Are We Entering a New AI Race Beyond Traditional LLMs?
Over the past few months, investors have started placing serious bets on AI systems built around reinforcement learning, world models, simulation environments, and self-improving architectures.
Just look at the headlines:
David Silver, the researcher behind AlphaGo, AlphaZero, and AlphaStar, raised $1.1 billion at a $5.1 billion valuation for Ineffable Intelligence.
Yann LeCun, Turing Prize winner and former Meta Chief AI Scientist, raised $1.03 billion for AMI Labs.
Recursive Superintelligence, founded by a UCL professor and former scientist at Google DeepMind, raised $500 million at a $4 billion valuation.
That’s over $2.6 billion invested into three labs with no product and no revenue.
The underlying idea is simple. Current AI systems learn mostly from human-generated data. LLMs are like libraries: vast systems built from everything humans have written, said, and created. However, scaling human-generated data alone may eventually hit limits.
So a parallel race has started.
These labs are building AI that learns the way a laboratory works — through experimentation. Trying, failing, adjusting, trying again. The goal is AI that can discover things humans never taught it and reach places human data doesn’t cover.
LLMs learned to read the map humans drew. This generation is learning to explore without the map.
As an investor, a few things stand out.
First, the founder is the entire bet at this stage. Silver is arguably the most celebrated applied reinforcement learning researcher in the world. LeCun is one of the godfathers of deep learning. These are the people who built the field. The valuation reflects scientific credibility and track record. There is almost nothing else to price yet.
Second, these are long-horizon bets. Nobody is expecting a product — or returns — anytime soon. Frontier AI research has always been expensive, uncertain, and slow-moving until suddenly it isn’t.
Ineffable Intelligence puts the ambition plainly on its website:
“If successful, this will represent a scientific breakthrough of comparable magnitude to Darwin.”
That’s the scale these researchers are working at.
One more thing worth noting is where these labs are being built. AMI Labs is headquartered in Paris, with teams in New York, Montreal, and Singapore. Ineffable Intelligence and Recursive Superintelligence are in London.
The AI race has been heavily US-centric for the past decade. It now looks like the foundational research driving this next wave is coming from a broader geography.
That’s a shift worth watching.
Are we entering a new phase of AI — beyond traditional LLMs?
Hit reply and tell me what you think.
The Internet Has a New User
We’ve spent years optimizing websites for humans and search engines, then for AI. Now there’s a new audience emerging: AI agents.
And agents “see” the internet very differently from people.
A beautiful website can be almost unusable for an AI agent trying to complete a task.
Google recently published an interesting piece on this shift, calling it the move from UX to AX. The information we publish today needs to be readable not just by search crawlers and LLMs, but also by agents.
An agent researching a task doesn’t care how your navigation is designed. It needs to understand:
what your product does;
how systems connect;
what actions are possible;
how confidently it can navigate the workflow.
This creates an entirely new layer of infrastructure around the internet.
I expect we’ll see a generation of tools and startups focused specifically on agent-ready systems — platforms that restructure company data for AI discovery and optimize workflows for agent interaction.
We’re already working with a startup in this space, building exactly that kind of infrastructure for developers.
This reminds me of the mobile transition.
At first, mobile-friendly design felt optional. Then it became impossible to ignore. Later, it simply became part of the standard.
Agent-ready infrastructure is likely following the same path.
The question is how fast.