Mirendil has signed a multiyear cloud agreement worth more than $100 million with Google Cloud to support its work on self-improving AI systems. The deal gives the startup access to Google TPUs, Nvidia GPUs and managed training infrastructure through Google Cloud’s AI Hypercomputer platform.
The agreement comes weeks after Mirendil emerged from stealth with a $200 million seed round led by Andreessen Horowitz and Kleiner Perkins, followed by an investment from Nvidia. The startup says its broader goal is to build AI systems that can increasingly automate AI research itself.
Google Cloud Will Provide TPUs and Nvidia GPUs
Mirendil will use Google Cloud’s AI Hypercomputer for model pre-training, post-training, inference and research workloads. The infrastructure combines Google’s own TPU accelerators with Nvidia hardware, allowing the company to assign different types of work to the systems best suited for them.
Co-founder Harsh Mehta said that flexibility is important because modern AI workloads do not all perform best on the same hardware. Using different accelerators for different tasks could help Mirendil lower computing costs as its systems scale.
Google Cloud said the partnership also reflects a shift toward treating AI infrastructure as a complete system rather than focusing only on individual chips. Its official announcement said Mirendil will use both TPUs and Nvidia infrastructure for its frontier AI research.
Mirendil Is Building AI That Improves AI
Mirendil describes its work as self-accelerating AI research, where AI systems progressively take on more of the experimentation and engineering involved in improving future models. The company ultimately wants these systems to handle ambitious research goals with less continuous human intervention.
Chief executive Behnam Neyshabur said such systems could also be applied outside AI development. One example would be giving an AI a long-term goal related to Alzheimer’s disease and allowing it to keep researching, learning and improving its performance over time.
The company says the same approach could eventually accelerate research across medicine, biology, materials science and other fields. Mirendil outlined the partnership and its research direction in its official Google Cloud announcement.
Former Anthropic Researchers Lead the Startup
Mirendil was founded by Neyshabur, Mehta, Shayan Salehian and Tara Rezaei. Its founding team includes researchers and engineers previously affiliated with Anthropic, Google DeepMind, xAI and OpenAI.
Neyshabur previously worked on AI research at Anthropic and Google, while Mehta helped develop Anthropic’s internal automated research systems. Andreessen Horowitz said the founders are focused on automating the end-to-end AI research process rather than simply building another general-purpose model.
Mirendil’s $200 million seed financing represented a large early commitment for a newly formed AI lab. Its new Google Cloud contract, worth more than half that amount, highlights how quickly access to computing infrastructure has become one of the largest expenses for frontier AI companies.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.