Recursive Superintelligence is a frontier AI research lab founded in London in late 2025 with a singular and audacious goal: to build an AI system that improves itself. The team believes continuous self-improvement — rather than the prevailing paradigm of large, static pretraining runs followed by fine-tuning — is the path toward more capable and ultimately superintelligent systems. The company name reflects the thesis that recursive self-enhancement is the key mechanism that could compound capability over time.
The company was co-founded by Tim Rocktäschel, a professor of AI at University College London and a former scientist at Google DeepMind, and Richard Socher, the former chief scientist at Salesforce and founder of You.com. Within a few months it assembled an unusually senior team of roughly 20 people, including former OpenAI researchers Josh Tobin, Jeff Clune, and Tim Shi, alongside talent from Google and Meta. The concentration of well-known researchers signaled to investors that the lab could compete for frontier talent against far larger players.
In April 2026, just about four months after incorporation, Recursive Superintelligence raised a $500 million Series A at a $4 billion valuation. The round was led by Google Ventures with participation from Nvidia, and was so heavily oversubscribed that reported total commitments could reach $1 billion. The deal exemplified a 2026 trend in which frontier AI labs founded by marquee researchers command enormous valuations within months of launch.
The company's research direction is deliberately unconventional and, by its founders' own framing, difficult to fully explain to outsiders. Rather than chasing incremental benchmark gains on existing model families, it focuses on architectures and training loops that allow an AI to continually learn from its own outputs and interactions. This places it in the same emerging cohort as other self-improvement and continuous-learning labs that captured outsized investor attention in early 2026.
As a very young, research-heavy organisation, Recursive Superintelligence has no public product and operates at the speculative edge of AI. Its progress is closely watched as a test of whether self-improving systems can move from theory toward demonstrable, compounding capability — and whether the lab can convert elite talent and abundant capital into a durable technical lead.