The growing impact of AI on energy consumption is becoming clear to more people. Analysts from Goldman Sachs predict that the energy demand for AI data centers will increase by 160% by 2030. According to Gartner, this could lead to up to 40% of existing AI data centers facing operational limitations due to power shortages.
The Stuttgart-based start-up Q.ANT aims to change this. Their “photonische Native Processing Unit” (NPU), designed for AI applications and complex simulations, is mounted on a PCI-Express card, making it fully compatible with existing server environments. The company claims that it has at least 30 times higher energy efficiency and significant improvements in computing speed compared to traditional CMOS technology.
Tests and simulations for image recognition show that Q.ANT’s NPU can train models significantly faster and achieve better results with fewer parameters. It also offers faster solutions for partial differential equations in physics simulations, simplifies time series analysis, and enhances efficiency in solving graph theory problems. However, scientific publications to support these claims are not yet available.
Michael Förtsch, CEO of Q.ANT, states, “Silicon is great for building transistors, but it’s not suitable for optics. That’s why I proposed the thin-film lithium niobate material system, though everyone said it would never enter the foundries of the world. My response was, let’s build the pilot lines ourselves. We did, and we set up our first pilot line in Stuttgart, showing we can build these chips.” He adds that the company can now produce a new generation of chips every three months because they have full control over the production process. “It’s possible to make a mark in logic technology in Germany again.”
Photonics technology is innovative because photonic chips compute with light instead of electrons. This means there are no electrical losses or heat generation in the chips, making them an energy-efficient alternative. However, despite startups like Lightintelligence and Lighton offering initial photonic chips, the technology has not yet been widely successful.
Q.ANT also coordinates the PhoQuant funding program of the Federal Ministry of Education and Research, where 13 partners are researching photonic quantum computers. This has led to Europe’s largest sampling-based quantum computer, PaQS (Paderborn Quantum Sampler), now operational at the University of Paderborn. A second device is expected to start operations soon at the Fraunhofer Institute for Optics and Precision Engineering (IOF) in Jena.