AI lab Mirendil has signed a multi-year partnership with Google Cloud valued at over $100 million to secure compute capacity for its self-improving AI research.

The deal, confirmed exclusively by TechCrunch, mirrors two trends in the industry: cloud giants making large infrastructure commitments to startups, and AI companies locking down compute as they scale.

The funding is roughly half of what Mirendil raised in a seed round at a $1 billion valuation in late June.

Under the agreement, Mirendil gains access to Google’s TPUs and Nvidia GPUs, along with managed training clusters, which it will use to pursue recursive self-improvement — an AI approach where systems iteratively improve themselves.

Mirendil’s co-founders previously worked at Anthropic, which is also researching this concept.

The startup believes its technology could eventually automate much of scientific and AI research, helping advance fields like medicine, biology, and materials science.

Co-founder Harsh Mehta emphasized the importance of matching workloads to the right hardware, noting that Google’s multiple chip types allow the lab to mix and match accelerators and lower costs for itself and its customers.

Google’s Amin Vahdat stated that AI progress now depends on orchestrating entire systems rather than just chip-level performance.

For Google, the deal provides a strategic partner building frontier recursive self-improving AI that can later be offered to enterprise customers.