Designing the chips required for advanced AI currently takes two to three years. Ricursive Intelligence aims to compress that cycle into weeks by having AI systems assist in the design process.
At TechCrunch Disrupt 2026, co-founders Anna Goldie and Azalia Mirhoseini present their work on closing the loop between artificial intelligence and hardware development.
What happens when AI designs the chips that power it?
Goldie and Mirhoseini previously co-led AlphaChip at Google. That system generated chip layouts in hours, a task that previously required human designers much longer periods. Their work contributed to multiple generations of Google’s Tensor Processing Units.
Ricursive Intelligence extends that concept by building tools to automate and accelerate the entire design workflow. The system learns across different chips, meaning experience gained from one project improves the approach for the next.
This creates a feedback loop: AI designs better hardware, that hardware supports more powerful AI, and those systems subsequently help develop the next iteration.
Secure your Disrupt pass with up to $200 in savings to hear directly from the founders working to build this system.
From AlphaChip to a billion startup
Both founders co-founded Google’s ML for Systems team. They were early employees at Anthropic and senior staff research scientists at Google DeepMind.
Goldie holds a Ph.D. in computer science from Stanford and was named one of MIT Technology Review’s 35 Innovators Under 35. Mirhoseini is an assistant professor of computer science at Stanford and founder of its Scaling Intelligence Lab.
Ricursive launched in late 2025. Investors moved quickly. Within four months, the company raised $335 million at a $4 billion valuation, including a $300 million Series A. Nvidia is among the investors.
The company wants AI to automate complex parts of chip design, from component placement through design verification. By accelerating development, Goldie and Mirhoseini believe they can open the door to new architectures and ultimately more capable, efficient AI.
How fast can the AI-hardware loop move?
AI models have advanced rapidly, but the underlying hardware operates on a different development cycle. Ricursive bets that AI can help narrow that gap.
At Disrupt, Goldie and Mirhoseini bring the perspective of researchers who proved AI could design real-world chips and are now building a company around that idea.
Their session is one of 200+ sessions across six industry stages, roundtables, and breakouts at Disrupt, October 13–15 at Moscone West in San Francisco. More than 10,000 founders, investors, operators, and tech leaders are expected, along with 250+ speakers and 300+ exhibiting startups.
Matchmaking, dealmaking, and networking create opportunities to connect with the founders, investors, and builders shaping what comes next.
What it means
For people building AI, the implication is a potential end to hardware bottlenecks. If chip design shrinks from years to weeks, the pace at which new models can be trained and deployed will likely increase. This could allow faster iteration on capabilities that currently wait on silicon cycles.
Secure your ticket before midnight PT. Get a second pass for 50% off select ticket types.
