Mirendil inks $100M+ Google Cloud deal to scale self-improving AI

TechCrunch reported exclusively that AI company Mirendil has signed a Google Cloud partnership worth more than $100 million to expand its computing infrastructure. The deal is centered on powering research into self-improving AI systems — systems designed to accelerate scientific discovery and the AI development process itself.

Self-improving AI is one of the most watched, and most contested, research directions in the field: systems that refine their own reasoning and training pipelines based on their outputs and feedback, steadily improving capability with less human intervention. By betting big on this approach and anchoring its compute on Google Cloud, Mirendil is signaling a move from concept-stage work toward scaled experimentation in machines that improve machines.

For Google Cloud, the $100 million-plus agreement is another heavyweight order in its AI cloud business. In the compute arms race, cloud providers are competing not only for training workloads but for the lock-in effect of frontier research customers — once a research program depends on a platform's infrastructure and tooling, switching costs rise sharply over time.

For Mirendil, ample compute is a precondition for self-improving experiments to work at all: these systems typically require repeated training and self-evaluation loops, consuming far more compute than conventional model development. The agreement secures the resources to sustain that cadence, while staking the company's thesis on whether the self-improving paradigm actually delivers.

TechCrunch's exclusive disclosed the deal's size and direction but not its duration, technical specifics, or Mirendil's product roadmap. Worth watching next are whether self-improving systems produce verifiable results in domains like scientific discovery, and whether this deal spurs more frontier AI labs to sign similarly large compute agreements with cloud vendors.