Skip to main content
Mirendil CEO shakes hands with Google Cloud executive, celebrating a $100M+ AI partnership.

Editorial illustration for Mirendil Secures USD 100M+ Google Cloud Deal for Self-Improving AI

Mirendil Lands $100M Google Cloud Deal for AI

4 min read

Mirendil has signed a multi-year deal with Google Cloud worth more than $100 million, the AI lab's co-founder and CEO Benham Neyshabur told TechCrunch. The agreement gives the startup access to Google's TPUs, Nvidia GPUs, and managed training clusters, all aimed at building what Mirendil calls self-improving AI. That's roughly half of the $1 billion valuation Mirendil landed in seed funding just this past June.

The deal fits a pattern that's become routine in AI right now: cloud providers dangling massive infrastructure commitments to lock in promising startups, while those startups race to grab as much compute as they can before they need it. Google has been especially aggressive on this front.

Mirendil's bet is on recursive self-improvement, systems that get better at their own research and development without constant human intervention. It's a concept Anthropic, where Mirendil's founders previously worked, has explored, and one that's spawned a small wave of new startups, including Recursive Superintelligence and Ricursive Intelligence. Neyshabur's team wants to build something that eventually functions like an entire frontier AI lab, automating the kind of scientific work that could speed up progress in medicine, biology, and materials science.

AI lab Mirendil has signed a multi-year partnership with Google Cloud to source compute capacity for its self-improving AI research, TechCrunch has exclusively learned.

Why this matters

A $100 million-plus compute deal for a startup most of us hadn't heard of last month tells you where the money is actually flowing right now. Google Cloud isn't just selling TPUs and GPUs to Mirendil, it's placing a bet that recursive self-improvement research is worth subsidizing at scale, alongside Nvidia hardware and managed training clusters. For founders, the lesson is blunt: infrastructure access has become a competitive moat, and cloud providers are willing to write large checks to lock in labs working on frontier capabilities before they're proven out.

For researchers, Neyshabur's claim that Mirendil's AI could eventually replace the work of an entire frontier lab deserves real scrutiny rather than applause. That's an enormous claim from a company that hasn't shipped a public model. We'd want to see benchmarks, papers, or at least a working demo before treating "self-improving AI" as anything more than a compelling pitch to Google's sales team.

Watch whether Mirendil publishes research or just burns compute quietly. The deal size is confirmed. The technology's ability to justify it is not.

Common Questions Answered

What resources does Mirendil gain access to through its Google Cloud deal?

Mirendil receives access to Google's TPUs, Nvidia GPUs, and managed training clusters as part of the multi-year partnership worth over $100 million. These computational resources are specifically aimed at supporting the AI lab's research into self-improving AI systems.

How does the $100M Google Cloud deal compare to Mirendil's seed funding valuation?

The Google Cloud deal is worth more than $100 million, which represents roughly half of the $1 billion valuation Mirendil secured in seed funding just months earlier in June. This demonstrates the significant computational investment required to scale self-improving AI research.

Why is Google Cloud willing to subsidize Mirendil's self-improving AI research at scale?

Google Cloud is placing a strategic bet that recursive self-improvement research is worth subsidizing at scale, viewing it as a significant area for future AI development. By providing infrastructure access alongside Nvidia hardware and managed training clusters, Google Cloud is positioning itself as a key enabler of cutting-edge AI research.

What competitive advantage does infrastructure access provide to AI startups like Mirendil?

Infrastructure access has become a critical competitive moat in the AI industry, with cloud providers willing to write large deals to support promising research. For founders, this means that securing compute capacity through partnerships with major cloud providers is now essential to scaling self-improving AI development.

LIVE17:23NVIDIA Unveils Cosmos 3, an Open Physical AI 'Omni-Model