Recursive Superintelligence has raised $650 million at a $4.65 billion valuation in a funding round led by GV and Greycroft, according to the company.
The round also included participation from AMD Ventures and NVIDIA, placing the stealth-mode AI lab among a growing group of venture-backed companies pursuing autonomous AI systems.
Founded in 2025, Recursive Superintelligence was started by former leaders and researchers from OpenAI, Google DeepMind, Meta AI, Salesforce AI and Uber AI.
The company’s founding team includes Richard Socher, Tim Rocktäschel, Jeff Clune, Josh Tobin, and Tim Shi.
Recursive Superintelligence is developing self-improving AI systems designed to automate parts of the research process, including model architecture, training methods, evaluation, and research direction.
The company argues that the next stage of AI development will require systems that can improve how they learn, rather than relying only on larger models and greater computing power.
Its approach is based on building software that can generate, test, and refine new capabilities in a continuous cycle with less direct human supervision.
The company compares the process to biological evolution, where improvements accumulate over time and lead to more advanced forms of intelligence.
The funding comes as private capital continues to flow into AI labs pursuing artificial general intelligence, autonomous agents and research automation, even as venture investors remain more selective in other technology sectors.
The deal highlights how AI funding is moving beyond enterprise software and into capital-intensive research labs that require large amounts of compute and technical talent.
Recursive Superintelligence currently operates from San Francisco and London and has grown to more than 25 researchers and engineers, according to the company.
The company plans to use the funding to expand its compute infrastructure and research operations.
It is also preparing to run its first “Level 1” autonomous training system, according to the company. A public launch is planned for mid-2026.
The deal adds to a competitive field that includes startups working on world models, reinforcement learning, and safety-focused superintelligence research.
Recursive Superintelligence is seeking to differentiate itself by attempting to automate more of the AI development pipeline, rather than focusing only on scaling existing model architectures.