Mirendil has signed a multi-year, $100 million-plus partnership with Google Cloud to expand its compute infrastructure. The deal will support Mirendil's research into self-improving AI systems aimed at accelerating scientific discovery and AI development.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
AI lab Mirendil has struck a deal worth over $100 million with Google Cloud to secure TPU and GPU compute for advancing self-improving AI research. The deal underscores the growing ties between cloud giants and startups, as well as the insatiable demand for compute in cutting-edge AI.
According to an exclusive report from TechCrunch, AI lab Mirendil has signed a multi-year agreement with Google Cloud to secure compute resources for its self-improving AI research. Mirendil co-founder and CEO Benham Neyshabur revealed that the deal is worth over $100 million, roughly half of the company's seed round, which closed at a $1 billion valuation in late June.
The deal reflects two major trends in the AI industry: cloud giants attracting startups with massive infrastructure investments, and AI companies racing to lock in compute to ensure scalability.
Self-improving AI, also known as recursive self-improvement, refers to AI systems that iteratively enhance themselves. Mirendil's co-founders come from Anthropic, which is also researching this concept. Additionally, startups like Recursive Superintelligence and Ricursive Intelligence have been founded around this goal.
Mirendil believes this process will automate much of science and AI research, helping scientists make progress in fields like medicine, biology, and materials science. Neyshabur said that AI can mimic human scientists by learning new domains, accumulating knowledge, and gradually improving performance. Users only need to point out a problem, and the AI will continuously improve over time.
He gave an example: an AI system could continuously research Alzheimer's disease, improving its knowledge and performance, thus allowing AI to set ambitious goals and make sustained progress.
Training self-improving AI requires enormous computational power. Co-founder Harsh Mehta said that training increasingly depends on matching the right workloads to the right hardware. Google offers a variety of chips, and this flexibility allows Mirendil to mix and match workloads with accelerators, reducing costs not only for itself but also for its customers.
Google's Amin Vahdat emphasized in a statement that AI progress is no longer just about chip-level performance, but about orchestrating the entire intelligent system and pushing the physical limits of scaling.
Neyshabur said that Mirendil's software and systems layer helps customers better utilize Google's hardware, providing the cloud giant with a potential edge over competitors. In return, Google gains a strategic partner in building frontier recursive self-improving AI, which it can market to enterprise customers in the future.
This report is based on an exclusive from TechCrunch, where the deal amount and company statements come from Mirendil's CEO, representing source claims rather than independent verification. Other background information, such as company valuation and co-founder backgrounds, also comes from the same source.
Mirendil's $100 million deal with Google Cloud highlights the immense compute demands in self-improving AI and the strategic alignment between cloud providers and AI startups. Although the deal details are not independently confirmed, this move indicates that recursive self-improving AI is moving from concept to tangible investment.
Primary report
Primary source