Mirendil inks $100M+ Google Cloud deal for self-improving AI

AI lab Mirendil on August 6, 2026 disclosed a multi-year partnership with Google Cloud worth more than $100 million to secure compute capacity for its self-improving AI research.

The sum represents roughly half of the $200 million in seed funding the startup raised at a $1 billion valuation in late June 2026, co-founder Benham Neyshabur said.

Under the partnership, Mirendil gains access to Google’s TPUs, Nvidia GPUs, and managed training clusters for building AI that iteratively improves itself.

Self-improving AI, also called recursive self-improvement, refers to systems that keep enhancing their own performance.

Neyshabur said such systems can be pointed at a problem and continuously improve, eventually automating research across medicine, biology, and materials science.

Co-founder Harsh Mehta stressed that training self-improving AI demands enormous compute, and Google’s multiple chip options let Mirendil assign the right workloads to the right accelerators, cutting costs for itself and its customers.

Google SVP Amin Vahdat said AI advancement now hinges on orchestrating entire systems of intelligence and overcoming physical scaling constraints.

The partnership positions Google to offer Mirendil’s software to enterprise customers, while Mirendil benefits from Google’s hardware ecosystem.

Mirendil’s co-founders previously worked at Anthropic, which has explored recursive self-improvement, while startups like Recursive Superintelligence and Ricursive Intelligence have recently emerged with similar ambitions.

The deal reflects a broader trend of cloud providers committing large infrastructure resources to AI startups racing to lock in compute capacity as they scale.