Recursive Superintelligence inks $410M Amazon compute deal to scale self-improving AI

Recursive Superintelligence, an AI research company that launched publicly just months ago, has locked in a $410 million compute contract with Amazon Web Services. The multi-year arrangement will help the startup develop systems capable of improving themselves without ongoing human intervention.

The $650 million-funded firm emerged from stealth in May with a focus on recursive self-improvement (RSI), a compute-heavy approach that prioritizes automation over traditional headcount. Rather than staffing up like conventional tech companies, Recursive is shifting resources toward powering autonomous AI agents that can iterate on their own capabilities.

Scale and future deals#

CEO Richard Socher characterized this agreement as merely the opening move in a broader infrastructure strategy. "Today's announcement is likely going to be one of the smallest compute deals we're going to sign in the next few years," he told TechCrunch, signaling the company's appetite for even larger allocations as it scales development.

The $410 million commitment represents the bulk of Recursive's capital to date, yet leadership views it as foundational rather than limiting.

AWS partnership approach#

Unlike typical venture arrangements between cloud providers and AI labs—which often blend investment with infrastructure commitments—this deal is purely a compute and services contract. Amazon gains no equity stake but gains a high-profile customer whose specialized needs could inform future infrastructure improvements.

Jason Bennett, AWS VP for startups and venture capital, framed the partnership as part of a broader strategy: "We're going to co-develop infrastructure purpose-built for these types of company," suggesting AWS is positioning itself to support a wave of companies pursuing similar self-improvement pathways.

Product timeline#

Recursive plans to release early products leveraging RSI technology before year-end. Socher promised tangible tools within months rather than quarters, targeting an October launch for usable prototypes that customers can experiment with directly.