Recursive Superintelligence, a London-based AI lab operating in stealth mode until now, has secured $650 million in a funding round that values the company at $4.65 billion. The round was led by GV (formerly Google Ventures) and Greycroft, with additional participation from AMD Ventures and NVIDIA. Previous reports had indicated the company was seeking at least $500 million at a $4 billion pre-money valuation, with the round ultimately oversubscribed and potentially heading toward $1 billion before closing.
Founders with deep AI pedigree
The startup was founded in 2025 by a group of prominent researchers and engineers drawn from the world’s leading AI organisations. The co-founders include Richard Socher, former chief scientist at Salesforce and founder of You.com, who is recognised for foundational work in natural language processing, including word vectors, contextual vectors and prompt engineering, with over 180,000 academic citations. Tim Rocktäschel, a professor of AI at University College London and former scientist at Google DeepMind, is also on the founding team, alongside Jeff Clune, Josh Tobin and Tim Shi, all of whom previously worked at OpenAI. The team — which also includes alumni from Meta AI and Uber AI — currently numbers fewer than 30 researchers and engineers, operating from offices in San Francisco and London.
The concentration of former OpenAI staff reflects a broader industry trend often referred to as the “OpenAI mafia”, with former employees of the organisation collectively raising billions of dollars in venture funding for new AI ventures. Recursive Superintelligence intends to use the fresh capital to secure large-scale compute infrastructure and to run its first “Level 1” autonomous training system.
Automating AI research: the core technology
Recursive Superintelligence’s central technological thesis is that the next major leap in artificial intelligence will come not from simply building larger foundation models, but from automating the research process itself. The company is developing systems capable of improving their own architecture, training methods, evaluation processes and research direction without continuous human oversight.
The approach is explicitly compared by the company to biological evolution, where discoveries accumulate over time to produce increasingly advanced forms of intelligence. The aim is to build software that continuously generates and refines new capabilities in an open-ended cycle — creating systems that not only learn tasks but independently discover better ways to learn. This contrasts sharply with the current focus of many leading AI companies, which remain concentrated on scaling large foundation models.
Recursive’s philosophy places it in a rapidly growing competitive field. AMI Labs, founded by Yann LeCun, is exploring world models. Ineffable Intelligence, founded by David Silver, focuses on reinforcement learning; NVIDIA has recently partnered with Ineffable to build reinforcement-learning agents. Safe Superintelligence is pursuing safety-first superintelligence research. Recursive Superintelligence, by contrast, is attempting to automate the entire AI development pipeline — from research direction to training and evaluation — in a single, self-improving loop.
Strategic investors and wider context
The participation of both NVIDIA and AMD Ventures in the round signals strategic interest in the AI hardware and software ecosystem. NVIDIA has committed significant capital to AI equity investments in 2026, including a substantial investment in OpenAI, and has also announced a global NOVA initiative for startups, committing £2 billion to catalyse the UK’s AI startup ecosystem. GV, as the venture capital arm of Alphabet Inc., operates independently but leverages Alphabet’s resources and typically holds a long-term investment horizon. Greycroft, founded in 2006, has shown a strategic shift towards AI startups since 2023.
The company plans a public launch in mid-2026, with ambitions extending beyond AI research into broader scientific discovery over time. Recursive Superintelligence intends to use the funding to scale compute infrastructure and research operations, with a public launch targeted for mid-May 2026.
