Mirendil has locked down a compute deal worth more than $100 million with Google Cloud. The AI lab, founded by former Anthropic researchers, announced the multiyear partnership on August 6, securing access to Google TPUs, Nvidia GPUs, and managed training clusters for its self-improving AI work.
The commitment equals about half of the seed round Mirendil raised at a $1 billion valuation in late June. The startup is chasing recursive self-improvement, building systems that iteratively get better at their own tasks, and its long-term ambition is software capable of doing the work of an entire frontier AI lab.
CEO Behnam Neyshabur sees the technology accelerating scientific discovery in medicine, biology, and materials science. Co-founder Harsh Mehta argues the key is routing workloads to the right hardware, and that Google’s mix of accelerator types lets the lab cut costs for itself and its customers.
Google gets value on two levels. Mirendil’s systems layer is expected to squeeze more out of Google’s hardware, and the cloud giant gains a strategic partner building frontier recursive self-improving AI that it can eventually offer to enterprise customers.
The startup joins a growing field, including Recursive Superintelligence, which signed a $400M compute agreement with Amazon in July.
