Chinese scientists have unveiled LightGen, an all-optical computing chip that demonstrates a significant leap in performance and energy efficiency compared to conventional AI processors, particularly for generative tasks. The chip, developed jointly by researchers at Shanghai Jiao Tong University and Tsinghua University, reportedly surpasses the capabilities of Nvidia’s leading A100 chip by over 100 times in speed and efficiency, according to research published in the journal Science on December 19, 2025.
Unlike traditional AI chips that rely on the flow of electrons through silicon transistors, LightGen utilizes photons – particles of light – to perform computations. This approach, known as optical computing, leverages optical interference to execute calculations, resulting in substantially faster processing speeds and reduced energy consumption. Professor Chen Yitong, leading the research at Shanghai Jiao Tong University, stated that LightGen “provides a new way to bridge the new chip architectures to daily complicated AI without impairment of performance and with speed and efficiency that are orders of magnitude greater, for sustainable AI.”
The development of LightGen comes as conventional silicon-based AI chips face increasing limitations in power consumption, scaling, and cost. Generative AI models, which are capable of creating realistic images and videos, demand immense computing power and energy. As silicon architectures approach physical limits, researchers have been exploring alternative computing paradigms, with optical computing emerging as a promising solution. The LightGen chip integrates more than 2 million photonic neurons on a single chip, enabling the generation of high-resolution images, including 3D scenes, and the creation of videos.
While LightGen demonstrates exceptional performance in specific generative tasks, It’s not intended as a direct replacement for NVIDIA-style GPUs for general-purpose computing. Instead, it represents a new computing architecture optimized for narrowly defined AI workloads, particularly those involving vision and generative image creation. The chip’s design allows for relatively effortless fabrication using older manufacturing processes, contrasting with the cutting-edge techniques required for producing advanced silicon chips.
Two distinct optical chip designs are currently being developed in China: ACCEL, a hybrid chip combining photonic components with analog electronic parts developed by Tsinghua University, and LightGen, the all-optical generative AI processor from Shanghai Jiao Tong University. Both chips aim to address the limitations of traditional electronic chips, but LightGen represents a more fundamental shift by eliminating electronic bottlenecks entirely.
The research team at Shanghai Jiao Tong University indicates that LightGen is scalable and could be further developed to enhance its capabilities. The implications of this technology extend beyond performance gains, potentially offering a viable path for scaling AI hardware beyond the constraints of Moore’s Law. However, the long-term impact and potential for widespread adoption remain to be seen.
