Recursive Superintelligence Raises $500M at $4B Valuation From GV + Nvidia, Four Months After Founding

Recursive Superintelligence, founded only four months ago, has closed at least $500M in funding at a $4B pre-money valuation. The round was led by GV (formerly Google Ventures) with Nvidia participating. Reporting indicates the round was sufficiently oversubscribed that the final total could extend toward $1B.

Team and founding

What Recursive Superintelligence is building

The company’s stated mission is autonomous self-improving AI: a system that writes, evaluates, and refines its own code and model architecture without human researchers in the loop at each step.

The approach sits in the same intellectual lineage as AutoML, meta-learning, and the theoretical work on recursive self-improvement that Socher and others have discussed for years, but aimed at production-grade foundation models rather than narrow-task ML systems.

Current status per the reporting:

• Research-phase only as of late April 2026
• Not yet tested over long autonomous self-improvement cycles
• Public launch planned roughly one month out from the April 9 announcement interview, placing it ~mid-May 2026

Why the funding round is notable

$500M at $4B valuation for a four-month-old company with no shipped product is one of the more aggressive AI seed/Series A configurations of 2026. Comparisons:

The pattern is consistent: founder reputation + thesis clarity is commanding AI-era seed valuations unprecedented in historical venture capital. GV and Nvidia backing Socher + Rocktäschel is the 2026 equivalent of Sequoia backing Brin and Page.

Where this fits in the April 2026 AI funding landscape

Per Crunchbase’s Q1 2026 data:

• $300B total global venture funding in Q1 2026 (up 150% QoQ and YoY)
• $242B of that ($80%) went to AI companies
• Four mega-rounds (OpenAI $122B, Anthropic $30B, xAI $20B, Waymo $16B) alone represented 65% of global VC

Recursive Superintelligence’s $500M is small relative to those four, but the valuation multiple relative to company age is more extreme than any of them.

What to watch

1. May 2026 public launch. What does a self-improving AI system look like when you can actually demo it? Real output quality, real failure modes, real cost math.

2. Nvidia + GV alignment. Both are strategic investors, not just financial. Nvidia benefits from anything that burns more GPU cycles; GV gets a parallel bet alongside Google’s own research programs.

3. Ex-DeepMind / ex-OpenAI researchers joining. The roster is one of the stronger single-company concentrations outside the existing frontier labs. Follow-on hiring signals will matter.

4. Regulatory response. Self-improving AI is the exact scenario discussed in most AI-safety policy papers. Expect EU AI Act and US executive-branch responses over the next 3-6 months if the demo is credible.

Editorial read

The valuation is extreme by any pre-2024 standard and aggressive even in 2026 terms. But the team is unusually strong (Socher + Rocktäschel together), the thesis is thesis-clear, and the strategic investor set (GV + Nvidia) is thoughtfully chosen.

For aipedia.wiki readers: this is a watch-list company, not a tool-to-pick-today company. If Recursive Superintelligence ships something meaningful in May 2026, it enters the tool catalog immediately. Until then, it’s a research bet that affects the competitive dynamics of AI tooling rather than a product choice.

Open questions

• What does self-improving mean in practice? Full model re-training? Architecture search? Prompt/tooling optimization?
• How do the founders frame alignment and safety concerns in their internal roadmap?
• Does the $500M (or $1B) runway include compute commitments, or is that separate?
• What is the go-to-market: API product, research lab-as-a-service, or something else?

Most of these resolve on or after the mid-May launch.