Mirendil Signs $100M-Plus Google Cloud Deal for Self-Improving AI

What Happened

AI startup Mirendil has signed a partnership agreement with Google Cloud valued at more than $100 million, the company announced Wednesday. The deal is structured to expand Mirendil's compute infrastructure and support ongoing research into self-improving AI systems, according to reporting by TechCrunch.

Background

Mirendil is focused on developing AI systems capable of iterative self-improvement, a research area that involves AI models refining their own capabilities over successive cycles without equivalent increases in human-directed training. The field sits at one of the more active and closely watched frontiers in AI development, attracting investment from both established technology companies and venture-backed startups.

Google Cloud has pursued a series of infrastructure partnerships with AI-focused companies as demand for large-scale compute resources has grown. Cloud providers including Google, Microsoft Azure, and Amazon Web Services have each moved to position their platforms as primary infrastructure layers for AI research and deployment.

Terms of the Deal

The agreement exceeds $100 million in total value, according to TechCrunch, which described the reporting as exclusive. Specific contract duration, payment structure, and the types of Google Cloud services covered were not disclosed in available wire reports. The partnership is described as covering compute infrastructure, the foundational layer of hardware and networking resources required to train and run large AI models.

Mirendil has not publicly stated the current scale of its models, the specific benchmarks its self-improving systems have reached, or the timeline by which it expects the expanded infrastructure to affect its research output.

Why Compute Scale Matters for This Research Area

Self-improving AI research is computationally intensive by design. Systems that revise their own parameters or architectures require repeated training runs, evaluation cycles, and comparison against prior versions, all of which multiply the demand for processing power relative to conventional model development. Access to large-scale cloud infrastructure is widely treated in the industry as a prerequisite for advancing this category of research at speed.

The Google Cloud partnership is intended to remove a ceiling on that process, allowing Mirendil to run more extensive experiments than its existing infrastructure would support.

Context: Leadership Changes at Google DeepMind

The Mirendil announcement arrives alongside reported leadership changes at Google DeepMind, Google's primary in-house AI research organization. The Guardian reported this week that DeepMind's chief executive is stepping down, and separately, two senior engineers are departing to launch a new startup identified in other reports as Discovery Loop. Those departures have prompted commentary about competitive dynamics within AI research, though Google has not publicly addressed the strategic implications of the changes.

Mirendil's deal with Google Cloud is a commercial infrastructure agreement and is separate from Google DeepMind's research operations. Google Cloud and Google DeepMind operate as distinct units within Alphabet.

What the Partnership Does Not Cover

Available reports do not specify whether Mirendil retains full intellectual property rights over research conducted using Google Cloud infrastructure, a standard point of negotiation in large cloud partnerships. No regulatory filings related to the deal have been reported. Mirendil has not announced customer deployments, product releases, or external benchmarking of its self-improving AI systems in connection with this announcement.

What Comes Next

Mirendil has not announced a public timeline for research milestones or product releases tied to the expanded Google Cloud infrastructure, and no regulatory review of the partnership has been reported.