Azalia Mirhoseini

Her AlphaChip project was one of the first uses of deep reinforcement learning on a real engineering problem, producing chip layouts in hours.

With future co-founder Anna Goldie she built an AI tool that Google used to help design several generations of its TPU chips.

Helped introduce Mixture-of-Experts architectures for language models, showing efficient training of models with over 130 billion parameters.

Still runs a Stanford lab alongside the startup; its KernelBench benchmark has been adopted by Meta, Nvidia and Cognition AI.

Ricursive raised a $300 million Series A, including from Nvidia's venture arm, at a $4 billion valuation.

Describes Ricursive's goal as a loop: AI makes better chips, and those chips power the next generation of AI.