Oriole Networks raises a £10m seed round to build AI ‘super-brains’
capable of training LLMs 100x faster with a fraction of energy
Posted
Oriole Networks, a UCL spinout using light to train LLMs a hundred times faster with only a fraction of power – has raised £10 million in Seed funding to revolutionise AI performance and adoption, and solve AI’s critical energy problem. The round, which is one of the UK’s largest Seed raises in recent years, was co-led by us via the UCL Technology Fund, Clean Growth Fund, XTX Ventures and Dorilton Ventures. It was supported by Innovate UK Investor Partnership.
Founded in 2023 by UCL scientists, Professor George Zervas, Alessandro Ottino and Joshua Benjamin, with seasoned CEO James Regan, Oriole Networks has developed a novel way of using light to connect thousands of AI chips together. Once connected, the power of each individual graphics processing units (GPUs) is combined to form a “super-brain.” This super-brain can be used to train advanced Large Language Models a hundred times faster, with a thousandth of the latency, while using a tiny fraction of energy.
This not only addresses rising concerns about the “obscene energy demands of AI”, but has the power to revolutionise time critical tasks, such as algorithmic trading, increase AI adoption, and accelerate machine learning algorithmic progress in a low carbon world. This will bring huge gains for all companies working with, and on AI, but particularly data centres. Data centres play a critical role in the evolution and proliferation of SaaS and its shift towards AI, yet are facing systemic problems and unsustainable power consumption due to increased demands.
Oriole Networks is a spinout from University College London and its IP is licensed through UCL’s technology transfer company, UCLB. It is being led by CEO Regan who already has a track record of building successful tech companies from university spinouts, having spun out EFFECT Photonics and built it to a half-a-billion-dollar company.
George Zervas, CTO, said: “AI computational needs are increasing by 10 times every 18 months. This leads to distributed training and inference across large numbers of xPUs. Collective data movement across the servers in the data centre becomes a bottleneck which in turn limits the training and inference completion time. This requires a fundamental shift in the co-design of next generation networked systems.”
James Regan, CEO, added: “As the demand for compute continues to increase, it is critical to find new solutions that can address these challenges in a sustainable and carbon efficient manner. Our novel approach to harness the power of light has already demonstrated significant technical performance improvements, up to 100 times speed up in completion time and 40 times improvements in energy consumption.”
“It’s rare to have such depth of innovation over many years at UCL combined with an experienced entrepreneur with domain knowledge and a massive market that is looking for this solution. This is going to be an exciting journey.”
David Grimm, Partner
Daniel Freeman, General Partner at Dorilton Ventures, said: “We invest in companies in the IT infrastructure, data science, and cyber security segments whose products support computationally driven businesses. Over the last decade, compute performance has improved ten times faster than networking performance, so HPC environments are highly network constrained. Oriole’s exciting approach can unlock the latent potential in existing infrastructure.”
Meanwhile, Beverley Gower-Jones OBE, Managing Partner of Clean Growth Fund, emphasised the criticality of sustainable development: “The world’s data centres already consume as much electricity as the whole of the UK, and it is rising rapidly, threatening to consume as much as the whole of Europe unless something is done. This radical approach to Net Zero innovation is exactly what is needed.”
Sifted has more here
The Times has more here.
Other news
