Neurophos

Neurophos is a photonic AI inference hardware company founded in 2020 and spun out of Duke University. It develops silicon photonics and programmable metasurfaces designed to perform matrix multiplications optically - at the speed of light - with the aim of replacing electronic computing for power-constrained AI inference workloads. The company's stated focus is the rising energy demand of AI data centres; its engineering approach is to rethink the underlying hardware rather than to optimise existing electronic architectures.

Its flagship product is the T100 Optical Processing Unit (OPU), an optical processing unit built on metamaterials that shrink optical components by a factor of 10,000 relative to traditional silicon photonics. This density is claimed to enable exaflop-scale AI inference performance on standard CMOS manufacturing lines. Neurophos quotes 235 TOPS/W for the T100, against 7 TOPS/W for the NVIDIA B200 - a stated 30x advantage in performance per watt.

The relevant technical domains are optical computing, programmable and active metasurfaces, metamaterials, optical matrix multiplication, chip design and energy-efficient computing. Target industries include AI infrastructure and data centres, hyperscale cloud providers, semiconductor manufacturing and high-performance computing.

Neurophos raised a $110m Series A from five named investors: Gates Frontier, M12, Aramco Ventures, Bosch Ventures and Space Capital. Founding technical leadership includes CTO Tom Driscoll alongside co-founders Patrick Bowen and Andrew Traverso. The company is based in the United States and traces its roots to Duke University and Metacept Inc.