The agreement grants Mirendil access to a hybrid infrastructure of Google’s proprietary TPUs and Nvidia GPUs. By leveraging managed training clusters, the lab aims to develop models capable of recursive self-improvement—systems designed to automate scientific research in complex fields like medicine and materials science. Mirendil’s founders, who previously worked at Anthropic, believe these models can mimic human expertise, allowing the AI to learn and solve problems iteratively.
In section Startups & Technology
Mirendil Secures $100M Google Cloud Deal to Fuel Recursive AI Research
Mirendil has committed more than $100 million to a multi-year partnership with Google Cloud, securing massive compute capacity for its self-improving AI systems. This deal, confirmed by CEO Benham Neyshabur, represents roughly half of the capital the startup raised in its recent $1 billion valuation seed round.
Co-founder Harsh Mehta emphasized that the partnership hinges on hardware flexibility. By matching specific workloads to the appropriate accelerators, the lab intends to optimize performance and reduce costs. For Google, the deal serves as a strategic play to integrate frontier research into its ecosystem. Amin Vahdat, Google’s SVP of AI infrastructure, noted that the collaboration focuses on orchestrating complex systems to bypass traditional scaling constraints. As Mirendil refines its software layer to maximize hardware efficiency, Google gains a foothold in the competitive race to commercialize autonomous, self-improving intelligence.
Comments (0)
No comments yet. Be the first!