Former Google Chief Scientist Jeff Dean suggests that artificial intelligence can reduce the development cycle for semiconductor chips from the current 18–30 months to 3–6 months. This approach relies on reinforcement learning and automated search to optimize design workflows.
The concept builds on Google's AlphaChip system, which was detailed in a 2021 Nature paper. AlphaChip automates chip layout tasks in under six hours and has been used in production for Google's Tensor Processing Units to support machine learning infrastructure.
Industry adoption is already underway. In September 2026, Architect Labs developed its Redwood inference chip from specification to proof-of-concept in less than two weeks. OpenAI also utilized large language models to design its Jalapeño accelerator.
OpenAI completed the RTL-to-tape-out phase for its chip in nine months, achieving a 10% reduction in chip area compared to manual designs. The total development time for the Jalapeño project was under 20 months.
Jeff Dean departed Google in mid-2026 to co-found Discovery Loop. The startup focuses on creating systems for recursive self-improvement in hardware design automation to further accelerate the development of specialized silicon.
