Oxford spin-out and AI accelerator startup Lumai raises over $100 million to reduce AI computation costs with optical acceleration

With AI models growing and power hunger, data centers are struggling to keep up. The cost of training and execution of large-scale language models (LLMS) continues to climb, and silicon chips reach performance limits. Energy use in US data centers is expected to triple by 2028. This consumes up to 12% of the country’s electricity.

Lumai, an AI accelerator startup that spun from Oxford University, believes the answer lies in optics, not silicon. Founded in 2022, the company was born from a collaboration between five experts: Tim Weil, Xianxin Guo, Alex Lovsky, Thomas Barrett and James Spall.

Today, Lumai announced it has raised more than $10 million to expand its optical computing technology, and has pledged to reduce costs and energy use in AI data centers. The round is led by Constructor Capital and will be supported by IP Group and new supporters such as Photonventures, Journey Ventures, Liftt, Qubits Ventures, State Farm Ventures, and Tis Inc.

This funding will help Lumai double the team, accelerate product development and expand its presence in the US

Lumai’s technology completely avoids silicon by performing AI core arithmetic operations within beams of light moving through 3D space. The optical accelerator fits standard PCIE form factors and makes it easier to deploy in existing systems. Lumai uses free space optics to overcome the limitations of silicon photonic computing and hopes to bring an 8-pop AI inference accelerator to the market in the coming years.

The company claims that hardware can reduce AI inference costs by 10% of current solution bills, offering 50x more performance while using 10% of the power supply.

“The future of AI requires a fundamental breakthrough in computing,” said Tim Weil, CEO and co-founder of Lumai. “Current LLMS costs are unsustainable, and next-generation AI won’t happen without a major shift. Lumai’s innovative optical computing design overcomes scalability challenges that have dramatically reduced power consumption that can hinder others and reduce AI costs.”

Investors are betting that Lumai’s approach can restructure AI infrastructure. “Life and intelligence are large-scale carbon- and electron-based neural models trained for 2 billion years. Fossil fuels are a by-product of this evolution, and if we continue to use electronic-based computing, it may not produce enough energy to create better models. Technology is an important step in improving matrix multiplication, as is the advancement Quantum Computers offers for other computational scenarios.”

“Lumai technology has the potential to change the future of AI as we solved the challenges of optical computing to provide low-cost, scalable solutions,” said Dr. Lee Thornton of IP Group. “We are proud to continue to support Lumai as we embark on the next phase of our journey.”

Ewit Roos of Photonventures said, “Lumai is not innovative. It fundamentally reshaping the future of AI computing. Lumai presents one of the most compelling opportunities in next-generation data center technology and places it at the forefront of the AI ​​revolution.”

Startups are already gaining industry traction. He recently won the “Best Onlient Technology” at the Global OCP Future Technologies Symposium and completed the Intel Ignite London program, whose co-founder was recognized by the Royal Academy of Engineering and the Photonics 100 List.

With fresh support and momentum gaining, Lumai aims to rewrite the rules for AI infrastructure.