London-based startup developing Optical Tensor Processing Units (photonic AI inference chips) combining SRAM with photonics to optimize AI model inference efficiency and lower costs versus traditional GPU/HBM designs.