Mirendil Is in Talks to Raise Up to $1 Billion at a $5 Billion Valuation

Mirendil, an AI startup founded by former Anthropic researchers, is reportedly in talks to raise up to $1 billion at an approximately $5 billion valuation, including the new investment. The prospective round would back the company’s effort to build AI systems that improve themselves with limited human assistance.

The product bet comes before the product

Mirendil says it wants to create widely accessible models that can improve themselves with little to no human help. The approach, known as recursive self-improvement, aims to have AI contribute to the work of developing later AI systems rather than relying entirely on researchers for each advance.

The reported funding would help Mirendil launch its own model and train future ones. Bloomberg’s sources said the company plans a frontier model for engineering and research work by the beginning of 2027, putting the proposed financing behind a planned release rather than an available product.

A proposed price, not a completed round

Kleiner Perkins is reportedly discussing leading the new financing, while Andreessen Horowitz is discussing an investment. Both firms led the earlier seed round. Mirendil declined to comment, and neither the investment amount nor the valuation is final.

The reported $5 billion figure is a proposed valuation including the new investment, not a measure of cash already raised or a completed market price. It nonetheless shows the scale of investor interest being contemplated for a company with more than 20 employees and an ambitious research agenda.

Compute and safeguards shape the bet

Training and running AI models is costly, and Mirendil is pursuing a cheaper, more automated way to build them. It has announced a Google partnership using Google Cloud TPUs to scale model development, and it works with Nvidia computing infrastructure; Nvidia is also an investor.

Self-improving AI also raises security concerns if it is achieved. Mirendil has told stakeholders it wants to broaden access to advanced AI while addressing the technical challenges of strong safeguards. How those safeguards would work remains an open question for the company as it seeks new backing.