The success of Data Centers over the last two decades has led them to consume 2-3% of the global power consumption; the equivalent of the total UK electricity needs. Predictions estimate that by 2040 the power consumed by Data Centers will be equivalent to the electricity needs of the whole of Europe. Meanwhile, Large Language Models (LLM) demand is growing 10 times every 18 months, far outpacing the computational improvements of GPUs and other custom processors/accelerators. This creates a significant gap between compute performance and LLM application requirements that increases 256 times every 18 months. Due to Moore's Law slowdown, electronic switched networks that are used to interconnect servers, GPUs and storage systems lead to significant computational and application overhead that can be as high as 90%. Thus only 10% of the time a 10,000 GPU cluster is used to perform the computational tasks of LLM training and 90% of the time is used to exchange information. The same challenges exist more broadly to any parallel-distributed computing tasks running in data centers. This project, following pioneering research conducted over the last two decades and significant IP, will prototype a photonic network solution that will eliminate the need for electronic packet switching. It will deliver a step-change in performance and energy efficiency at a large scale. It will prototype a photonic integrated circuit to provide ultra-fast, intelligent, switching functionality using trusted high-volume manufacturing and demonstrate its end-to-end network capabilities.