# UK ARIA Tests Lumai Optical AI Servers in Data Centres Advanced Research and Invention Agency evaluates light-based computing for AI inference efficiency. - **Source:** [The Next Web](https://brifing.si/source/afe24b20-b275-4c19-b9a6-8bb3c8babae7) - **Published:** 2026-11-03 - **Topic:** [Business](https://brifing.si/c/business) - **Original article:** - **On Brifing:** https://brifing.si/s/b070406b-b389-4b7d-b567-e9596259637b ## Optical Computing Pilot Program The UK’s Advanced Research and Invention Agency is deploying Lumai’s Iris servers within its Scaling Inference Lab. This initiative marks the first optical computing platform integration for the lab, which is backed by £50 million in funding. The project aims to evaluate how light-based hardware performs under real-world AI model workloads. ## Technical Mechanism of Lumai Lumai, an Oxford University spin-out established in 2021, utilizes light beams to perform complex AI mathematical calculations. By replacing traditional transistor switching with optical processing for heavy computations, the system offloads specific tasks. Standard electronic components remain responsible for the remaining operational functions, creating a hybrid architecture for modern data centres. ## Targeting AI Inference Prefill The hardware specifically targets the prefill stage, which occurs when an AI model initially processes a user prompt. While the optical chips handle the initial computational heavy lifting, standard silicon chips continue to generate the final output. This division of labor aims to optimize the specific phase where latency often accumulates. ## Energy Efficiency Claims Lumai claims its technology could reduce data centre power consumption by up to 90 percent. Furthermore, the company asserts the system is 16 times more energy-efficient per request than silicon-only setups. These figures currently remain unverified, making independent validation a primary objective for the Scaling Inference Lab’s upcoming testing phase. ## Lab Infrastructure and Partners The Scaling Inference Lab is operated by the nonprofit organization CommonAI. Beyond Lumai, the facility already hosts Oriole, another UK startup focused on using light to interconnect AI chips. These deployments occur within a commercial data centre environment to ensure the hardware faces authentic cooling and power constraints during the evaluation. ## Strategic Leadership Perspective David Lewis, the chief science and technology officer at ARIA, identified Lumai’s technology as the most mature optical compute platform evaluated by the agency. This endorsement highlights the strategic importance of optical hardware in the UK's broader research agenda. The agency is prioritizing scalable solutions to address the rising energy demands of AI. ## Competitive Landscape Analysis The optical computing sector is seeing significant activity, with Salience Labs recently raising 125 million dollars for optical switch development. Meanwhile, Nvidia is investing billions into light-based data centre connections to improve throughput. Additionally, the UK inference-chip startup Fractile is currently raising capital at a valuation of 6.5 billion dollars. ## Future Implications for AI The success of these tests could redefine how data centres manage the massive energy footprint of generative AI models. By moving away from purely silicon-based architectures, the industry seeks to bypass current thermal and power limitations. The lab’s findings will provide the first independent benchmark for optical computing’s viability at scale. ## Sources covering this story - https://thenextweb.com/news/aria-scaling-inference-lab-lumai-optical-servers — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._