Yet another startup wants to crack the LLM code but this time using light; optical pioneer Oriole Networks wants to train LLMs 100x faster with a fraction of the power

A couple of years ago, Israeli startup CogniFiber made headlines with Deeplight, a fiber-optic cable which could, “process complex algorithms within the fiber itself before the signal hits the terminal.” At the time, we warned this technology wouldn’t reach end users in the near future, and was unlikely to appear in laptops or smartphones anytime soon.

However, eeNews Embedded is now reporting on Oriole Networks, a UK-based startup using light for a different purpose – to create efficient networks of AI chips.

The technology can reportedly train LLMs up to 100 times faster than conventional methods while drastically reducing power usage, and this research aims to mitigate the growing energy consumption of data centers driven by the rapid expansion of AI workloads and the increasing demand for high-performance computing.

20 years of photonics research

“Our ambition is to create an ecosystem of photonic networking that can reshape this industry by solving today’s bottlenecks and enabling greater competition at the GPU layer. Building on decades of research, we’re paving the way for faster, more efficient, more sustainable AI,” said James Regan, CEO of Oriole Networks.

The company’s roots lie in optical network research from University College London (UCL) and Oriole’s unique IP is based on the work of founding scientists Professor George Zervas, Alessandro Ottino, and Joshua Benjamin.

The startup has already drawn attention from a number of investors keen to find solutions for AI’s increasing energy demands.

With plans to release its early-stage products by 2025, Oriole Networks hopes to reshape AI’s infrastructure by making it faster, more energy-efficient,…