Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini are banking on artificial intelligence to design more efficient chips as a solution to what they describe as an impending compute shortage. Speaking at TechCrunch Disrupt on Wednesday, the pair outlined how their company is building AI systems that can create chip designs, absorb lessons from every cycle, and refine future iterations—all aimed at speeding up and automating the traditionally slow chip design process.

Goldie projected that demand for computing power will surge by "at least 10,000x" over the next five years, a forecast she characterized as "somewhat potentially conservative." She argued that constructing 10,000 times more data centers isn't viable, and that the industry instead needs chips that operate more efficiently, consume bandwidth more effectively, and make better use of manufacturing capacity in wafers and fabrication plants. The founders say AI is already challenging conventional chip design by discovering layouts that human engineers initially found surprising—such as a circular, "donut-shaped" arrangement developed by AlphaChip, the AI system Goldie and Mirhoseini previously created—but which ultimately delivered superior performance. Unlike traditional methods that divide chips into dozens of smaller blocks, Ricursive optimizes entire chips at once, enabling the AI to generate complex patterns including wave-like designs the founders plan to reveal in greater detail later.

According to Goldie, AI-driven automation in chip design will free up human engineers to concentrate on creative problem-solving and application development rather than execution. "I think the way everything is shifting is that humans come bring their great ideas and creative thoughts, and execution is handled by AI," Goldie said. "And this is happening in chip design, software development, and other places." The founders also noted that shorter design cycles—potentially reducing chip development time from six months or two years down to roughly a week—would make it economically feasible to create custom chips for niche applications in areas like space, augmented and virtual reality, and biological computing, while simultaneously improving energy efficiency.

The founders' thesis rests on a simple calculus: if computing demand multiplies by four orders of magnitude while data center expansion can't keep pace, then each chip must deliver vastly more performance per watt and per square millimeter of silicon. AI-designed chips, they argue, can unlock unconventional architectures that human designers might overlook or dismiss, squeezing more computational throughput from the same physical and energy footprint. By automating the iterative refinement process, Ricursive aims to compress design timelines so radically that producing specialized chips for smaller markets becomes profitable—a shift that could democratize access to custom silicon beyond the handful of hyperscale tech companies that can afford lengthy, expensive development cycles today. The accelerated pace could also let chipmakers respond faster to emerging workloads, matching hardware to software needs in near real-time rather than locking in specifications years in advance. If the founders' vision pans out, the bottleneck in computing may shift from fabrication capacity and power budgets to how quickly design teams can iterate—and AI, they believe, is uniquely positioned to break that barrier. Whether the industry embraces AI-generated layouts as readily as it has adopted other automation tools remains an open question, though early results from AlphaChip suggest resistance may fade once the performance gains become undeniable.