AI lab Mirendil has signed a multi-year partnership with Google Cloud to secure computing power for its self-improving AI research, TechCrunch has learned exclusively. The deal, valued at over $100 million, gives the startup access to Google's TPUs and Nvidia GPUs, plus managed training clusters. Mirendil's co-founder and CEO, Behnam Neyshabur, confirmed the figure, noting it's roughly half of what the company raised in seed funding at a $1 billion valuation in late June.

The agreement reflects two broader industry trends: cloud providers courting startups with big infrastructure commitments, and AI companies locking in compute capacity as they scale. Mirendil's co-founders previously worked at Anthropic, where similar recursive self-improvement research is underway. Other startups, including Recursive Superintelligence and Ricursive Intelligence, have also emerged around this goal.

Self-improving AI, also called recursive self-improvement, refers to systems that iteratively enhance themselves. Mirendil believes this approach could automate much of scientific and AI research, helping in fields like medicine, biology, and materials science. Neyshabur said the goal is to point an AI at a problem and have it improve with time, using Alzheimer's research as an example of an ambitious target.

Training such systems demands enormous compute. Co-founder Harsh Mehta explained that matching workloads to the right hardware is key, and Google's variety of chips allows flexibility to lower costs for both Mirendil and its customers. Google's Amin Vahdat, SVP and chief technologist for AI and infrastructure, said AI progress now depends on orchestrating entire systems, not just chip performance. In return, Google gains a strategic partner building frontier recursive AI that could eventually be offered to enterprise clients. Mirendil hopes its AI will one day handle the work of an entire frontier AI lab.

Source: TechCrunch