From invisibility cloaks to AI chips: Neurophos raises $110M to build tiny optical processors for inferencing

Neurophos, a photonics startup originating from Duke University, has successfully raised $110 million in a Series A funding round to advance its groundbreaking optical processors aimed at enhancing AI computing capabilities. Founded by a team that includes David R. Smith, who was known for creating a functional invisibility cloak using metamaterials, Neurophos is now tackling the pressing challenge of scaling up computing power in AI labs while minimizing energy consumption. The startup's innovation, a "metasurface modulator," boasts unique optical characteristics that allow it to function as a tensor core processor. This technology is essential for matrix vector multiplication, a fundamental operation in AI, particularly during the inference phase. Unlike traditional silicon-based GPUs and TPUs, Neurophos's optical processing unit can house thousands of these modulators on a single chip, significantly outperforming existing silicon GPUs in both speed and efficiency during AI tasks. Neurophos's optical chip is designed to operate at an impressive 56 GHz, achieving a peak performance of 235 Peta Operations per Second (POPS) while consuming only 675 watts. In contrast, Nvidia's B200 AI GPU provides a mere 9 POPS at 1,000 watts. This substantial difference in efficiency could make Neurophos a formidable competitor in the AI chip market, particularly as it prepares to launch its first products by mid-2028. The funding round was led by Gates Frontier, Bill Gates' investment firm, with participation from major players including Microsoft’s M12 and several venture firms. Despite entering a competitive landscape dominated by established companies like Nvidia, Neurophos is confident that its innovative approach to optical computing will create a significant competitive advantage. Dr. Patrick Bowen, CEO of Neurophos, emphasized the importance of energy efficiency in enhancing processing speed, stating, "If you want to go fast, you have to solve the energy efficiency problem first." The company’s approach could redefine the future of AI processing, as the technology is not only faster but also more efficient than the current silicon-based systems. As the demand for AI solutions continues to surge, the need for advanced computational efficiency becomes even more critical. Dr. Marc Tremblay from Microsoft has noted the urgent requirement for breakthroughs in computing technologies that match the advancements in AI models themselves, aligning perfectly with Neurophos’s mission to transform the landscape of AI processing.