Ricursive Intelligence Targets Chip Design Bottleneck to Solve Compute Crisis
AI-generated image · Bay Street Wire
Founders Anna Goldie and Azalia Mirhoseini argue that automating silicon architecture is the only way to meet a projected 10,000x increase in compute demand.
The primary constraint on AI growth may not be the number of available GPUs, but the slow, human-led cycle of designing the silicon itself. According to reporting from TechCrunch, Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini are developing AI designed to automate and accelerate the chip design process to combat a looming compute crisis.
Speaking at TechCrunch Disrupt, Goldie noted that compute demand could increase by at least 10,000x over the next five years. She argued that meeting this demand by simply building more data centers is unsustainable, suggesting instead that the industry must create chips that more efficiently utilize wafers, fabs, and bandwidth.
Traditional chip design is a slow process that can take months or years, forcing companies to be extremely selective about which chips they develop. Ricursive Intelligence aims to reduce this timeline to as little as a week, which Goldie says would make custom chips financially viable for specialized applications in AR, VR, space, and biological computing.
Unlike traditional methods that partition chips into small blocks, Ricursive's approach optimizes the entire chip. This allows the AI to discover unconventional, high-performance layouts. As an example, the founders cited AlphaChip, a system they previously developed that created a successful, "donut-shaped" arrangement of components that surprised human engineers. Goldie stated that while AI handles the execution and intricate patterns, humans can focus on creative ideas and application development.