Exclusive: Mirendil Signs Over $100M Google Cloud Deal to Boost Self-Improving AI

AI lab Mirendil has entered a multi-year agreement with Google Cloud to secure compute resources for its self-improving AI research, as exclusively reported by TechCrunch.
This partnership reflects two key trends in the AI sector: major cloud providers are attracting startups with substantial infrastructure commitments, while AI companies are aggressively securing compute deals to ensure access as they expand.
The agreement is valued at over $100 million, according to Mirendil’s co-founder and CEO, Benham Neyshabur. This figure represents approximately half of the seed funding Mirendil raised at a $1 billion valuation in late June.
The deal grants the startup access to both Google’s TPUs and Nvidia GPUs, along with managed training clusters for developing its self-improving AI. The company aims to create an AI capable of performing the work of an entire frontier AI lab.
Self-improving AI, or recursive self-improvement, involves systems that iteratively enhance themselves. This concept is being explored by major labs like Anthropic, where Mirendil’s co-founders previously worked. Several startups, including Recursive Superintelligence and Ricursive Intelligence, have recently emerged with similar goals.
Mirendil believes this process will automate much of scientific and AI research, aiding progress in fields such as medicine, biology, and materials science.
Neyshabur suggests that AI can replicate how human scientists learn new domains, accumulate knowledge, and gradually improve performance. “You can have a self-improving AI where you can point a problem at it and it keeps getting better with time,” he explained.
“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 added. “This technology allows us to set goals that are ambitious for AI, and the AI would keep making progress.”
Training self-improving AI requires significant computing power. Co-founder Harsh Mehta noted that training increasingly involves matching the right workloads to the appropriate hardware.
“These models are really good at working with different workloads and chips, and assigning the right workloads to the right chips,” Mehta stated. “[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.”
This flexibility is central to Google’s AI infrastructure strategy. Amin Vahdat, SVP and chief technologist of AI and infrastructure at Google, stated in a press release that AI advancement is no longer just about chip-level performance, “but how we orchestrate entire systems of intelligence and break through the physical constraints of scaling.”
Neyshabur indicated that Mirendil’s software and systems layer help customers maximize Google’s hardware, giving the cloud provider a competitive edge. In return, Google gains a strategic partner developing frontier recursive self-improving AI — technology it can eventually offer to enterprise clients.
Related article
QueryStory urges caution on AI-generated content
In 2009, Shapor Naghibzadeh discovered the power of compelling storytelling while working as a Google sysops engineer. During Operation Aurora, when state-sponsored hackers targeted the tech giant, he was summoned to a crisis room to clarify server a
French Startup ZML Unveils Free Tool to Accelerate AI Inference Across Multiple Chips
Nvidia’s era of unchecked market supremacy may be winding down, yet competitors and alternatives are emerging from every angle.ZML, a prominent French AI startup backed by Turing Award laureate Yann LeCun, has unveiled inference-performance software
Anthropic Unveils Opus 4.6, a Controversial Content Generator
Anthropic’s universal usage standards for Claude prohibit the model from generating sexually explicit content, including depicting or requesting sexual intercourse or sex acts, generating content related to sexual fetishes or fantasies, or engaging i
Related Special Topic Recommendations
Comments (1)
0/500

AI lab Mirendil has entered a multi-year agreement with Google Cloud to secure compute resources for its self-improving AI research, as exclusively reported by TechCrunch.
This partnership reflects two key trends in the AI sector: major cloud providers are attracting startups with substantial infrastructure commitments, while AI companies are aggressively securing compute deals to ensure access as they expand.
The agreement is valued at over $100 million, according to Mirendil’s co-founder and CEO, Benham Neyshabur. This figure represents approximately half of the seed funding Mirendil raised at a $1 billion valuation in late June.
The deal grants the startup access to both Google’s TPUs and Nvidia GPUs, along with managed training clusters for developing its self-improving AI. The company aims to create an AI capable of performing the work of an entire frontier AI lab.
Self-improving AI, or recursive self-improvement, involves systems that iteratively enhance themselves. This concept is being explored by major labs like Anthropic, where Mirendil’s co-founders previously worked. Several startups, including Recursive Superintelligence and Ricursive Intelligence, have recently emerged with similar goals.
Mirendil believes this process will automate much of scientific and AI research, aiding progress in fields such as medicine, biology, and materials science.
Neyshabur suggests that AI can replicate how human scientists learn new domains, accumulate knowledge, and gradually improve performance. “You can have a self-improving AI where you can point a problem at it and it keeps getting better with time,” he explained.
“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 added. “This technology allows us to set goals that are ambitious for AI, and the AI would keep making progress.”
Training self-improving AI requires significant computing power. Co-founder Harsh Mehta noted that training increasingly involves matching the right workloads to the appropriate hardware.
“These models are really good at working with different workloads and chips, and assigning the right workloads to the right chips,” Mehta stated. “[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.”
This flexibility is central to Google’s AI infrastructure strategy. Amin Vahdat, SVP and chief technologist of AI and infrastructure at Google, stated in a press release that AI advancement is no longer just about chip-level performance, “but how we orchestrate entire systems of intelligence and break through the physical constraints of scaling.”
Neyshabur indicated that Mirendil’s software and systems layer help customers maximize Google’s hardware, giving the cloud provider a competitive edge. In return, Google gains a strategic partner developing frontier recursive self-improving AI — technology it can eventually offer to enterprise clients.
QueryStory urges caution on AI-generated content
In 2009, Shapor Naghibzadeh discovered the power of compelling storytelling while working as a Google sysops engineer. During Operation Aurora, when state-sponsored hackers targeted the tech giant, he was summoned to a crisis room to clarify server a
French Startup ZML Unveils Free Tool to Accelerate AI Inference Across Multiple Chips
Nvidia’s era of unchecked market supremacy may be winding down, yet competitors and alternatives are emerging from every angle.ZML, a prominent French AI startup backed by Turing Award laureate Yann LeCun, has unveiled inference-performance software
Anthropic Unveils Opus 4.6, a Controversial Content Generator
Anthropic’s universal usage standards for Claude prohibit the model from generating sexually explicit content, including depicting or requesting sexual intercourse or sex acts, generating content related to sexual fetishes or fantasies, or engaging i





Home






