Lumai has launched the Iris Nova server, described as the first optical computing system capable of running billion-parameter large language models in real time. The development marks a major step forward in AI infrastructure by addressing performance and energy efficiency challenges in large-scale computing.

Unlike traditional silicon-based systems, Iris Nova uses light (photonic computing) for processing, enabling faster inference speeds while significantly reducing power consumption. The system reportedly achieves up to 90% lower energy usage compared to conventional architectures, helping ease the increasing pressure on data centers as global energy demand for computing continues to rise.

Figure 1. Lumai Launches First Optical Computing System

According to Lumai CEO Dr. Xianxin Guo, shifting computation from electrons to photons enables substantial performance gains while dramatically improving energy efficiency, signaling a transition toward a post-silicon era in AI computing. Figure 1 shows Lumai Launches First Optical Computing System.

Luma Iris Nova Server Achieves Real-Time Billion-Parameter LLM Inference

Lumai’s Iris Nova server introduces an optical computing approach that replaces traditional silicon-based acceleration with light-based processing to meet the growing demands of AI workloads and data centers.

Designed to address what the company calls the “Energy Wall,” the system delivers faster inference performance and improved efficiency while reducing energy consumption by up to 90% compared to conventional architectures. This development comes as global data center power demand is projected to double by 2030, according to industry forecasts.

Originating from research at the University of Oxford in 2021, the technology uses three-dimensional optical computation to bypass the limitations of two-dimensional silicon chips, enabling massive parallelism and high-throughput token processing. The Iris Nova server combines a hybrid design, pairing digital control systems with an optical tensor engine for core computations, allowing integration into existing data center infrastructure.

Supported by interest from organizations such as the Advanced Research and Invention Agency (ARIA), the system is currently being evaluated by hyperscalers, enterprises, and research institutions as a potential alternative path for scaling AI compute beyond traditional electronic processors.

The drive toward more powerful artificial intelligence is increasingly limited not by software advances, but by the physical constraints of silicon-based hardware. Conventional processors are reaching bottlenecks in performance scaling, power efficiency, and heat management, requiring far greater energy for only incremental gains.

In response, this new computing approach replaces electrons with photons, using light to perform computations. By exploiting spatial parallelism in a three-dimensional structure, it enables massive parallel processing, allowing millions of operations to be carried out simultaneously within a compact optical volume.

University of Oxford Research Powers Lumai’s AI Infrastructure Breakthrough

Lumai originated as a spinout from optical computing research at the University of Oxford in 2021, with the goal of addressing the growing computational demands of artificial intelligence infrastructure.

The company’s technology fundamentally rethinks AI computation by using light instead of electricity, performing calculations within a three-dimensional optical medium [1]. This approach overcomes the limitations of traditional two-dimensional silicon chips by enabling massive spatial parallelism, allowing millions of operations to run simultaneously and significantly improving throughput for large AI workloads.

This innovation comes at a time when data centers are under increasing strain from rising energy requirements—often referred to as the “Energy Wall”—and the slowing efficiency gains of conventional silicon-based scaling.

References:

1. https://quantumzeitgeist.com/lumai-launches-the-worlds-first-optical-computing-system/

Cite this article:

Janani R (2025), World’s First Optical Computing System Launched by Lumai, AnaTechMaz, pp.196