ResolightAI

Today, AI infrastructure is still largely endpoint-focused. Enormous investment is going into better compute platforms and co-packaged optics (CPO), with one clear objective: move data out of the GPU more efficiently, at just a few picojoules (pJ) per bit, so AI systems can scale beyond a single rack. But CPO by itself is not enough. The purpose of getting data out of the GPU is to connect GPUs into a larger AI compute cluster. In today’s roadmap, once a bit leaves the GPU, it may still pass through multiple electronic switches — each with additional optical interfaces, each forcing light-to-electronics and electronics-to-light conversion, and each adding power, latency, and complexity. Resolight.ai believes the real breakthrough comes from coupling CPO with the right network architecture. The data should be converted once at the transmitter and once at the receiver — and remain in light throughout the network. This is the promise of Resolight’s native photonic processor: light-speed optical switching designed for AI scale-up networks. Resolight.ai’s architecture is built to reduce energy beyond what CPO can achieve on its own, unlock much larger scale-up radix, and enable AI clusters that are larger, faster, and more efficient. Read more at: resolight.ai/ #SiliconPhotonics #AIHardware #CPO #photonic #deeptech