Nvidia partners with AI startup Ineffable on RL systems

Nvidia said on May 13 that it will partner with London-based AI startup Ineffable Intelligence to build large-scale reinforcement learning systems.

The company was founded by former Google DeepMind scientist David Silver.

Ineffable raised a US$1.1 billion seed round in April from Sequoia and Lightspeed, with backing from Nvidia, Google, and other investors.

The companies said their engineers will build infrastructure for AI models that learn from experience rather than human data.

It will use Nvidia’s Grace Blackwell chips and Vera Rubin platform.

The deal adds to a wave of funding for new AI labs launched by former researchers from DeepMind, OpenAI, Anthropic, and xAI.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

The tie-up tackles the growing “data wall,” when the supply of high-quality human-made training data starts to run out 1.
David Silver, founder of Ineffable Intelligence, compares today’s dependence on human data to a finite “fossil fuel” 2.
His company backs reinforcement learning as a “renewable fuel,” with AI agents improving through trial and error in simulations 3.
That approach creates training data as it goes. It also puts different pressure on hardware such as memory bandwidth and interconnects, the links that move data between chips and systems 4.

The agreement goes beyond funding. By placing engineers inside the project to “codesign the infrastructure,” Nvidia gets an early look at the hardware and software large-scale reinforcement learning will need 5.
That gives Nvidia room to tune future systems such as its Vera Rubin platform, which could help it keep selling hardware if AI development moves past today’s leading training methods 5.
Pairing technical help with capital and computing access gives Nvidia an edge 6.
The setup gives Nvidia quick feedback from advanced AI work. Rival labs may find it harder to get the specialized infrastructure needed to keep up 7.