Hardware

Ricursive Intelligence Automates AI Chip Design

Ricursive Intelligence is building AI models to automate chip design in a week, aiming to keep pace with an estimated 10,000-fold surge in compute demand over the next five years.

TechCrunch AI21 hrs agoHardware
Image: TechCrunch AI

Speaking at TechCrunch Disrupt, Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini outlined their plan to tackle an impending computing shortage by automating hardware design. The founders project that global demand for compute capacity will grow by "at least 10,000x" across the next five years. Rather than attempting to solve this surge purely by building thousands of new data centers, the startup focuses on generating far more efficient microchips through iterative, self-improving artificial intelligence.

Instead of relying on conventional semiconductor design methods—which involve breaking silicon layouts down into dozens of isolated blocks—Ricursive Intelligence optimizes entire chips at once. This holistic approach builds upon the co-founders' earlier work with AlphaChip, an AI framework that gained attention for creating an unconventional, donut-shaped arrangement of components. Goldie and Mirhoseini revealed that their latest algorithms are generating intricate, non-traditional structures, including wave-like patterns, that outperform legacy human-engineered layouts.

Drastically shrinking the development timeline is central to the startup's mission. Traditional microchip design typically requires between six months and two years, forcing manufacturers to be highly selective about which specialized architectures reach production. Goldie noted that reducing this design phase down to roughly a single week makes custom silicon financially feasible for niche applications, such as biological computing, space systems, and augmented or virtual reality devices, while offering greater overall energy efficiency.

Ultimately, the founders envision a paradigm shift where human developers supply high-level creative ideas while AI models handle the complex execution of physical layout and software logic. By automating the arduous process of semiconductor architecture, Ricursive Intelligence hopes to make custom, energy-efficient chips widely accessible without exhausting global power grids or physical manufacturing resources.

This is our own summary of reporting by TechCrunch AI

More in Hardware