Neurophos OPU (www.neurophos.com)

Neurophos has unveiled its innovative Optical Processing Unit (OPU), which compresses the power and size of traditional GPU racks into a single unit while achieving an astonishing 0.47 ExaOPS—making it the world's most efficient AI chip. This breakthrough utilizes Photonic Tensor Cores that are 10,000 times smaller than conventional components, enabling unprecedented performance scalability. Protected by a robust portfolio of 300 patents, the OPU represents a leap in computational efficiency equivalent to 30 years of traditional transistor scaling, targeting the growing demand for Terawatt compute capabilities. The implications for the AI and machine learning community are significant. This technology not only reduces the physical footprint and power requirements—scaling down from over 1,000 nuclear plants' worth of power for GPUs to just 10 plants for OPUs—but also positions the OPU as a game-changer in the race towards high-performance computing. With evaluations expected to start in 2026 and the first systems anticipated in early 2028, the OPU could dramatically reduce costs and energy consumption in AI infrastructure, paving the way for more efficient, large-scale deployments in various applications.