Analysis
Mirendil, an AI research lab founded by former Anthropic researchers including CEO Behnam Neyshabur, signed a multi-year partnership with Google Cloud worth more than $100 million for compute capacity, according to TechCrunch. The deal gives Mirendil access to both Google's TPUs and Nvidia GPUs through managed training clusters, which the company will use to develop what it calls self-improving AI -- systems designed to accelerate their own research and development cycle. Mirendil has previously raised $200 million in seed funding.
The company's stated ambition is unusually direct for a lab this young: build AI capable of eventually taking on the research and engineering work of an entire frontier AI lab. That's the same broad direction Anthropic, OpenAI and DeepMind have all pointed toward publicly this year, using language like "AI exponential" and "intelligence explosion" to describe models that increasingly help build the next generation of themselves.
For Google, the deal is as much about distribution as compute revenue -- landing a startup building frontier recursive self-improving AI as a strategic partner gives Google a technology it can eventually offer enterprise customers through its own cloud platform, the same playbook it has run with other AI-native startups building on Google Cloud rather than AWS or Azure. Pulse has previously covered Mirendil's seed round and early lab formation, and this compute deal is the first real test of whether that early funding turns into a working self-improving system.
