Licensable GPU.
GPU silicon IP with a software ecosystem that you can license, customize, and tape out. Designed to scale from chiplet to SoC to ETM.
It Starts with
one core.
OxCore is foundational GPU IP — a single, licensable compute core that fuses CPU scheduling, GPU tensor computation, and CUDA-compatible programmability into one unified architecture.
Scale it from a single chiplet at the edge to a full SoC, all the way to an ETM in the data center. Tape it out under your own brand. Run your existing CUDA software stack without modification. OxCore is the building block that semiconductor companies and data center operators need to build custom AI silicon without vendor lock-in.
A Custom Fusion of CPU, GPU
and TPU Architecture.
- Queuing-based async work dispatcher
- Nano Agents in Silicon
- Hardware task scheduler
- Scalable: 1 to 1,000s of cores
- Purpose-built for AI workloads
- Native CUDA / PTX support
- Future-proof programmable design
OxCore in Action.
Our chiplet-based architecture successfully works on FPGA hardware and generates tokens running GPT-4 class models in simulation. By breaking free from monolithic GPU designs, we're creating a new paradigm for AI hardware.
Build your next chip on