AI Designs a Chip in 2 Weeks—A Job That Usually Takes Years

A startup says it used AI to design and verify a computer chip in just two weeks—a process that typically takes over a year and costs hundreds of millions of dollars. The company, Architect Labs, says its AI handled the detailed design work while two human architects set and refined the overall plan. The result is a chip called Redwood, which can run AI models like Meta’s Llama and Alibaba’s Qwen. Ebrahim Hussain, the 20-year-old cofounder and CEO of Architect Labs, said the speed matters because AI models evolve faster than the chips they run on. He added that after running an AI model on Redwood, the model found ways to improve the chip itself—a "self-reinforcing loop" he believes could lead to AI building better hardware over time. The startup emerged from stealth in June with $24 million in seed funding led by Kindred Ventures. Google’s former chief scientist Jeff Dean and executives from OpenAI and Nvidia also backed the company. Researchers called the two-week timeline "impressive," but noted that the chip hasn’t been manufactured yet. So far, Redwood has only been tested on reprogrammable hardware called a field-programmable gate array (FPGA). Those tests suggest the finished chip could be more efficient than Nvidia’s Jetson Orin Nano, an AI supercomputer used in robotics. Hussain said the company plans to send Redwood to Taiwanese manufacturer TSMC for production but hasn’t set a date because its AI is still improving the design. Experts say manufacturing will be the real test. Hao Zheng, an assistant professor at the University of Central Florida, said it will reveal how fast, efficient, and reliable the chip actually is. Matthew Guthaus, a professor at UC Santa Cruz, warned that AI "can also make silly mistakes and needs close supervision and double-checking," as errors can cost millions if not caught before production. Hussain says Architect’s AI checks its own work for mistakes, reducing the need for human oversight. He believes AI could usher in a "designless" era for the chip industry—making custom chips accessible without the large teams traditionally required.