- Recursive Superintelligence raised $500 million at a $4 billion valuation—four months after founding, with no publicly available product.
- The startup wants to build an AI that improves itself without human help, a concept many researchers see as the most likely path to superintelligence.
- GV led the round with Nvidia joining; the deal was so oversubscribed that Recursive could pull in as much as $1 billion.
Half a billion dollars for a company that hasn’t officially launched. Four months of existence for a $4 billion valuation. No product demos, no benchmarks, no GitHub repo—just a team of 20 people and an idea that sounds like the premise of a cautionary AI paper.
Recursive Superintelligence, founded by former Salesforce chief scientist Richard Socher and Google DeepMind veteran Tim Rocktäschel, has raised at least $500 million in a round led by GV, with Nvidia participating, according to The Decoder, citing the Financial Times as the original source. The round was so oversubscribed that the company could end up pulling in as much as $1 billion.
The founding roster reads like a who’s-who of AI research: Socher ran Salesforce’s AI division as chief scientist before founding You.com, while Rocktäschel is a professor at University College London and was previously a principal scientist at DeepMind. The roughly 20-person team also includes former OpenAI researchers and alumni from Google and Meta.
What ‘Recursive Self-Improvement’ Actually Means
Recursive Superintelligence is chasing the holy grail of AI research: a system that gets smarter on its own. The concept of recursive self-improvement—an AI that identifies its own weaknesses, designs fixes, and implements them without human oversight—is what many researchers consider the most plausible path to artificial general intelligence. Each cycle of improvement makes the next cycle better. In theory, the feedback loop accelerates exponentially.
In practice, nobody has built one that works over extended periods. The Financial Times reported that the concept remains in the research phase and hasn’t been tested over long stretches of time. The team hasn’t published papers demonstrating the approach, hasn’t released benchmarks, and hasn’t disclosed what architectures they’re exploring.
The investor bet here isn’t on a proven technology—it’s on a team pedigree combined with a thesis that self-improving AI is inevitable. DeepSeek is raising $300 million for the first time after years of bootstrapping, and Britain’s sovereign AI fund just committed $675 million to domestic AI infrastructure. The capital sloshing through the system right now is staggering—even by 2026 standards.
The AI Funding Arms Race Rewards Elite Pedigrees
The pattern is unmistakable: the most eye-popping AI raises go to teams with the shiniest academic and corporate pedigrees, regardless of whether they have a working product. Sequoia recently led a $1 billion seed round for another ex-Google researcher’s lab. The AI model race has investors throwing capital at anyone who can plausibly claim a path to superintelligence.
Socher’s track record helps. You.com, his previous company, hit a $1.5 billion valuation after raising $100 million from investors including Salesforce Ventures. Rocktäschel’s academic work on reinforcement learning and open-endedness is widely cited. Together, they bring both research credibility and operational experience—a combination that’s currently commanding irrational premiums.
The oversubscription tells its own story. When a four-month-old company with no revenue can attract enough demand to potentially double its raise, it says less about the company and more about the market. Investor FOMO around superintelligence is driving valuations that would’ve seemed satirical two years ago. In April 2026, $4 billion pre-money for a 20-person team with no product barely registers as unusual.
The round closes a week after Claude Mythos demonstrated that today’s frontier AI already has capabilities that alarm security researchers. Whether Recursive’s self-improving approach turns out to be visionary or vaporware, the money has already spoken.
GV and Nvidia co-leading a half-billion-dollar bet on unproven recursive self-improvement research, with the round potentially doubling to $1 billion, is the kind of deal that defines the current AI moment.