Optical computing reaches commercialization inflection point: Lightelligence PACE 3 photonic chip with 256x256 matrix, first real-world deployed photonic computing at WAIC 2026. Published: July 27, 2026 Want to read in a language you're more familiar with? Lightelligence demonstrates first real-world deployed photonic computing at WAIC with Tianshu Light Cube access control system, announces PACE 3 chip with 256x256 photonic matrix for low-latency inference. WAIC 2026 marked optical computing shift to commercialization led by Lightelligence. The company demonstrated Tianshu Light Cube as first real-world photonic computing system in access control, and announced PACE 3 chip with 256x256 photonic matrix for low-latency inference. Founder Dr. Shen Yichen describes the stage as creating a hardware lottery for future applications, paralleling NVIDIA GPU gaming roots that enabled neural networks. Lightelligence places simulators in classrooms so next-gen algorithms grow on photonic hardware. The photoelectric hybrid architecture addresses the core bottleneck: while photonic computing offers bandwidth advantages, lower transmission loss, and natural parallel processing capability that theoretically breaks the power and bandwidth limits of electronic chips, practical deployment requires integration with electronic systems for control and data conversion. The primary barrier to scaled deployment has been a chicken-and-egg problem: downstream customers lack incentive to restructure algorithms and software for unverified hardware, while upstream supply chains lack market demand to amortize optical device manufacturing costs. Lightelligence approach combines optical interconnect, optical switching, and photonic computing capabilities across its product lines to create integrated photoelectric solutions that do not require customers to completely rebuild their software stack. PACE 3 is designed specifically for large model inference rather than training, leveraging photonic compute-in-memory architecture for matrix-vector multiplication the dominant computational pattern in neural network inference with lower latency and higher energy efficiency than electronic counterparts. The chip photonic matrix size deliberately targets what co-founder CTO Dr. Meng Huaiyu calls the sweet spot. While media and investor attention often focuses on matrix size as a metric, the practical reality is that large model computations require efficient coordination across matrices of different sizes, not a single oversized matrix. NVIDIA Tensor Core operates at 64x64 and Google TPU stabilized at 256x256 for years, validating the sweet spot principle. The WAIC forum focused on where to deploy first. Lightelligence argues photonic computing is not making a better key for GPUs but creating an entirely new ecosystem. The access control deployment proves long-term operational stability. As silicon photonics matures, costs follow the semiconductor trajectory for broader deployment.

(Yicai) April 28 -- Shares of Xizhi Technology, also known as Lightelligence, surged on their first day of trading in Hong Kong, with the Chinese company becoming the world's first artificial intelligence optical computing solutions developer to go public.