ARIA will deploy Lumai’s Iris servers as the lab’s first optical computing platform. The Scaling Inference Lab is funded with £50 million by ARIA and is run by the nonprofit CommonAI. The setup involves installing Lumai’s light-based AI processors across multiple machines in a commercial data centre to evaluate how they handle real AI models, power consumption, and cooling requirements. Lumai, a spin-out from Oxford University founded in 2021, uses beams of light to perform AI maths while conventional electronics handle other tasks. Salience Labs, another Oxford-origin project, recently raised $125 million for optical switches. Lumai claims its approach could reduce data centre power usage by up to 90% and that the system is at least 16 times more energy efficient per request than silicon alone when paired with standard chips, though none of these figures has been independently verified yet. The lab already hosts Oriole, a UK startup that uses light to link AI chips. Nvidia has invested heavily in light-based intra-data-centre connections, and Fractile, another UK inference-chip startup, is reportedly seeking substantial funding. The project aims to test Lumai’s technology in a real-world setting rather than a lab-only environment.