NVIDIA partners with British startup

Nvidia is not just selling more chips here. It is picking a side in a pretty important argument about what comes after today’s large language models. On Wednesday, May 13, Nvidia said it will work with London-based Ineffable Intelligence, the new AI lab founded by former DeepMind star David Silver, on infrastructure for reinforcement learning at scale. The basic idea is simple — if current AI mostly learns by absorbing human-made text, this partnership is about building systems that learn more from experience itself. ### Who is the British startup? The company is Ineffable Intelligence, a U.K. lab founded in late 2025 by David Silver, a University College London professor and the former leader of DeepMind’s reinforcement learning team. Silver is one of the key people behind AlphaGo and later systems like AlphaZero — the projects that made “learn by playing” feel like more than a toy idea. Ineffable says its mission is to build a “superlearner” that can keep discovering knowledge and skills without leaning on human data as the main crutch. (blogs.nvidia.com) ### What did Nvidia actually agree to do? This is an engineering-level collaboration, not an acquisition and not a vague “strategic partnership” with no machinery behind it. Nvidia and Ineffable say their engineers will work together on the training pipeline needed to feed reinforcement-learning systems at very large scale. The work begins on Nvidia’s Grace Blackwell systems and is also meant to explore the upcoming Vera Rubin platform — basically a sign that Nvidia wants to shape the hardware-software stack for this style of AI before the field fully arrives. (cnbc.com) ### Why is reinforcement learning the point? Reinforcement learning means an AI system improves through trial and error. It acts, gets feedback, adjusts, and tries again. That is how AlphaGo got so strong. But it is much harder to do this in broad, messy environments than in games with neat rules. Silver’s pitch is that researchers have gotten pretty good at building systems that know what humans already wrote down, but the harder problem is building systems that discover genuinely new knowledge for themselves. (blogs.nvidia.com) Nvidia is now publicly backing that thesis. ### Why does Nvidia care beyond selling GPUs? Because if AI shifts from scraping human-produced data toward constant simulated experience, the compute problem changes. You need hardware and software that can generate, store, and learn from enormous streams of interaction, not just train once on giant text corpora. That is a huge opportunity for Nvidia. If the next wave of frontier AI depends on reinforcement-learning infrastructure, Nvidia does not want to be a parts supplier standing outside the lab — it wants to help define the lab. (cnbc.com) That is the real significance here. ### Why is David Silver such a big deal? Because he is not just another ex-Big Tech founder with a deck and a slogan. Silver spent years building the strongest proof that experience-driven learning can outperform hand-fed human knowledge in the right setting. Investors clearly think that track record matters. In April, Ineffable announced a $1.1 billion seed round at a $5.1 billion valuation, with Sequoia and Lightspeed co-leading and Nvidia, Google, Index, DST Global, and the U.K. (blogs.nvidia.com) Sovereign AI Fund also participating. For a company this young, that is an enormous vote of confidence. ### Why is the U.K. angle part of the story? Because Nvidia has already been leaning hard into Britain as an AI base. In September 2025, it announced a £2 billion commitment to the U.K. AI startup ecosystem, framing the country as a place where universities, researchers, startups, and compute could all line up. This new partnership fits that pattern almost perfectly — elite local research talent paired with Nvidia infrastructure and capital. (cnbc.com) ### What is the catch? The catch is that this is still a bet, not a breakthrough. Ineffable has a mission, a famous founder, huge funding, and now Nvidia’s backing — but no public product and no proof yet that reinforcement learning can scale into a general-purpose “superlearner” outside narrower domains. That is why this news matters. It shows where one of the most powerful companies in AI thinks the next hard problem is. (investor.nvidia.com) ### Bottom line? This partnership says something bigger than “Nvidia likes a startup.” It says the company thinks the next frontier may be AI that learns from doing, not just from reading. If that bet works, the center of gravity in AI could shift from bigger text models toward systems that generate their own experience — and Nvidia wants to build the machinery underneath that shift. (blogs.nvidia.com) (ineffable.ai)