Ricursive Intelligence bets on AI-designed chips

The startup is building AI that designs chips and improves with each iteration, aiming to speed up chip design.

Ricursive Intelligence, a startup founded by Anna Goldie and Azalia Mirhoseini, wants AI to design the chips that AI runs on. The founders laid out their approach on stage at TechCrunch Disrupt this week.

The idea

The company is building AI systems that design chips, learn from each design iteration, and use that experience to improve the next one. The goal is to automate and accelerate a process that today takes large teams of specialized engineers a long time.

The founders' argument starts with demand. Goldie said the industry expects compute demand to grow by something like ten-thousand-fold over the next five years, and possibly more. Building that many more data centers is not realistic, so the alternative is chips that make better use of power, bandwidth, wafers and fabrication capacity.

Track record

Goldie and Mirhoseini previously developed AlphaChip, an AI system for chip layout. One of its results was an unusual, roughly donut-shaped arrangement of components that surprised engineers at first but turned out to work well. The founders see that as evidence that AI can find layouts humans would not try, rather than simply copying existing practice faster.

Goldie framed the shift more broadly: people contribute ideas and direction, while AI handles more of the execution, a pattern she said is already playing out in software development as well as chip design.

Why it matters for builders

Most developers will never design a chip, but chip efficiency sets the price of everything built on top of it. Inference cost per token, the size of models you can run locally, and how quickly providers can add capacity all depend on hardware improvements. If AI-assisted design shortens chip development cycles, the effect shows up downstream as cheaper and faster APIs.

It is also a clear example of the "AI improving AI infrastructure" loop that many labs are pursuing. For builders, the takeaway is to design products that benefit as model calls get cheaper: features that are too expensive today, such as running an agent continuously or processing every document a user uploads, may become affordable sooner than expected.

You can compare current model APIs and pricing options in our APIs & Model Platforms category.

Source: TechCrunch. This article is PickAI’s own summary and analysis based on the original reporting.