AWS to provide $400 million in compute for Recursive Superintelligence

Recursive Superintelligence announced on Tuesday a multi-year, $400 million compute agreement with Amazon Web Services. The AI company emerged from stealth in May with $650 million in funding and plans to use the capacity for open-ended, self-improving systems.

Why compute is central to the strategy

The commitment represents most of Recursive’s funding to date. Unlike AI laboratories that direct substantial budgets towards staffing and operations, the company intends to place more capital directly into computing resources and automate parts of product development.

Its focus is recursive self-improvement, or RSI: systems that can contribute to their own advancement with less human involvement. Researchers disagree on whether RSI will produce a sudden breakthrough or develop gradually, but either path could demand significant and flexible infrastructure.

Terms of the AWS agreement

The deal does not include an investment by Amazon, distinguishing it from the hybrid funding and cloud arrangements used by several major AI laboratories. Socher nevertheless expects Recursive to sign larger compute contracts as its systems scale.

“Likely going to be one of the smallest compute deals we’re going to sign in the next few years.”

AWS will commit engineering resources to Recursive’s requirements. Jason Bennet, vice president for startups and venture capital at AWS, said the parties would jointly develop infrastructure designed for companies pursuing this model of AI research.

“Part of the agreement is that we’re going to co-develop infrastructure purpose-built for these types of company.”

Products expected within months

Recursive intends to turn RSI research into products rather than treat self-improvement solely as a theoretical milestone. Socher said the first tangible and usable examples should appear around October, before the end of the year.

For businesses, the agreement shows that advanced AI development increasingly depends on long-term access to specialised compute, not only model expertise. Companies planning agent-based products should evaluate capacity, cloud architecture and infrastructure partnerships early, before compute becomes a constraint on delivery.