Mirendil, an artificial intelligence research laboratory, has entered into a substantial partnership with Google Cloud valued at more than $100 million. The multi-year agreement will provide the startup with essential computing resources to advance its work on self-improving AI systems. According to Mirendil’s co-founder and CEO Behnam Neyshabur, this investment represents approximately half of the company’s seed funding round completed in June at a $1 billion valuation.
The infrastructure deal grants Mirendil access to Google’s specialized processors, including TPUs and Nvidia GPUs, along with managed training clusters. These resources will support the company’s research into recursive self-improvement technology, which enables AI systems to iteratively enhance their own capabilities. Mirendil’s long-term vision involves creating AI that can autonomously conduct scientific research across fields including medicine, biology, and materials science.
The partnership reflects broader industry dynamics, as major cloud providers compete to secure relationships with emerging AI companies while startups race to lock in computing capacity. Co-founder Harsh Mehta emphasized the importance of matching different computational workloads to appropriate hardware types, a capability Google’s diverse chip offerings provide. For Google, the deal represents a strategic opportunity to develop frontier AI technology that could eventually reach enterprise customers.
Self-improving AI represents an emerging focus for multiple organizations, including established labs like Anthropic, where Mirendil’s co-founders previously worked. The technology aims to automate significant portions of scientific and research development by allowing AI systems to progressively deepen their knowledge and improve performance over time.
Originally reported by Techcrunch →
About the Author
