Computing With Light Photonic Processor NPU 2.0 Aims to Solve the Energy Problem

Related Vendors

Q.ANT GmbH

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

With the second generation of its photonic processor, Q.ANT is launching the NPU 2.0. Energy consumption is said to be 30 times lower than CMOS processors, and performance is 50 times higher.

The rapid development of artificial intelligence is hitting a fundamental limit: energy consumption. Modern AI applications like ChatGPT already require over 564 megawatt-hours daily. This corresponds to 100 times the annual consumption of a German household. GPU-based data centers now consume up to 40% of their energy solely for cooling the overheated hardware. While computing power is increasing exponentially, energy demand is growing even faster. This development is unsustainable and puts the entire AI revolution at risk.

Against this background, the Stuttgart-based company Q.ANT presents an exciting approach: the second generation of its photonic Native Processing Unit (NPU 2.0). This technology promises to fundamentally solve the energy problem of AI by computing with light instead of electrons.

Light Instead of Silicon: A Physical Paradigm Shift

The basic principle of photonic data processing breaks the physical laws of conventional semiconductor technology. While CMOS processors perform complex mathematical operations through millions of transistor circuits, photonic processors utilize the properties of light for native analog computations. Light not only moves faster than electrons but also generates almost no heat and requires significantly less energy for signal transmission.

"For years, AI development has been advancing rapidly, while we have struggled to keep up with its energy supply. Energy is the new challenge in performance enhancement," explains Dr. Michael Förtsch, CEO of Q.ANT. With the NPU 2.0, the company demonstrates up to 30 times lower energy consumption with 50 times higher performance for complex AI and HPC workloads compared to traditional silicon processors.

Enhanced Nonlinear Processing Opens Up New Application Fields

The second generation of the Q.ANT NPU offers significant improvements over the first version. The enhanced nonlinear processing core was specifically optimized for neural network models, reducing the required number of parameters and training depth. At the same time, accuracy increases in image learning, classification, and physics simulation.

These properties open up entirely new application fields that are not achievable with digital circuits. These include physical AI, robotics, next-generation computer vision, industrial intelligence, physics-based simulation, as well as data analysis and automatic pattern recognition. Particularly in manufacturing, logistics, and inspection, photonic processors can execute nonlinear neural networks far more efficiently than conventional systems.

Immediate Integration into Existing Infrastructures

A key advantage of the Q.ANT solution is its seamless integration into existing data center environments. The Native Processing Server (NPS) is delivered as a turnkey 19-inch rack-mountable server and contains multiple second-generation NPUs. It can easily be integrated with CPUs and GPUs in existing HPC infrastructures via PCIe and C/C++/Python APIs.

This compatibility allows companies to leverage the advantages of photonic processing without needing to overhaul their entire IT infrastructure. The Q.PAL library (Q.ANT Photonic Algorithm Library) provides developers with efficient nonlinear algorithms and functions for complex workloads.

The advances in photonic data processing are remarkable. While digital data processing required decades for its development, Q.ANT achieved the leap from simple digit recognition to image classification and image learning within just one year. "What took ten years in digital data processing, we achieved with photonics in just one year," emphasizes Dr. Förtsch.

This speed demonstrates the enormous potential of the technology and suggests that photonic computing scales faster than conventional CMOS technology. At Supercomputing 2025 in St. Louis, Q.ANT is showcasing live how the NPU 2 learns images within seconds using a nonlinear neural network.

Economic And Ecological Impacts

The drastic reduction in energy consumption not only makes AI applications more sustainable but also more economical. Computer vision systems, previously considered too computationally intensive, suddenly become profitable with photonic processors. In manufacturing, error detection systems, object tracking, and inventory optimization can be implemented with fewer parameters, significantly reducing energy costs.

At the same time, photonic processors open up new possibilities for hybrid AI models that combine statistical logic with physical modeling. This enables advancements in areas such as drug research, material design, and adaptive optimization, where both nonlinear complexity and extreme energy efficiency are crucial.

The Q.ANT servers with the latest processor generation NPU 2 are now available for order and can be delivered in the first half of 2026. The turnkey servers can be used in any data center and mark the beginning of a new era of sustainable, high-performance computing. (heh)

Future-Proof Authentication with Universal RFID Readers

Universal RFID Readers: One Solution for All Credentials