Ricursive Intelligence is using AI to rethink chip design
Ricursive Intelligence is building AI systems that design chips, learn from each iteration and aim to make specialised silicon faster to develop.
Ricursive Intelligence says AI-designed chips could help meet the rapidly growing demand for computing power, as the startup works towards automating more of the silicon-design process.1
The company’s founders, Anna Goldie and Azalia Mirhoseini, discussed that plan at TechCrunch Disrupt. Their pitch is not simply to use AI as another chip-design tool: they want systems that can design chips, learn from each iteration and use those lessons to improve later designs.1
Why chip design is under pressure
AI companies are demanding far more computing capacity, but building data centres and manufacturing new processors cannot happen at the same speed as software development. Goldie told TechCrunch that demand for compute could be 10,000 times higher within five years, a projection she described as potentially conservative.1
Ricursive argues that more efficient designs could make better use of data-centre bandwidth, wafers and fabrication capacity. The founders also say that AI can explore layouts engineers might not naturally try, including unconventional arrangements that can still prove effective.1
That approach builds on work associated with AlphaChip, an AI system developed before Ricursive was formed. The founders said the system produced unusual layouts, including a circular “donut-shaped” arrangement of chip components, which engineers initially found surprising but ultimately successful.1
From long projects to custom silicon
The broader ambition is to make specialised chips more practical. Traditional chip design can take six months, a year or longer, which forces companies to be selective about what they build. Ricursive says a much faster automated process could make custom silicon viable for more applications, including space, augmented and virtual reality, and biological computing.1
Ricursive describes itself as a frontier AI lab focused on chip design and says it is backed by $335 million from Sequoia, Lightspeed, DST and NVentures.2
Our opinion
The interesting part of Ricursive’s pitch is not the familiar claim that AI can make engineers faster; it is the attempt to shorten the feedback loop between designing silicon and learning from the result. That could eventually make more specialised hardware economical, but the hard test will be whether designs survive manufacturing constraints and deliver measurable gains outside a presentation. For now, Ricursive has a credible research lineage and a very large problem to solve, rather than a finished answer to the compute crunch.