en.Wedoany.com Reported - Mirendil, a startup founded by former Anthropic researchers, announced on June 24 the completion of a seed funding round, raising $200 million at a $1 billion valuation. The company aims to build an AI platform capable of automating the work of AI researchers, seeking to break the monopoly of large AI labs on self-improving technology. Andreessen Horowitz and Kleiner Perkins co-led the round, with Nvidia also participating.
Founders Behnam Neyshabur and Harsh Mehta met at Google in 2019, joined Anthropic in late 2024, and left in December 2025. Current CEO Neyshabur spent over five years at Alphabet, co-leading Gemini's reasoning research. The team of about 20 includes members from Anthropic, xAI, Google DeepMind, and OpenAI, as well as early xAI member Shayan Salehian and 23-year-old MIT graduate Tara Rezaei.
Mirendil plans to build an AI system capable of designing experiments, finding the right settings, evaluating models, and running the next round of training, packaging this capability into a platform open to other organizations. Neyshabur describes it as "AI for AI for science," rather than AI directly for science. For example, a university biology lab without a machine learning team could use the platform to build drug target models, compressing work that would normally take months into just a few days.
This business model targets restrictions in large AI lab service terms that prohibit using their tools to build competing services. As of May, Anthropic stated that its Claude model wrote over 80% of the company's code, but its terms of service forbid customers from using its results to build competitive AI. Matt Bornstein of Andreessen Horowitz noted that when labs refuse to provide means for enhancing models, they are acting as "rational economic agents," and "structurally, there must be an independent company."
The technology involves recursive self-improvement, where AI rewrites its own code without supervision. Anthropic has pointed out potential dangers in this direction, suggesting it could slip out of human control. The founders argue it is the "shortest path" to faster science and can be supervised rather than avoided. This argument comes at a time when Anthropic withdrew access to its most powerful Mythos and Fable models after the Trump administration imposed export controls.
Mirendil's valuation is set against a backdrop of a capital flood in the AI sector. According to Crunchbase data, AI accounted for nearly half of global venture capital investment in 2025, approximately $202 billion, up over 75% year-over-year. The AI infrastructure market generated nearly $337 billion in revenue by the end of 2025, projected to reach $1.2 trillion by 2030. Among lab spin-offs, Ilya Sutskever's Safe Superintelligence raised $6 billion at a $32 billion valuation, and Mira Murati's Thinking Machines Lab raised $2 billion at a $12 billion valuation, both without delivering products. A closer comparison is Periodic Labs, also backed by Andreessen Horowitz, which raised $200 million to apply AI to materials science.
The founders advocate freeing AI research from a handful of labs and putting it in the hands of thousands of users, a democratization argument accompanying every public challenge in Silicon Valley's frontier. Mirendil plans to launch models and products in the coming months. If AI can automate its own research, the current advantages of large labs—thousands of employees and years of accumulated knowledge—could gradually erode. This $200 million will test whether the idea is timely or premature.
