Google and NVIDIA Bet on This $4-Billion AI Company That Aims to Replace Scientists
TechFlow Selected TechFlow Selected
Google and NVIDIA Bet on This $4-Billion AI Company That Aims to Replace Scientists
The funding myth surrounding self-learning AI tells us one thing: this AI arms race is even “overwhelming” researchers themselves.
Navigating Web3 tides with focused insightsAuthor｜Hualin Wuwang
In 1956, a group of scientists gathered at Dartmouth College to formally debate the question “Can machines think?” for the first time. They optimistically believed the problem could be solved in a single summer.
Seventy years later, the question remains unanswered. Yet one company—just four months old—has already secured $500 million in funding and achieved a $4 billion valuation, solely on the claim that it has discovered a path toward AI that can conduct research autonomously and evolve itself.
That company is Recursive Superintelligence.
Google Ventures (GV) led the round, with NVIDIA participating as a co-investor. Their respective positions within the AI ecosystem need no elaboration. That both firms jointly backed a startup with no publicly released product warrants careful analysis.
01 “Removing Humans from the Loop”
First, what exactly is Recursive Superintelligence building?
Founded by Richard Socher, former Chief Scientist at Salesforce, the company’s core team hails from Google DeepMind and OpenAI. This is not an unfamiliar combination—over the past two years, engineers and researchers departing top-tier labs to launch startups have formed a clear wave.
Richard Socher’s X profile; Sam Altman clearly follows this talent. | Source: X
Socher is not the typical Silicon Valley founder who joins a major tech firm merely to “polish his resume.” Born in Germany in 1983, he studied under AI pioneers Andrew Ng and NLP authority Christopher Manning at Stanford University. He completed his Ph.D. thesis in 2014 and received Stanford’s Computer Science Department’s Best Dissertation Award that year.
Richard Socher is one of the key figures who brought neural network methods into mainstream natural language processing (NLP). His early work on word embeddings, contextual embeddings, and prompt engineering directly laid the technical foundations for today’s BERT and GPT-series models. His Google Scholar citation count exceeds 180,000.
In the same year he earned his doctorate, Socher founded the AI startup MetaMind, which Salesforce acquired via strategic acquisition just two years later. Thereafter, he served as Chief Scientist and Executive Vice President, leading Salesforce’s AI strategy for several years and overseeing the rollout of enterprise AI products such as Einstein GPT.
After leaving Salesforce, he launched the AI search engine You.com in 2020 and closed its Series C round in 2025 at a $1.5 billion valuation. This time, he has shifted his focus from search to a more foundational challenge.
Thinking Machines Lab, Safe Superintelligence, Ineffable Intelligence, Advanced Machine Intelligence Labs—each brandishes the label “ex-XX large-model core team,” and each tells a story about “the next generation of AI.”
Yet Recursive’s entry point is far more radical than most peers’.
Its core proposition is “self-learning AI”—not making AI better at answering questions, but enabling AI to autonomously execute the full scientific research workflow: formulating hypotheses, designing experiments, evaluating results, and iterating directions. In other words, it aims to fully remove human researchers from this loop.
This direction is not new—but Recursive embeds it within an extremely pragmatic commercial logic. Today, top-tier AI researchers command annual salaries of $15–20 million. If a system can perform equivalent work at lower cost and higher speed, the economic model underpinning frontier research would be fundamentally rewritten.
Investors clearly see this logic. Reports indicate the funding round was oversubscribed, with the final size potentially reaching $1 billion.
02 Google and NVIDIA Bet Simultaneously
GV led the round; NVIDIA participated. This investor lineup itself sends a signal.
Google’s rationale is straightforward. For years, DeepMind has been the foremost explorer of “AI for Science”: AlphaFold cracked the protein folding problem; AlphaGeometry defeated top human competitors in mathematical olympiads.
But DeepMind’s approach uses AI to solve specific scientific problems. Recursive aims to do something more fundamental: enable AI systems to autonomously drive the scientific discovery process itself. To Google, this represents both competition and a compelling hedge.
More importantly, earlier this month Google announced a multi-generational AI infrastructure partnership with Intel. This signals Google’s accelerated push across AI infrastructure layers. Its investment in Recursive is a piece in that broader strategic chessboard—whoever builds the fastest models, Google wants a stake.
NVIDIA’s rationale is even more direct. The core bottleneck for self-learning AI isn’t algorithms—it’s compute. If AI must autonomously run experiments and iterate models, the required GPU cluster scale grows exponentially. By investing in Recursive, NVIDIA is, in effect, investing in its own future orders.
The fact that both companies invested simultaneously also conveys a subtler message—that this sector may have reached the “if you don’t invest now, you’ll miss out” stage.
03 Is a $4 Billion Valuation Reasonable After Just Four Months?
When most people first saw the $4 billion figure, their immediate reaction was likely, “Here we go again.”
AI startup valuation bubbles are nothing new over the past two years. A PDF, a demo, a few slides—and names from top-tier labs—have repeatedly unlocked hundreds of millions of dollars in funding. This is no longer legend in Silicon Valley or London; it’s routine.
Yet upon closer inspection, Recursive differs from typical “PowerPoint unicorns” in several ways.
First, the weight of its founding team. Richard Socher possesses genuine academic credentials in NLP—not just the “ex-big-tech” halo. His core team’s experience at DeepMind and OpenAI means they’ve directly grappled with frontline research pain points.
Second, the fact that the round was oversubscribed. This implies demand vastly exceeds supply—investors are scrambling to get in, not being persuaded to join.
Still, a $4 billion valuation for a four-month-old company with no public product rests entirely on expectations—not reality. It’s essentially paying for a direction, not a product or revenue stream.
This pricing logic is becoming increasingly common in the AI era, reflecting investors’ deep-seated fear of “missing the next OpenAI.” Safe Superintelligence secured its sky-high valuation years ago with virtually no product—the mere name Ilya Sutskever was its hardest asset.
Recursive is replicating that playbook. This is not criticism—it’s an objective observation.
04 Behind the Door Labeled “Self-Learning”
The name “Recursive Superintelligence” already reveals the company’s ambition quite clearly.
“Recursive” refers to recursion—a function calling itself—and serves as a core mechanism in many complex algorithms. Applied to AI research, “recursive superintelligence” implies a system capable of continuously optimizing itself and ascending in a spiral-like fashion.
This concept is not new; its extreme version is the “intelligence explosion”—a scenario where, once a system surpasses a critical threshold, it autonomously accelerates its own evolution, ultimately attaining intelligence levels incomprehensible to humans. This remains one of the central concerns in AI safety.
Yet Recursive is almost certainly nowhere near that level yet. A more realistic interpretation is that it’s attempting to build a system capable of autonomously driving the scientific exploration loop—aiming to drastically reduce the human labor and time costs of AI research.
If it succeeds, the impact won’t be confined to the AI community. Drug discovery, materials science, physics, and other fields could enter an era where rapid progress occurs “without human scientists involved.”
Of course, that remains an “if.”
In the AI industry, the distance between claiming something and actually delivering it is never linear.
05 The Logic of the Wave
Since the second half of 2025, wave after wave of researchers have departed top-tier labs to launch startups: Thinking Machines Lab, Safe Superintelligence, Ineffable Intelligence—the list keeps growing.
Recursive is the newest—and currently highest-valued—entrant in this wave.
The underlying structural reason is simple: competition among OpenAI, Anthropic, and Google DeepMind has transformed these elite labs into increasingly corporate entities—with KPIs, compliance requirements, and internal politics.
Researchers truly committed to the most radical frontiers often feel freer to strike out on their own.
Meanwhile, capital market logic reinforces this trend. For top-tier researchers backed by major tech firms, the current window for entrepreneurship may be historically optimal—investors are more willing than ever to pay for “direction.”
The central question of this wave isn’t “Who will succeed?” but rather “What does success mean?”
If Recursive proves the feasibility of self-learning AI, it will rewrite the foundational paradigm of AI research. If it fails, after burning through its $500 million war chest, all that may remain is another overhyped concept.
Both outcomes are genuinely possible.
Four months. A $4 billion valuation. That number excites—and alarms. As the AI arms race evolves, even “how research itself is conducted” has become a battlefield.
The question debated by scientists at Dartmouth for a summer is now being addressed—not by humans, but by AI: using AI to research AI, pursuing superintelligence via recursion.
No one truly knows where this road leads. But clearly, Google and NVIDIA have decided: regardless of destination, they cannot afford to be absent.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
极客公园
