Mirendil Secures $100M+ Google Cloud Deal for Self-Improving AI

AI research firm Mirendil has secured a significant multi-year partnership with Google Cloud, valued at over $100 million, to access crucial computing resources. This collaboration will provide Mirendil with Google's Tensor Processing Units (TPUs) and NVIDIA GPUs, along with managed training clusters, to advance its work on self-improving artificial intelligence.

Mirendil's co-founder and CEO, Benham Neyshabur, stated that the deal is worth upwards of $100 million, a figure that represents approximately half of the company's seed funding. The startup's objective is to develop AI systems capable of iterative self-improvement, a concept known as recursive self-improvement, aiming to automate scientific research and accelerate progress in fields such as medicine and materials science.

Neyshabur explained that this AI technology could mimic human scientists' ability to learn, accumulate knowledge, and enhance performance over time, enabling ambitious goal-setting for AI systems. Co-founder Harsh Mehta highlighted the importance of matching workloads to the right hardware, noting that Google's flexible chip offerings allow Mirendil to optimize costs for itself and its future customers.

Google's Amin Vahdat emphasized that AI advancement relies on orchestrating entire systems of intelligence and overcoming physical scaling constraints. Mirendil's software layer is expected to enhance the utility of Google's hardware, potentially giving the cloud provider a competitive edge while establishing Mirendil as a strategic partner in frontier AI development.