Recursive Superintelligence signs $410M AWS compute deal for self-improving AI
The RSI startup plans to release its first products by October, using massive compute for self-improvement.
Recursive Superintelligence (RSI), an AI lab focused on building open-ended self-improving systems, has signed a $410 million compute deal with Amazon Web Services. The agreement, announced Tuesday, is a multiyear deal designed to provide flexibility as the company scales its systems. CEO Richard Socher, a former Salesforce chief scientist, stated this is likely the smallest compute deal the company will sign in the coming years. Since emerging from stealth in May with $650 million in total funding, Recursive has emphasized that most of its budget goes toward compute rather than headcount, aiming to automate its own product development. 'For us, it's less about headcount and more about agent count,' Socher said.
Unlike major AI labs that often include an investment component in such deals, Amazon's involvement is purely a compute commitment. However, the scale allows AWS to co-develop purpose-built infrastructure for Recursive's unique needs, potentially attracting other foundation-level AI companies, according to Jason Bennett, AWS VP for startups and venture capital. Recursive is pursuing recursive self-improvement (RSI), a long-hyped inflection point where AI can improve itself without human input. While some predict imminent breakthroughs, others see it as a continuum. Socher expects to release tangible products 'within a few months, not quarters or years,' with the first examples arriving as early as October.
- Recursive Superintelligence signed a $410 million compute deal with AWS, multiyear, with flexible scaling.
- Company emerged from stealth in May with $650 million total funding; CEO calls this 'the smallest compute deal we'll sign.'
- No investment component; AWS will co-develop infrastructure for foundational AI companies; products expected by October.
Why It Matters
Recursive's focus on self-improving AI could accelerate autonomous product development, reducing reliance on human-led iteration.