Mirendil Signs Major Google Cloud Deal for Self-Improving AI

AI laboratory Mirendil has entered into a multiyear strategic partnership with Google Cloud to expand its research into self-improving AI. According to exclusive information from TechCrunch, the deal is worth more than $100 million and clearly highlights two key trends in today’s technology market: cloud giants attracting startups through major infrastructure commitments, and AI companies racing to secure computing power. TechCrunch.com reports .

According to Mirendil founder and CEO Behnam Neyshabursh, the deal is worth approximately half the amount the startup raised in a seed round last June at a $1 billion valuation. The partnership will give the young company access to Google’s TPU chips and NVIDIA GPUs. The startup will also receive dedicated managed training clusters for developing its AI systems.

Recursive Improvement and Future Prospects

The Mirendil team hopes that the AI it is developing will eventually be able to perform the work of an entire leading AI laboratory. Known as recursive self-improvement, this concept involves systems iteratively improving themselves. Notably, major laboratories such as Anthropic, where Mirendil’s founders previously worked, are also pursuing this direction. In recent months, several startups, including Recursive Superintelligence and Ricursive Intelligence, have been founded with similar goals.

According to startup representatives, this process could help automate much of scientific and AI research, enabling scientists to make significant advances in fields such as medicine, biology, and materials science. Behnam Neyshabur believes AI can imitate human scientists’ ability to explore new fields, accumulate knowledge and experience, and gradually improve its performance.

Computing Power and Cost Optimization

However, training self-improving AI requires enormous computing power. According to laboratory co-founder Harsh Mehta, training modern models increasingly depends on matching the right workloads with the appropriate hardware.

Thanks to the flexibility of the various chips and accelerators offered by Google, the company can match workloads with the most suitable hardware. This could significantly reduce costs not only for Mirendil itself but also for customers using its systems.