TechCrunch Disrupt 2026: Ricursive Intelligence’s Anna Goldie and Azalia Mirhoseini on when AI starts designing its own hardware

Meru AI, a startup founded by former Google Brain researchers Anna Goldie and Azalia Mirhoseini, is developing artificial intelligence systems that can design their own specialized hardware. Speaking at TechCrunch Disrupt, the co-founders outlined their strategy to overcome a critical bottleneck hindering AI advancement: the limitations of current-generation hardware.

The Hardware Performance Gap

As AI models grow in complexity, they are increasingly constrained by the general-purpose nature of hardware like GPUs. While effective, these chips are not optimized for specific, sophisticated tasks. The alternative, creating custom silicon, is a traditionally slow and expensive process dominated by human engineers, with design cycles that cannot keep pace with the rapid evolution of AI software.

An AI-Powered Design Engine

Meru AI addresses this challenge by using reinforcement learning to automate the chip design process. The platform deploys an AI agent that explores a vast array of potential hardware architectures, making decisions about component configurations like memory size and compute units. Its performance is measured and rewarded based on the efficiency of the resulting chip design for a particular AI model.

Mirhoseini compared the process to DeepMind's AlphaGo, which learned to master the game of Go. In this case, the AI learns to build superior hardware. This approach enables the simultaneous “co-design” of software and hardware, ensuring the final chip is perfectly tailored to its intended AI application. The company reports its system has already produced designs that outperform those created by human experts on certain tasks.

Future Outlook

This automation of hardware design could dramatically accelerate progress across the AI landscape. Goldie projects that within five to ten years, AI could be responsible for designing the majority of its own hardware. By removing the traditional constraints of chip development, Meru AI’s platform could empower a new generation of more powerful and efficient artificial intelligence systems, potentially democratizing access to custom-built silicon for a wider range of companies.