OXMIQ Raises $35M to License AI Chip Designs Instead of Making Them

Why an IP-first model matters

Developing bespoke AI silicon typically demands both deep capital and years of engineering — a barrier that keeps most companies out of the market and concentrates power among a handful of established chipmakers. OXMIQ's approach is to sell architectural intellectual property instead of finished chips, allowing clients to own their computing roadmaps while avoiding the heavy manufacturing costs that have sunk many hardware ventures.

For the Africa–Europe technology corridor, the significance lies in access. A licensing-based model lowers the entry threshold for smaller design teams and system builders that lack the resources of incumbent giants. That could let developers and startups outside the traditional silicon hubs participate in custom AI hardware design without launching prohibitively expensive full-chip programs. Keller, as cited by Ventureburn, framed an open GPU foundation as essential to removing artificial limits on AI innovation.

Inside the technology

At the core of OXMIQ's stack is OxCore, a scalable, licensable GPU core that combines three compute functions in one system: a CUDA-compatible GPU engine, a tensor processing engine, and a CPU orchestration engine. Consolidating capabilities that normally require separate physical chips, the design leans on near-memory computing to cut data movement and reduce energy use.

OxCore is paired with OxQuilt, a chiplet integration architecture that OXMIQ says is not tied to any single foundry or memory type. According to Ventureburn, this flexibility lets customers mix logic process nodes and packaging options across different supply chains.

On the software side, OXMIQ offers OxPython, a layer intended to run existing CUDA and PyTorch code on OxCore hardware without code changes. The company describes this as preserving software portability while ensuring compatibility with newly released AI models.

What comes next

OXMIQ is currently demonstrating its architecture on FPGA setups and says it is working with semiconductor companies, robotics firms, and AI system builders that want more control over their hardware. The pitch is timed to a market moment in which demand for AI processing is reportedly growing faster than physical data center capacity.

Whether the licensing approach can loosen the grip of dominant chip vendors remains to be seen, but the funding signals investor appetite for alternatives to the capital-intensive, build-your-own-chip status quo.

Emile Burger is leaving his role as chief executive of Tarsus Distribution after a tenure that also saw him serve as the company's finance chief.