NVIDIA has announced a new engineering-focused collaboration with Ineffable Intelligence to accelerate the development of large-scale reinforcement learning infrastructure for next-generation AI systems.

The partnership comes shortly after Ineffable Intelligence emerged from stealth mode under the leadership of AI researcher David Silver, widely recognised as one of the architects behind AlphaGo and a pioneer in reinforcement learning technologies.

The collaboration will focus on building advanced infrastructure capable of supporting “superlearners” AI systems that continuously improve through experience and trial-and-error learning. Unlike traditional AI training methods that rely heavily on fixed human-generated datasets, reinforcement learning systems dynamically generate data through interactions with their environments, requiring highly optimised computing architectures and real-time feedback loops.

Jensen Huang, founder and CEO of NVIDIA, said the next frontier of artificial intelligence lies in systems that can independently discover new knowledge rather than simply reproducing existing human understanding. He noted that the collaboration aims to co-design hardware and software infrastructure capable of supporting reinforcement learning at unprecedented scale.

According to David Silver, current AI systems have largely mastered learning from existing human knowledge, but future breakthroughs will depend on systems that learn directly from experience. He added that this transition demands entirely new approaches to model architectures, training algorithms, and computational infrastructure.

Engineering teams from both companies are now working on reinforcement learning pipelines built initially on NVIDIA’s Grace Blackwell platform, with future exploration planned on the upcoming NVIDIA Vera Rubin architecture. The initiative aims to optimise interconnects, memory bandwidth, and serving systems required for complex reinforcement learning workloads.

The companies believe the collaboration could help unlock large-scale AI systems capable of discovering scientific, industrial, and technological breakthroughs through autonomous learning and simulation-based training environments.