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

AI research firm Mirendil has finalized a multiyear agreement with Google Cloud, securing over $100 million in compute capacity to fuel its ambitious development of self-improving artificial intelligence. The strategic partnership, exclusively reported, underscores a dual trend in the AI sector: major cloud providers are actively engaging startups with significant infrastructure commitments, while AI companies are proactively seeking extensive compute resources to support their scaling operations.

The landmark deal, confirmed by Mirendil co-founder and CEO Behnam Neyshabur, significantly boosts the startup's capabilities. It provides access to Google's Tensor Processing Units (TPUs) and Nvidia's Graphics Processing Units (GPUs), alongside managed training clusters. Mirendil aims to leverage this powerful combination to advance its unique approach to self-improving AI, a technology that iteratively enhances its own performance. The ultimate goal is for this AI to autonomously conduct the work typically handled by an entire frontier AI research laboratory.

The concept of recursive self-improvement in AI, where a system refines its own algorithms and knowledge base, has been a long-standing pursuit for leading AI organizations, including Anthropic, where Mirendil's founders previously worked. The emergence of startups like Recursive Superintelligence and Ricursive Intelligence also signals a growing focus on this specialized area of AI development. Mirendil believes this technology can revolutionize scientific discovery by automating complex research processes in fields such as medicine, biology, and materials science.

Advancing Scientific Discovery Through AI

Behnam Neyshabur elaborated on the vision, drawing parallels between AI's potential and human learning. "You can have a self-improving AI where you can point a problem at it and it keeps getting better with time," he stated. "How can we have an AI system that keeps doing research, keeps improving its own knowledge and performance when it comes to Alzheimer’s disease? This technology allows us to set goals that are ambitious for AI, and the AI would keep making progress." This capability promises to accelerate breakthroughs by enabling AI to continuously learn and adapt.

However, the training of such advanced AI systems demands immense computational resources. Mirendil co-founder Harsh Mehta highlighted the increasing importance of optimizing workloads for specific hardware. "These models are really good at working with different workloads and chips, and assigning the right workloads to the right chips," Mehta explained. "[Google] provides multiple kinds of chips… This flexibility allows us to ultimately mix and match workloads with the right kind of accelerators, and then lower the cost not just for us, but also for our customers using our systems." This emphasis on hardware flexibility is a key element of Google Cloud's offering.

Amin Vahdat, SVP and chief technologist of AI and infrastructure at Google, emphasized Google Cloud's strategic approach. "AI advancement isn’t just about chip-level performance anymore, but how we orchestrate entire systems of intelligence and break through the physical constraints of scaling," Vahdat remarked. Mirendil's sophisticated software and systems layer is expected to enhance the efficiency of Google's hardware, providing the cloud provider with a competitive edge. In return, Google gains a vital partner in developing cutting-edge recursive self-improvement AI, technology that could eventually be offered to enterprise clients seeking advanced AI solutions.

This substantial investment by Google Cloud into Mirendil, which recently secured significant seed funding at a $1 billion valuation, positions both companies at the forefront of the next wave of AI innovation. The collaboration is expected to yield significant advancements in AI's ability to tackle complex scientific challenges and drive progress across multiple disciplines.