Recursive Superintelligence, a San Francisco-based artificial intelligence (AI) startup, exited stealth mode this month with a $650 million funding round, positioning itself to pursue what its founders describe as one of the most elusive goals in contemporary AI research: a system capable of improving itself without human intervention.
Funding and investors
<cite index="2-1">The startup emerged from stealth on Wednesday with $650 million in funding from investors including Greycroft and GV.</cite> <cite index="8-1">The round valued the company at $4.65 billion</cite>, and <cite index="8-14">was described as heavily oversubscribed.</cite> <cite index="8-12,8-13">The funding will be used in part to secure the large-scale compute infrastructure required to run these experiments. The company currently operates from offices in San Francisco and London, with a team that has expanded beyond 25 researchers and engineers.</cite>
Founding team
The company is led by Richard Socher, the former chief scientist at Salesforce and founder of AI search engine You.com. <cite index="8-9">He leads the company alongside seven co-founders: Yuandong Tian, formerly a research scientist director at Meta's Fundamental AI Research lab (FAIR), where he led work on reinforcement learning, Large Language Model (LLM) reasoning, and AI-guided optimisation; Tim Rocktaschel, a professor of AI at University College London and former principal scientist at Google DeepMind; Alexey Dosovitskiy, one of the authors of the Vision Transformer (ViT)</cite>, alongside others including Peter Norvig and Cresta co-founder Tim Shi.
Technical approach
<cite index="3-12,3-13">Recursive Superintelligence is focused on creating what researchers call a recursively self-improving AI model — a system that can autonomously identify its own weaknesses, design fixes, and implement them without human intervention. This concept, often described as a holy grail in contemporary AI research, would represent a fundamental shift in how AI systems evolve.</cite>
The company's central technical bet is on "open-endedness." <cite index="2-6">The startup focuses on open-endedness, a concept inspired by biological evolution where AI systems co-evolve through iterative competition, rather than relying on human-designed benchmarks or supervised fine-tuning.</cite> <cite index="6-11,6-12">One practical example is "rainbow teaming," a technique where two AI systems co-evolve — one attempts to make the other produce harmful outputs, and the other learns to resist those attempts. This iterative process, Socher said, is now used in all major labs.</cite>
Product roadmap
<cite index="8-10,8-11">The first step, according to company materials, is to train a system with the capabilities of "50,000 doctors" to automate AI scientific research itself. From there, the company plans to run what it calls a "Level 1" autonomous training system, with a public launch targeted for mid-2026.</cite> Socher has said products are expected within "quarters, not years," pushing back on the "neolab" classification often applied to research-first AI startups.
Competitive context
<cite index="8-18,8-19,8-20">What distinguishes Recursive Superintelligence from these efforts is that none of the major laboratories has organised an entire company around recursive self-improvement as its core commercial thesis. OpenAI, Anthropic, and Google DeepMind all use AI to assist their research workflows, but their businesses are built around selling models and Application Programming Interface (API) access. Recursive is betting that the self-improvement loop itself is the product.</cite>
The viability of the thesis remains debated. <cite index="8-22">Anthropic co-founder Jack Clark has estimated a roughly 60% probability that a system capable of training a more powerful successor on its own, without human involvement, will exist by the end of 2028, and a 30% chance by 2027.</cite> Skeptics counter that recursive improvement loops may converge on diminishing returns rather than runaway acceleration.