A few AI labs have pulled away from the pack over the last few years, and they might be able to maintain this advantage going forward.
Google DeepMind CEO Demis Hassabis believes that the leading players in the artificial intelligence space are creating a self-reinforcing cycle of improvement. Speaking on the 20VC podcast, Hassabis argued that the current crop of top-tier labs is beginning to distance itself from the competition by using their current models to architect the next.
“There are three or four leading labs now, of which we are one,” Hassabis noted. “I think the gap is starting to pull away because a lot of these tools also, of course, help you build the next generation. So things like coding tools and math tools—it’s getting harder and harder, I would say, to eke out the same gains from just the same ideas.”
Hassabis suggests that as the industry moves past the low-hanging fruit of the current architectural paradigm, the ability to innovate at the fundamental level will become the primary differentiator.
“I think those labs that have the capability to invent new algorithmic ideas are going to start having a bigger advantage over the next few years, as the last set of ideas are sort of having all the juice rung out of them,” he explained. “Give it a tool so that it can get the context it needs, and you’re just going to get better results.”
The implications of this “flywheel effect” are already becoming visible in the industry. For example, DeepMind recently demonstrated this recursive improvement with AlphaEvolve, an evolutionary coding agent powered by Gemini that discovered new algorithms to solve decade-old mathematical puzzles. By using AI to optimize the very code and math that underpins AI, frontier labs can accelerate their development speed far beyond what human engineers could achieve alone.
While this creates a formidable barrier to entry, the competitive landscape remains in flux. While labs like OpenAI and Anthropic continue to trade blows on leaderboards, the high cost of this “recursive” R&D—often involving thousands of specialized GPUs—may eventually consolidate the “frontier” into an even smaller, more exclusive circle. As Hassabis suggests, the era of “ringing the juice” out of old ideas is ending; the next phase of AI supremacy will be won by those who can build the best tools to invent the next ones.