Recursive Superintelligence Raises $650M to Build Self‑Improving AI

*TL;DR: Recursive Superintelligence closed a $650 million round at a $4.65 billion valuation and plans a public debut in mid‑2026 with self‑improving AI systems.

Context

Founded in 2025, Recursive Superintelligence brings together former leaders from OpenAI, Google DeepMind, Meta AI, Salesforce AI and Uber AI. The stealth‑mode lab operates out of San Francisco and London, where it already employs more than 25 researchers and engineers. Its core mission is to automate the AI research pipeline, allowing software to redesign its own architecture, training methods and evaluation metrics without human intervention.

Key Facts

- The company raised $650 million in a financing round led by GV (Google Ventures) and Greycroft, with AMD Ventures and NVIDIA also participating. The round values the startup at $4.65 billion. - Founders include notable figures such as Richard Socher, Tim Rocktäschel, Jeff Clune, Josh Tobin and Tim Shi, each previously heading AI teams at major tech firms. - Recursive Superintelligence plans to use the capital to acquire large‑scale compute resources and launch its first “Level 1” autonomous training system. - A public launch is scheduled for mid‑2026, coinciding with an expansion of research operations in both San Francisco and London. - The startup’s approach differs from competitors that focus on scaling ever larger foundation models; instead it seeks to create a recursive loop where AI systems continuously generate and refine new capabilities, akin to biological evolution.

What It Means

The infusion of $650 million signals strong investor confidence in a paradigm shift from model scaling to self‑directed AI research. By automating the discovery of better learning strategies, Recursive Superintelligence could accelerate progress across multiple AI domains and potentially reduce the time and cost of developing advanced models. The involvement of hardware players AMD and NVIDIA suggests the venture will secure cutting‑edge GPUs and specialized chips, essential for the massive compute demands of autonomous training cycles.

If the company meets its mid‑2026 launch target, it will join a growing cohort of AI firms pursuing open‑ended intelligence, challenging incumbents that rely on incremental model improvements. Success could reshape how AI research is funded and conducted, pushing the industry toward a model where machines drive their own evolution.

What to watch next

Monitor the rollout of Recursive Superintelligence’s Level 1 system and any partnerships that emerge as the firm scales its compute infrastructure across the UK and US.