Recursive Superintelligence signs $400M compute deal with AWS

The AI startup committed $400 million to Amazon Web Services for cloud computing capacity to develop self-improving systems.

Recursive Superintelligence, an AI startup pursuing self-improving systems, signed a $400 million multi-year compute agreement with Amazon Web Services, marking one of the largest cloud infrastructure commitments by a company still in stealth.

The company prioritizes "agent count" over human headcount, directing capital toward machine-based development rather than workforce expansion, according to company leadership. The approach reflects a bet that automated research cycles can accelerate progress faster than traditional hiring.

The deal, announced Tuesday, does not include an equity investment from Amazon. Instead, AWS will use insights from supporting Recursive Superintelligence's infrastructure needs to refine its offerings for other foundational AI companies. The startup emerged from stealth in May with $650 million in external funding, one of the largest initial raises for an AI company.

For AWS, the agreement locks in a $400 million revenue stream from a high-growth AI client and strengthens its position in the AI cloud sector, where hyperscalers are competing for specialized firms requiring massive computational environments. Recursive Superintelligence gains the computing resources needed to pursue recursive self-improvement, a research approach that remains unproven at commercial scale.

Competitive Landscape

The deal comes as AI startups race to secure compute capacity amid a supply crunch for Nvidia's H100 and B200 graphics processing units. AWS, Microsoft Azure and Google Cloud have each struck multi-billion-dollar partnerships with AI companies, with Microsoft's $13 billion investment in OpenAI and Google's ties to Anthropic setting the benchmark for strategic cloud-AI alliances. Recursive Superintelligence's $400 million commitment, while smaller in absolute terms, is notable for a company that has yet to release a commercial product.

The Self-Improvement Thesis

Recursive Superintelligence's core premise — that AI systems can enhance their own architecture without direct human intervention — remains in the research phase. The approach is compute-intensive by design, requiring vast clusters of GPUs to run iterative improvement loops. Ilya Sutskever's Safe Superintelligence, another startup pursuing similar goals, recently partnered with Nvidia to scale its research, showing the capital intensity required in this segment.

Investor Implications

Amazon shares trade at roughly 22 times forward earnings, with AWS contributing about 16 percent of total revenue but a disproportionate share of operating profit. The deal adds to a growing pipeline of AI-related cloud commitments that analysts at Morgan Stanley estimate could contribute $2 billion to $3 billion in incremental AWS revenue over the next two years. For Recursive Superintelligence, the challenge will be managing its $650 million cash pile while spending heavily on compute, with no timeline disclosed for a revenue-generating product.

This article is for informational purposes only and does not constitute investment advice.