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

AI lab Mirendil has signed a multiyear partnership with Google Cloud to source compute capacity for its self-improving AI research, TechCrunch has exclusively learned. The deal reflects two broader trends in the AI industry: cloud providers courting startups with large infrastructure commitments, and AI companies locking down as many compute deals as possible to secure capacity as they scale.

The deal is worth upward of $100 million, Mirendil co-founder and CEO Behnam Neyshabur told TechCrunch — roughly half of what the startup raised in its seed round at a $1 billion valuation in late June. Under the agreement, Mirendil gets access to both Google’s TPUs and Nvidia GPUs, along with managed training clusters it will use to develop self-improving AI. The startup hopes the technology will eventually be able to take on the work of an entire frontier AI lab.

Self-improving AI, also called recursive self-improvement, refers to AI systems that iteratively improve themselves — a concept major labs including Anthropic, where Mirendil’s co-founders previously worked, have also been pursuing. A handful of startups, including Recursive Superintelligence and Ricursive Intelligence, have recently formed around the same goal. Mirendil believes the approach could automate large portions of scientific and AI research, helping accelerate progress in fields such as medicine, biology, and materials science.

“You can have a self-improving AI where you can point a problem at it and it keeps getting better with time,” Neyshabur said, citing Alzheimer’s research as one example of a domain where such a system could keep improving its own knowledge and performance over time.

Training self-improving AI, however, demands enormous computing power. Co-founder Harsh Mehta said the work increasingly comes down to matching the right workloads to the right hardware, arguing that Google’s mix of chip types lets Mirendil combine workloads with the most suitable accelerators and lower costs both for itself and for its own customers. Amin Vahdat, Google’s SVP and chief technologist for AI and infrastructure, said in a statement that AI advancement is now less about chip-level performance and more about orchestrating “entire systems of intelligence.”

Neyshabur said Mirendil’s software and systems layer helps customers get more out of Google’s hardware — giving Google a potential edge over rival cloud providers — while Google, in turn, gains a strategic partner building frontier self-improving AI that it could eventually offer to enterprise customers.

Why it matters

The deal shows how compute-hungry AI labs and cloud providers are locking each other into multiyear commitments well before their technology is proven, tying cloud giants’ infrastructure bets directly to bets on unproven approaches like recursive self-improvement.