processing into a single IP block with open architecture, design teams can dramatically reduce the complexity and cost involved in silicon development. This enables faster iteration and more affordable customization aligned to specific AI workloads.
Moreover, Oxmiq’s OxQuilt chiplet fabric offers adaptable integration, allowing operators to configure designs across diverse process nodes and memory types while maintaining readiness for future innovations such as silicon photonics interconnects. Combined with software tools that support native CUDA and PyTorch workloads, this significantly lowers integration overhead, helping operators focus on AI software innovation rather than hardware adaptation.
What to watch next
In the coming months, attention will focus on Oxmiq’s ability to finalize and commercialize its proprietary IP offering and scale its engineering team to support broader adoption. Key milestones include releasing the first batch of OxCore IP for licensing and demonstrating effective use cases with early customers in the AI and semiconductor space.
Industry watchers should also monitor how the licensing model competes with incumbent chip IP providers and whether Oxmiq’s open architecture approach gains industry traction. Advances in supply chain agnostic packaging and integration, particularly the deployment of chiplets and silicon photonics, will be critical to validate the platform’s flexibility and performance promise.