Mirendil signs $100M-plus Google Cloud deal aimed at scaling research into self-improving AI

Mirendil, a technology company focused on advanced AI research, has signed a Google Cloud partnership valued at more than $100 million, according to Yahoo Finance. The agreement is intended to give the company expanded access to cloud compute capacity as it works on systems described as “self-improving” and aimed at speeding up both scientific discovery and the development of new AI capabilities.

The reported deal centers on scaling Mirendil’s compute infrastructure, a reference to the high-performance computing resources needed to train and run large AI models. In practice, such partnerships typically support workloads that require significant GPU and data processing capacity, particularly for experiments that iterate quickly and involve repeated model training or simulation.

Yahoo Finance frames Mirendil’s work as building toward AI systems that can improve their own performance over time. The emphasis on self-improvement implies an iterative development loop, where models or surrounding software components continually refine behavior, potentially reducing the time from research hypothesis to measurable results.

Alphabet, via Google Cloud, is increasingly positioned as an infrastructure provider for AI training and deployment, as companies across the industry seek scalable compute and tooling for model development. For Alphabet, deals like this reinforce the strategy of selling cloud capacity and AI services to customers attempting to push the frontier of machine learning.

The Yahoo Finance report does not provide additional contract specifics in the information available here, including the length of the arrangement, whether the figure is a minimum commitment or expected spend, or how costs are structured. It also does not spell out the exact configuration of the compute resources, such as the types of accelerators or workload priorities.

It is also not clear from the report what, if any, exclusivity or exclusivity-like obligations are included, nor whether Mirendil is building primarily for internal R&D or plans to offer outputs to external partners. The absence of disclosure around commercial terms makes it difficult to gauge how much of the agreement is tied to near-term deployment versus longer-horizon research.

From a sector standpoint, large cloud commitments for AI research have become more common as organizations move from proof-of-concept to sustained experimentation. Self-improving AI concepts are especially compute-intensive because they require repeated cycles of training, evaluation, and refinement, all of which can compound compute demand.

One important caveat is that this report is an exclusive market-news account and, based on the information provided here, does not include direct statements from Mirendil or Google Cloud executives, nor does it cite the full contract or regulatory filing. As a result, key details about deliverables, performance targets, and timelines remain uncertain. What is known is the headline value and the stated goal of scaling compute for self-improving AI research, and those specifics are likely to be refined as more information becomes public.