Q.ANT demonstrates AI on photonic hardware
Q.ANT has demonstrated generative AI and recurrent neural network workloads running on its second-generation photonic processor, highlighting the potential of photonic computing for future AI infrastructure.
Q.ANT has showcased the execution of complex AI workloads on its second-generation Native Processing Unit (NPU), demonstrating both a diffusion model and a recurrent neural network at ISC High Performance 2026 in Hamburg.
The demonstrations mark a significant milestone for photonic computing, showing that the company's hardware can support modern AI applications including generative image synthesis and time-series prediction.
For generative AI, Q.ANT ran a diffusion model for image-to-image synthesis, a workload that relies on repeated matrix operations and is widely used in image generation applications.
The company said the demonstration represents one of the most advanced diffusion models yet executed on photonic hardware.
Q.ANT also demonstrated NXAI's TiRex time-series prediction model, based on the xLSTM architecture.
The recurrent neural network is designed for applications such as financial forecasting, supply chain optimisation and weather prediction.
According to the company, its photonic architecture performs matrix operations using light rather than electronic transistors, targeting up to 30 times greater energy efficiency at the photonic circuit level compared with conventional processors.
The demonstrations build on earlier ecosystem developments, including the successful deployment of a PyTorch-based object detection model on Q.ANT hardware by software partner Daisytuner and recent commercial hardware orders through cloud provider IONOS.
Q.ANT said the latest results demonstrate the growing maturity of photonic computing platforms and their potential role in addressing the power and performance challenges facing next-generation AI infrastructure.
