On November 28, 2025, Stuttgart-based company Q.ANT unveiled its second-generation photonic processor, the NPU 2.0, which promises to address the escalating energy demands of modern artificial intelligence applications. The NPU 2.0 reportedly consumes 30 times less energy than traditional CMOS processors while delivering 50 times higher performance, a significant advancement given that current AI models like ChatGPT require over 564 megawatt-hours daily—equivalent to the annual energy consumption of 100 German households.
The increasing energy consumption of GPU-based data centers, where up to 40% of energy is used for cooling, raises concerns over the sustainability of AI’s rapid expansion. Dr. Michael Förtsch, CEO of Q.ANT, emphasized that energy has become the new challenge in enhancing performance. The NPU 2.0 aims to overcome the energy limitations faced by the AI sector by utilizing light for computation rather than conventional electron-based systems.
This shift to photonic data processing marks a significant departure from traditional semiconductor technology. While CMOS processors rely on complex networks of transistors to perform calculations, photonic processors exploit the unique properties of light to execute calculations with minimal energy expenditure and heat generation. “For years, AI development has been advancing rapidly, while we have struggled to keep up with its energy supply,” said Förtsch.
The NPU 2.0’s enhanced nonlinear processing core has been optimized for neural network models, reducing the number of parameters and the necessary training depth while simultaneously increasing accuracy in tasks such as image learning and classification. These advancements open up new application fields previously unattainable with digital circuits, including physical AI, robotics, and advanced computer vision.
Additionally, Q.ANT’s photonic processors excel in the manufacturing sector. They can efficiently handle nonlinear neural networks for tasks like error detection, object tracking, and inventory optimization, significantly cutting down energy costs and complexity compared to conventional systems.
The NPU 2.0 is available as a turnkey solution, housed in a 19-inch rack-mountable server that can integrate seamlessly with existing data center infrastructures. It connects with CPUs and GPUs via PCIe and supports C/C++/Python APIs, allowing businesses to adopt this technology without overhauling their IT setups. The Q.PAL library provides developers with efficient algorithms tailored for complex workloads.
The rapid advancements in photonic processing are striking. In just one year, Q.ANT has progressed from basic digit recognition to sophisticated image classification and learning, a leap that took a decade in traditional digital processing. This swift evolution underscores the potential of photonic technology to scale faster than conventional methods.
At the upcoming Supercomputing 2025 event in St. Louis, Q.ANT will showcase the NPU 2.0’s capabilities live, demonstrating its ability to learn images in seconds using a nonlinear neural network. The implications of this technology extend beyond mere efficiency; they promise a more sustainable and economically viable future for AI applications.
The reduction in energy consumption not only makes AI applications more sustainable but also more economically feasible. With photonic processors, systems involving computer vision can now become profitable, paving the way for new implementations in industries such as manufacturing and logistics. Moreover, advances in hybrid AI models that integrate statistical logic with physical modeling stand to benefit from both energy efficiency and nonlinear processing capabilities, enhancing areas like drug research and material design.
Q.ANT’s NPU 2, now available for order, is set to be delivered in the first half of 2026, marking a pivotal moment in the evolution of sustainable, high-performance computing. As this technology gains traction, it could signify a transformative shift in how data is processed, enabling more efficient and innovative applications in various fields.
Pusan National University Study Reveals New Framework for AI Accountability Distribution
2026 Sees AI Agents Transform Industries, Quantum Threats Challenge Cybersecurity Resilience
HUMAIN Launches Qualcomm AI Engineering Center in Riyadh, Boosting 200 MW Data Capacity
Dell Raises FY26 Revenue Outlook to $111.7B, Driven by $12.3B AI Server Orders
IREN Partners with Microsoft for $9.7B AI Transformation, Aiming for 140,000 GPUs by 2026
