In 1956, a group of scientists gathered at Dartmouth to formally discuss for the first time whether machines could think. They optimistically believed they could solve the problem in just one summer.
Seventy years later, the question still has no answer. But one company, just four months old, has raised $500 million in funding and reached a $4 billion valuation—simply because it claims to have found a way to make AI learn to conduct research and evolve on its own.
The company is called Recursive Superintelligence.
Led by Google Ventures (GV), with NVIDIA participating as a co-investor. The positions of both companies within the AI ecosystem require no elaboration. Their simultaneous investment in a startup that hasn’t even publicly launched a product warrants careful analysis of the underlying rationale.
01 "Remove people from the cycle"
First, let’s explain what Recursive Superintelligence is actually doing.
The company was founded by Richard Socher, former Chief Scientist at Salesforce, with a core team drawn from Google DeepMind and OpenAI. This is not an unusual combination—in the past two years, engineers and researchers leaving top laboratories to start their own ventures have formed a clear trend.
Socher is not the typical founder who comes from a big tech company to gain prestige. Born in Germany in 1983, he studied under AI pioneer Andrew Ng and NLP expert Christopher Manning at Stanford University, completing his doctoral dissertation in 2014 and earning the top doctoral thesis award in Stanford’s Computer Science department that year.
Richard Socher is one of the key figures who truly brought neural network methods into natural language processing—his early research on word vectors, context vectors, and prompt engineering directly laid the technical foundation for today’s BERT and GPT series models, with over 180,000 citations on Google Scholar.
In the year he graduated with his Ph.D., he founded the AI startup MetaMind, which was acquired by Salesforce through a strategic acquisition two years later. He then led Salesforce’s AI strategy for several years as Chief Scientist and Executive Vice President, overseeing the deployment of enterprise AI products such as Einstein GPT.
After leaving Salesforce, he founded the AI search engine You.com in 2020, which completed its Series C funding in 2025 with a valuation of $1.5 billion. This time, he shifted his focus from search to more fundamental questions.
Thinking Machines Lab, Safe Superintelligence, Ineffable Intelligence, Advanced Machine Intelligence Labs... each emerges with the label of "the former core team behind the top X large models," each telling a story of "the next generation of AI."
But Recursive's approach is more aggressive than most of its peers.
Its core proposition is "self-learning AI"—not making AI better at answering questions, but enabling AI to autonomously complete the entire scientific research process: formulating hypotheses, designing experiments, evaluating results, and iterating directions. In other words, it aims to fully remove human researchers from this cycle.
This isn't a new direction, but Recursive has placed it within an extremely practical business logic. With top AI researchers now earning salaries of $15 to $20 million annually, if a system can perform the same work at lower cost and greater speed, the economic model of cutting-edge research will be completely rewritten.
Investors clearly understood this logic. The funding round was reportedly oversubscribed, with a final size potentially reaching $1 billion.
02 Google and NVIDIA both invest
GV led the investment, with NVIDIA participating as a co-investor. This investor lineup itself is a signal.
Google's logic is easy to understand. For years, DeepMind has been the leading pioneer in the "AI for Science" domain, with AlphaFold solving the protein folding problem and AlphaGeometry outperforming top human competitors in mathematical competitions.
But DeepMind’s approach is to use AI to solve specific scientific problems, while Recursive aims to do something more fundamental—to enable AI systems to autonomously drive the process of scientific discovery itself. This represents both a competitive threat and a worthwhile hedge for Google.
More importantly, just at the beginning of this month, Google announced a multi-generational AI infrastructure partnership with Intel, signaling that Google’s efforts in AI infrastructure are accelerating across the board. Its investment in Recursive is one piece of this larger strategy—Google wants a stake in whichever model runs ahead.
NVIDIA’s logic is more straightforward. The core bottleneck for self-learning AI is not algorithms, but computing power. If AI is to autonomously run experiments and iterate models, the scale of GPU clusters required will grow exponentially. By investing in Recursive, NVIDIA is essentially betting on its own future orders.
Both companies acting simultaneously also sends a more subtle signal—that this sector may have reached the point where “if you don’t invest now, it’s already too late.”
Is a $4 billion valuation after four months reasonable?
When everyone first sees the $4 billion figure, their initial reaction is likely, “Here we go again.”
The AI startup valuation bubble is no longer a new topic these past two years. A PDF, a demo, a few slides, and a handful of names from top labs can unlock hundreds of millions of dollars—this is no longer legend in Silicon Valley and London, but everyday reality.
However, looking closely at Recursive's situation, there are a few differences from the typical "PPT unicorn."
First, the weight of the founding team. Richard Socher has genuine academic credentials in NLP, not just a veneer of prestige from a former big tech company. The core team’s experience at DeepMind and OpenAI means they have firsthand exposure to the real challenges in cutting-edge research.
Second, the fact that the financing was oversubscribed. This means market demand far exceeded supply, with investors eager to participate rather than being persuaded to do so.
But a $4 billion valuation for a company that is only four months old and has no publicly available product is based on expectations, not reality. It is essentially paying for a direction, not for a product or revenue.
This pricing logic is becoming increasingly common in the AI era, driven by investors' deep-seated fear of missing the next OpenAI. Safe Superintelligence also secured a massive valuation with almost no product— Ilya Sutskever’s name was its most valuable asset.
Recursive is copying the same path. This is not a criticism, but an objective observation.
04 What lies behind the door of "self-learning"?
The name "Recursive Superintelligence" clearly reveals the company's ambition.
"Recursive" means recursion. In computer science, recursion is a structure in which a function calls itself, serving as a core mechanism for many complex algorithms. In the context of AI research, "recursive superintelligence" suggests a system capable of continuously optimizing itself in a self-reinforcing, spiraling process.
This concept is not new; its extreme version is "intelligence explosion"—a system that, once it surpasses a certain threshold, can autonomously accelerate its own evolution, ultimately reaching an intelligence level beyond human comprehension. This has long been one of the core concerns in AI safety.
But what Recursive is currently doing is likely nowhere near that level. A more realistic interpretation is that it is attempting to build a system capable of autonomously driving a scientific exploration loop, with the goal of significantly reducing the human and time costs of AI research.
If it truly delivers, its impact won't be limited to the AI community—it could usher in a new era in drug discovery, materials science, physics, and other fields, where rapid progress happens even without direct human scientific involvement.
Of course, this is still "if."
The distance between claim and realization in the AI industry is never linear.
05 The Logic of the Wave
Since the second half of 2025, a wave of entrepreneurs has emerged from top laboratories, one after another. Thinking Machines Lab, Safe Superintelligence, Ineffable Intelligence… this list continues to grow.
Recursive is the newest and currently the most highly valued company in this wave.
The underlying structural reason is simple—competition among OpenAI, Anthropic, and Google DeepMind has made these leading labs increasingly resemble large corporations, with KPIs, compliance requirements, and politics.
Researchers who truly want to bet on the most aggressive directions often find it more liberating to strike out on their own.
Meanwhile, the logic of the capital market is reinforcing this trend. For top researchers backed by major tech companies, the current window for entrepreneurship may be the best in history—investors are more willing than ever to pay for “direction.”
The core issue of this wave is not "who will succeed," but "what does success mean?"
If Recursive ultimately proves the feasibility of self-learning AI, it will rewrite the foundational paradigm of AI research. If it fails, the $500 million in funding will be spent, leaving behind just another overhyped concept.
Both possibilities may be real.
Four months, a $4 billion valuation—this number is both exciting and concerning. Today, the AI arms race has evolved to the point where even "how to conduct research" has become a battleground.
Scientists debated an issue all summer at Dartmouth, and now someone plans to use AI to answer it—using AI to study AI, recursively heading toward superintelligence.
No one truly knows where this path leads, but clearly, Google and NVIDIA have decided that they cannot afford to miss it.