Mirendil Secures $100 Million Google Cloud Deal to Scale Self-Improving AI
Mirendil, a stealth San Francisco-based frontier AI research lab founded by prominent ex-Anthropic and Google Brain researchers, has officially announced a multi-year, $100 million-plus infrastructure partnership with Google Cloud. The deal will power Mirendil's mission to build autonomous, self-improving AI systems capable of automating the core processes of machine learning research and development.
The AI Hypercomputer Hybrid Advantage
Officially detailed on August 21, 2026, the partnership grants Mirendil access to Google Cloud's AI Hypercomputer, an "agent-native" infrastructure stack. Crucially, the deal does not lock Mirendil into a single hardware architecture; instead, Google Cloud will provide a hybrid mix of Google's custom Tensor Processing Units (TPUs) and NVIDIA's full-stack GPU systems.
This dual-hardware setup allows Mirendil to allocate different pre-training and post-training workloads to the most efficient chip architecture. The lab is building AI agents designed to automate the entire scientific experimentation loop, including proposing experiments, tuning hyperparameters, running training runs, and evaluating results autonomously.
Behnam Neyshabur, co-founder and CEO of Mirendil, explained the strategic importance of the infrastructure deal:
"Expanding on Google Cloud gives us the scale and flexibility to push those systems further and put frontier AI research capabilities in the hands of many more scientists and engineers to run that loop faster and at greater scale."
By leveraging Google Cloud's infrastructure, Mirendil aims to bypass the massive capital bottlenecks associated with building proprietary data centers, allowing a small, highly focused team of researchers to scale frontier-class model development rapidly.