Recursive Superintelligence emerged with a $650 million round and a $4.65 billion valuation to pursue recursive self-improvement rather than a conventional product roadmap.
What changed
Richard Socher and Tim Rocktäschel lead a research-heavy team split between London and San Francisco.
The company’s premise is that AI can propose hypotheses, design experiments, write code and change its own training loop.
Why it matters
No public benchmark or product has been shown, so investors are underwriting people, compute access and a very open research problem.
What to watch
The useful signal is not the headline number alone, but whether the product, contract or platform shift becomes repeatable over the next two quarters.
Sources: CocoLoop