Exclusive: Mirendil’s $100M+ Google Cloud Deal to Supercharge AI
Mirendil just signed a multi-year agreement worth over $100 million with Google Cloud. The frontier AI startup will use Google Tensor Processing Units and Nvidia GPUs to build self-improving artificial intelligence. They want to completely automate the scientific research process.
The Massive Compute Commitment
I have observed that cloud providers now race to lock down emerging AI laboratories. Mirendil raised a $200 million seed round barely two months ago led by Andreessen Horowitz and Kleiner Perkins, with Nvidia participating. Committing at least half that capital to Google Cloud before launching a public product shows how expensive frontier research has become.
To train models that evaluate and improve themselves, you need raw megawattage and flexible hardware. Google Cloud gives Mirendil access to managed training clusters on its AI Hypercomputer. They can shift workloads between Google TPU v5P chips and Nvidia systems instantly to optimize efficiency.
What Self-Improving AI Actually Means
Honestly, most people get this wrong. They think recursive self-improvement just means a chatbot that corrects typos. The reality is much bigger.
Mirendil wants to build “AI for AI for Science”. Instead of forcing every medical or biology lab to fine-tune a general model, the system identifies its own weaknesses, designs improvements, and integrates them autonomously. It removes the human bottleneck from the training loop.
Founders Behnam Neyshabur and Harsh Mehta built similar automated research platforms inside Anthropic and Google. Now they want to deploy that architecture for drug discovery, robotics, and materials science.
The Cloud Infrastructure Race
This partnership proves that compute capacity dictates who wins the next era of development. We track these massive infrastructure shifts in our AI news section. Google gets a seat at the table with an ambitious next-generation architecture while locking in a heavily funded customer.
Check out our breakdown of cloud hosts AI investment to see how Microsoft and AWS fight similar battles.
| Partner | Role in the Deal | Strategic Benefit |
| Mirendil | AI Lab | Secures TPU and GPU compute capacity |
| Google Cloud | Infrastructure Provider | Gains a high-profile frontier AI customer |
| Nvidia | Investor & Hardware Provider | Expands GPU footprint in managed clusters |
Can They Deliver on the Hype?
The commercial outcome remains entirely untested. Mirendil has announced no enterprise customers, no pricing, and no public benchmarks.
They have $200 million in the bank and a massive Google Cloud contract. Investors are betting entirely on the pedigree of a founding team drawn from Anthropic, xAI, Google DeepMind, and OpenAI. Their next milestone must be hard evidence that an automated research loop actually works in a real scientific setting.
FAQ Mirendil and Google Cloud
What is Mirendil?
Mirendil is a frontier AI startup founded by former Anthropic and Google researchers. They recently raised $200 million to build self-accelerating artificial intelligence systems for scientific research.
How much is the Mirendil Google Cloud deal worth?
The multi-year agreement between Mirendil and Google Cloud is worth more than $100 million. It grants the startup access to Google TPUs and Nvidia GPUs for model training through managed clusters.
What is recursive self-improvement in AI?
Recursive self-improvement happens when an AI system identifies its own flaws and updates its architecture without requiring a full human-led retraining cycle. Mirendil plans to use this approach to automate research in medicine, chemistry, and biology.
