O
£106K / year · London, England, GB
Olix builds Optical Tensor Processing Units, light-based AI accelerators using photonic SRAM architectures and rack-scale co-design to address memory, energy and speed limits in frontier AI inference.
Olix develops Optical Tensor Processing Units (OTPUs), light-based AI accelerators intended to address the memory, energy and speed constraints of GPU-based infrastructure for frontier AI inference. Founded in 2024 by James Dacombe and originally launched under the name Flux, the company focuses on fundamental architectural change rather than incremental improvement to existing designs.
The company's OTPU combines an SRAM architecture with photonic circuits. Olix states this design is intended to exceed HBM-based architectures on throughput per megawatt and total cost of ownership, while outperforming silicon-only SRAM architectures on interactivity and latency. Its second product, the OLIX Decode Accelerator 1 (DX-1), is described as the first accelerator architected specifically for decode, with rack-scale co-design spanning logic, data movement, packaging, optics and interconnect.
Olix's technical work spans photonics, optical computing, AI accelerator hardware, SRAM and HBM memory architectures, chip packaging, interconnect design and inference/decode optimisation. The company takes a full-stack systems approach, co-designing across layers of the stack rather than at the component level. It operates in the AI infrastructure and semiconductor sectors, serving frontier AI model training and inference, and holds a contrarian engineering position relative to dominant full-stack incumbents in silicon, software and supply chain.
Team size, funding and geographic presence are not disclosed. Further detail is available at olix.com.