Recursive Superintelligence Signs $410M AWS Compute Deal

A 30-Person Company With No Product Just Bought $410 Million of AWS

Recursive Superintelligence is spending most of what it raised in May on compute for a system meant to improve itself. Its co-founder calls it one of the smallest deals the company will sign.

Jul 29, 2026Source: AWS

AWS announced on 28 July that Recursive Superintelligence had signed a $410 million multi-year agreement for compute to run its automated AI-research system. The company emerged from stealth on 13 May with $650 million at a $4.65 billion valuation, led by GV and Greycroft with AMD Ventures and Nvidia participating. Most of that money now has a destination.

What they say they are building

Co-founder and chief executive Richard Socher — formerly chief scientist at Salesforce and co-founder of You.com — describes the approach as open-endedness in service of self-improvement: "Our unique approach is to use open-endedness to get to recursive self-improvement, which no one has yet achieved." Tim Rocktäschel, previously at Google DeepMind, is a co-founder; the founding team is drawn from OpenAI, DeepMind, Meta AI, Salesforce AI and Uber AI. AWS's Jason Bennett supplied the demand-side line: "Self-improving AI creates a compounding demand for compute."

The arithmetic

Twenty-five to thirty people, two offices, no shipped product, and $410 million committed to inference and training. Socher's framing of the ratio is the quotable part: "For us, it's less about headcount and more about agent count." He also called this "likely going to be one of the smallest compute deals we're going to sign in the next few years."

What is not disclosed

Neither party said which silicon. AWS's language is "purpose-built compute" with "the elasticity to run those loops in parallel at massive scale," which covers both its own Trainium and its Nvidia fleet. No contract term in years was given, and no investment component is bundled into the deal.

October

Socher told reporters initial products are targeted for October 2026. That is roughly five months from raise to shipping for a company whose stated objective is a system that improves itself — a deadline that will be the first external evidence of whether any of this works.

Why It Matters

The old signal for an AI startup's ambition was how many researchers it hired. This one spent 63% of its funding on compute for thirty people, which says the bet is that capability now comes from cycles rather than headcount — and October is when it gets checked.