Recursive co-founder Tim Rocktaschel predicts self-improving AI in two years

Tim Rocktaschel, co-founder of Recursive Superintelligence, says his company can build recursive self-improving AI within roughly two years.

Recursive Superintelligence, a London-based startup founded in mid-May 2026, has raised $650 million in funding at a valuation of approximately $4.65 billion. That round could swell to $1 billion. The investor list includes GV (formerly Google Ventures), Nvidia, and AMD.

What “recursive self-improving AI” actually means

Most AI today gets better because humans make it better. Researchers tweak architectures, gather more training data, fine-tune reward signals. Recursive self-improvement flips that dynamic. The idea is to build AI systems that can autonomously enhance their own capabilities, iterating on themselves without human intervention.

Recursive’s approach draws inspiration from two domains: evolution and automated scientific discovery. Instead of relying solely on traditional scaling methods, the company is betting that iterative self-improvement represents a different path to superintelligence.

The team behind the bet

Before co-founding Recursive, Rocktaschel served as a principal scientist at Google DeepMind, where he led self-improvement research efforts. The founding team includes Richard Socher as another co-founder, with the broader roster pulling from DeepMind, Salesforce, and other prominent tech firms. As of mid-2026, Recursive has fewer than 30 employees and has not released a public product.

A $4.65 billion valuation for a company with fewer than 30 employees and no public product works out to roughly $155 million per employee in implied value.

The involvement of Nvidia and AMD as investors adds another layer. Both companies sell the hardware that powers AI training and inference. Their participation suggests they see recursive self-improvement as a potential catalyst for even more compute demand, as self-improving systems might require more computational resources if each improvement cycle involves retraining or re-evaluating the system at scale.

The key metric over the next two years isn’t whether Recursive ships a product. It’s whether the company can demonstrate measurable, autonomous improvement in a controlled setting. The broader question is whether two years is an ambitious but realistic timeline. Rocktaschel’s track record at DeepMind lends credibility, but $650 million, and potentially $1 billion, is riding on Rocktaschel being right on schedule.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.