Today, we’re announcing a multi-year partnership with Google Cloud to power the next generation of self-accelerating AI. Mirendil will scale its training, inference, and AI research workloads on Google Cloud’s AI Hypercomputer — drawing on both Google Cloud TPUs and full-stack NVIDIA AI infrastructure.
Removing the bottleneck in AI progress
Progress in AI has been bounded by how fast humans can run the research loop: designing experiments, evaluating results, and deciding what to try next. Mirendil is building self-accelerating AI that can eventually take on the work of an entire frontier AI lab, automating and continuously improving the research process itself and making AI development faster and more autonomous. Our long-term mission is to democratize frontier AI R&D, so that teams in medicine, biology, materials science, and beyond can build specialized AI systems without needing the infrastructure of the world’s largest AI labs.
Scaling the research loop
Building self-improving AI systems requires running increasingly complex, end-to-end workloads across large compute environments: pre-training through post-training, reinforcement learning at massive scale, and research loops that run many experiments in parallel. Mirendil’s systems are designed to run across accelerators, optimizing the software stack above the chip layer and matching different workloads to the most effective compute available.
Google Cloud’s AI Hypercomputer, including managed training clusters co-designed with our team across compute, storage, networking, and control planes, gives us the TPUs and NVIDIA’s full-stack accelerated computing platforms to do that at frontier scale. This lets us run larger and more sophisticated research loops while reducing the operational complexity of keeping frontier-scale experiments running reliably, for both humans and agents.
“We are at an inflection point in which AI advancement is no longer just about chip-level performance, but how we orchestrate entire systems of intelligence and break through the physical constraints of scaling,” said Amin Vahdat, SVP and Chief Technologist, AI and Infrastructure at Google. “By building on Google Cloud’s AI Hypercomputer – leveraging co-designed hardware, networking, and software across the widest portfolio of TPUs and GPUs – Mirendil can run complex training workloads with maximum efficiency, accelerating the research loops that will power the next era of scientific insights.”
What’s next
This partnership deepens our relationship with Google and expands our compute foundation across the industry’s leading providers. It moves us toward a future where frontier AI R&D is accessible to everyone, raising our collective ambition to take on the most critical scientific problems.
Read more about the partnership on the Google Cloud blog.