Nvidia has long been the primary arms dealer in the generative AI gold rush, but its latest partnership signals a strategic pivot toward the next great technical hurdle. The chipmaker has announced a deep collaboration with Ineffable Intelligence, a secretive startup founded by David Silver, the visionary architect behind DeepMind’s AlphaGo. This venture aims to move past the limitations of Large Language Models (LLMs) by building the infrastructure for what Nvidia CEO Jensen Huang calls 'superlearners.'
Unlike current AI models that rely on massive, static repositories of human-generated text and images, Ineffable Intelligence is doubling down on Reinforcement Learning (RL). This approach enables AI to learn through trial and error, effectively allowing systems to 'experience' their environments and discover new knowledge that is not yet part of the human record. David Silver argues that while the industry has mastered imitating human knowledge, the future lies in systems that can autonomously solve problems where no human blueprint exists.
This shift from imitation to discovery carries profound implications for computing hardware. While training LLMs is a linear process of data ingestion, Reinforcement Learning requires a high-frequency loop of action, observation, and self-correction. To support this, Nvidia is integrating Ineffable Intelligence’s workflows into its Grace Blackwell and forthcoming Vera Rubin platforms. These systems are designed to handle the extreme memory bandwidth and real-time data generation required by agents that learn on the fly.
The financial backing for this vision is staggering, reflecting a broader trend of 'talent flight' from established labs like DeepMind, Meta, and OpenAI. Ineffable Intelligence secured a massive $1.1 billion seed round led by Sequoia Capital and Lightspeed, joined by Nvidia and various sovereign funds. This capital influx into new AI laboratories, including rivals like Recursive Superintelligence and AMI Labs, suggests that the market is already looking beyond the transformer-based architecture that defined the ChatGPT era.