Ricursive Intelligence Founders to Make Case for AI-Designed Chips at TechCrunch Disrupt 2026

Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini will appear on the Disrupt Stage at TechCrunch Disrupt 2026, where they plan to lay out their case for letting AI systems take a larger role in designing the hardware they run on. The session puts two of the better-known researchers in machine-learning-for-chip-design in front of an audience of founders, investors and engineers. Organizers are also running a pass promotion that ends today, with discounts of up to $200.

Goldie and Mirhoseini are not newcomers to this problem. Both spent years at Google, where they worked on using reinforcement learning to place components on chip floorplans — a job that has traditionally consumed months of human expert effort. Their 2021 work on chip floorplanning showed a machine could produce layouts in hours that matched or beat human engineers on key metrics, a result that pushed the idea from academic curiosity toward something chipmakers had to take seriously.

Ricursive Intelligence is their bet that the next step is not a single automated task but a closed loop: models that propose designs, get feedback from simulation and manufacturing data, and improve without a human in the middle of every iteration. That framing matters because chip design cycles are long and expensive, and the firms that can shorten them gain an edge in everything from data-center accelerators to edge silicon. If the loop genuinely closes, the argument goes, hardware roadmaps start to move at something closer to software speed.

The timing is pointed. AI companies are spending record sums on compute, and the bottleneck has shifted from whether models can be trained to whether enough suitable chips can be built and powered. Nvidia's dominance in accelerators has pushed hyperscalers and startups alike to explore custom silicon, and every one of those programs needs design talent that is in short supply. Automation that compresses layout, verification or architectural search is one of the few levers that could ease that constraint without waiting on a new fab.

What Goldie and Mirhoseini have not yet shown publicly is how far Ricursive's system reaches beyond floorplanning into architecture selection, verification and the handoff to fabrication — the stages where errors are costliest. Expect the Disrupt conversation to press on exactly that: how much of the pipeline is actually automated today, what still requires human sign-off, and whether customers are willing to trust a model with decisions that carry nine-figure consequences. Their answers will shape how quickly the industry treats AI-designed chips as routine rather than a demo.