Neurophos employee Patrick Bowen explains why to investor Benjamin Levy of Geometry

Neurophos is building AI inference chips that are two orders of magnitude faster and more energy-efficient than state-of-the-art GPUs. Its Optical Processing Unit uses a metasurface compute architecture with tunable optical elements roughly 10,000x smaller than conventional photonic approaches, delivering a step-change in compute density, speed, and power efficiency. This breakthrough targets the central constraint in AI infrastructure: manufacturing, powering, cooling, and deploying enough compute for hyperscale inference. By increasing compute per chip, per wafer, and per megawatt, Neurophos creates a fundamentally steeper scaling curve for AI. The company has strong strategic and commercial momentum. Microsoft engaged deeply and invested meaningfully in Neurophos’ 3x oversubscribed $110M Series A. Neurophos is engaged with most hyperscalers and has commercial negotiations underway. Neurophos has completed its first compute demonstration with a metasurface array and is taping out additional devices with improved reliability and performance. The company is working with the leading foundries, custom ASIC designers, optics suppliers, and packaging partners. Neurophos has tripled its headcount since closing its $110M Series A in January 2026, adding senior leaders from NVIDIA, Intel, Infinera, Qualcomm, Magic Leap, and other leading semiconductor and photonics companies.