Recursive Superintelligence has entered into a substantial $410 million compute agreement with Amazon Web Services (AWS), announced on July 28, 2026. The AI startup, which went public from stealth mode in May with $650 million in funding, focuses on open-ended, self-improving AI systems—a resource-intensive pursuit requiring significant computing power. This multi-year collaboration provides the scalability and flexibility necessary for Recursive to advance its automated product development efforts.

Founder and CEO Richard Socher highlighted that this deal represents the majority of Recursive’s total raised capital so far, but anticipates many more such agreements to come. Unlike typical AI companies that emphasize headcount, Recursive is channeling much of its budget toward increasing computational agents rather than expanding staff, aiming to automate its innovation cycles efficiently. AWS’s role in the partnership is strictly as a service provider with no investment stake, but the scale of support reflects a commitment to develop custom infrastructure that fits recursive AI companies' specific needs.

Self-improvement in AI, often referred to as recursive self-improvement (RSI), is considered a key turning point in the field. While the exact nature and timeline remain debated, Recursive is pioneering this frontier by applying RSI toward the creation of tangible products. Socher expressed confidence in delivering functional and useful AI-powered tools to users within a few months, projecting the first releases as early as October 2026.

AWS’s involvement also signals potential support for other foundational AI firms working in this developing space, as the partnership includes co-development of infrastructure tailored to these innovative ventures. This move highlights how cloud providers are strategically positioning themselves to enable the next wave of AI breakthroughs driven by self-upgrading systems, positioning Recursive Superintelligence as an important player on this cutting edge of technology advancement.