Lumai

An Oxford University spin-out building AI accelerators that do the heavy maths with beams of light.

About Lumai

Lumai is an Oxford company building a new kind of AI processor that uses light for its core calculations. AI systems spend most of their time multiplying large tables of numbers. Lumai encodes data into an array of light beams, passes them through a membrane that holds the table values, and uses a lens to add up the results before converting the answer back to an electrical signal. The company says this can cut the power needed for AI inference, the stage where a trained model answers questions, by around 90 per cent compared with conventional chips. Its planned products are modules that slot into standard data-centre racks.

Lumai began as OxONN and is a spin-out of the University of Oxford. The founding team is Professor Alex Lvovsky of Oxford's Department of Physics, Dr Thomas Barrett, Dr Xianxin Guo, who is the main inventor of the optical training technology, and James Spall, who co-developed the core optical computing method. Its stated mission is to make computing faster and more sustainable and to make AI inference in data centres cheaper, more power efficient and more scalable. The company says the cost of today's large language models is unsustainable.

Lumai is based in Oxford. A 2022 interview places it at the Wood Centre for Innovation in Headington, which provides lab and office space, and later coverage describes it as an Oxford company. It is working with a specialist optical manufacturer to build its modules and with system integrators on full rack-scale systems. It has said it plans a North American office, so some work may eventually involve colleagues in the United States. Its early targets included autonomous driving, robotics, drug discovery and cybersecurity as well as data centres.

The company started small, with four full-time optical engineers in 2022, and a 2025 article reports more than 20 staff across electronics, software and optical engineering. Tim Weil joined as chief executive in early 2022 with about 25 years of experience in deep-tech companies, while a later article names Dr Xianxin Guo as CEO, Dr James Spall as chief technology officer and Phil Burr as head of product, so roles have changed over time. The company describes its approach as building on peer-reviewed work, including a 2020 paper on optical vector-matrix multiplication. These are the company's own statements.

A company like Lumai employs optical engineers, electronics and hardware engineers, software and machine-learning engineers, manufacturing and supply-chain specialists, and product, sales and operations staff. It is a good example of how physics research at a university can grow into a product team, and why physics, engineering, computing and maths all matter in new AI hardware.

What they do

• {'desc': 'Optical matrix multiplication using beams of light in free space.', 'title': '3D optical computing'}
• {'desc': 'Rack-mounted modules for data-centre AI inference.', 'title': 'AI accelerator modules'}
• {'desc': 'Patented method for training optical neural networks optically.', 'title': 'Optical training research'}

Key facts

• University of Oxford spin-out, formerly OxONN (Oxford Trust interview, 2022)
• Founding team: Alex Lvovsky, Thomas Barrett, Xianxin Guo and James Spall (Oxford Trust)
• Based at the Wood Centre for Innovation, Headington, in 2022 (Oxford Trust)
• More than 20 staff and a planned North American office (EPDT article, 2025)

Website: www.lumai.ai