AI lab Mirendil has signed a multi-year partnership with Google Cloud to source compute capacity for its self-improving AI research.
The agreement is worth upwards of $100 million, according to Benham Neyshabur, Mirendil’s co-founder and CEO. That figure represents roughly half of the $1 billion valuation raised by the company during its seed round in late June.
The startup will access both Google’s TPUs and Nvidia GPUs, alongside managed training clusters designed for its self-improving AI projects. The goal is to build a system capable of performing the work of an entire frontier AI lab.
What is self-improving AI
Self-improving AI, also known as recursive self-improvement, describes systems that iteratively improve themselves. Major labs such as Anthropic, where Mirendil’s founders previously worked, have been pursuing this concept. A few other startups, including Recursive Superintelligence and Ricursive Intelligence, have recently emerged with similar ambitions.
Mirendil believes this process will automate significant portions of scientific and AI research, aiding progress in medicine, biology, and materials science. Neyshabur argues that AI can mimic how human scientists learn new domains, accumulate knowledge, and enhance performance over time.
“You can have a self-improving AI where you can point a problem at it and it keeps getting better with time,” he said.
“How can we have an AI system that keeps doing research, keeps improving its own knowledge and performance when it comes to Alzheimer’s disease?” he continued. “This technology allows us to set goals that are ambitious for AI, and the AI would keep making progress.”
Hardware and cost
Training these models demands enormous computing power. Harsh Mehta, Mirendil’s co-founder, noted that training is increasingly about matching specific workloads to the correct hardware.
“These models are really good at working with different workloads and chips, and assigning the right workloads to the right chips,” Mehta said. “[Google] provides multiple kinds of chips […] this flexibility allows us to ultimately mix and match workloads with the right kind of accelerators, and then lower the cost not just for us, but also for our customers using our systems.”
Amin Vahdat, SVP and chief technologist of AI and infrastructure at Google, stated that AI advancement is no longer just about chip-level performance, but about orchestrating entire systems of intelligence and breaking through the physical constraints of scaling.
Neyshabur said Mirendil’s software and systems layer help customers extract more value from Google’s hardware. This arrangement gives the cloud provider a strategic partner building frontier recursive self-improving AI, technology that could eventually be offered to enterprise clients.
What it means
For researchers and developers, this deal removes a major bottleneck: the need to secure massive, custom-built hardware to train complex models. By relying on a flexible cloud environment, Mirendil can focus on the software logic of self-improvement rather than fighting for physical chips. This shift means teams can iterate faster and potentially reduce the financial risk associated with building the next generation of autonomous research tools.
