Recursive's $650M Stealth Exit: The 14-Day Sourcing Window Before LinkedIn Catches Up
Recursive Superintelligence launched May 14, 2026 with $650M and 25 researchers. Here's how to source the second wave before LinkedIn titles update.
On Wednesday May 14, 2026, Richard Socher's Recursive Superintelligence came out of stealth with $650 million at a $4.65 billion valuation, eight named co-founders, and a team that has expanded beyond 25 researchers and engineers. If you are a recruiter or a founder competing for the same profile, you have roughly two weeks before those 25 people update their headlines and the rest of the cohort becomes opaque again. The named roster is small enough to enumerate by hand today. The second wave, which is where you actually have a chance, is sourceable only by upstream signals: paper authorship, prior employer, GitHub history.
This is the playbook for the next 14 days.
The roster, and why it's the wrong place to start
Recursive's announced co-founders are Richard Socher (CEO, ex-Salesforce chief scientist, You.com founder), Yuandong Tian (ex-Meta FAIR research scientist director), Tim Rocktäschel (UCL, ex-DeepMind principal scientist), Alexey Dosovitskiy (Vision Transformer co-author), Josh Tobin (ex-OpenAI), Caiming Xiong, Tim Shi (Cresta co-founder), and Jeff Clune. Peter Norvig sits on as an adviser, not a full co-founder. Socher's own launch post also names Chris Cummins and Jeffrey Pennington as early researchers on the team.
You are not going to poach any of these people. They just raised. Stop trying.
What you want is the second-degree graph: the people who will be hires #26 through #75, and the people who pitched and lost. Socher told reporters products are expected within "quarters, not years," and the company is targeting a public launch of a "Level 1" autonomous training system by mid-2026. That timeline implies aggressive engineering hiring inside 90 days. GV led the round with Greycroft, and both AMD Ventures and Nvidia participated, which is a tell: compute-secured rounds disproportionately hire infrastructure and distributed-training engineers first, not publishing researchers.
Why "neolab" framing will misroute your sourcing
Socher explicitly rejected the "neolab" label that gets pinned on research-first AI startups. He told reporters, "I want us to become a really viable company, to really have amazing products that people love to use." Read that as a hiring signal, not press positioning.
The Tim Shi axis is the tell. Shi scaled Cresta past 500 people; this is not his first time shipping. Josh Tobin was one of the first people at OpenAI and led their Codex and deep research teams. Pair that with a CEO who shipped You.com to a $1.5B valuation and you do not have an SSI-style monastery. You have a product company that happens to be staffed by ICML authors.
If you pitch candidates "go publish papers at the next big lab," you will misfire. The profile they need is applied ML, distributed training, RL infra, eval infrastructure, and product engineers who can read a paper and ship a feature. That is a very different sourcing query than "senior research scientist."
The named roster is bait. The hire profile is applied, not academic, and the second wave is where every reply-rate point matters.