The substrate becomes a design target of the thing it runs

Models laying out the chips that will run their successors, and fabs built around a recursive loop where masks, fabrication and iteration sit in one building. The bottleneck stops being human design time. The thread opens on 20 March 2026 with Verkor’s agent taking a RISC-V CPU from concept to tape-out in twelve hours; two days later TERAFAB was announced with a recursive design loop under one roof. In April GPT-5.5 topped KernelBench for writing the GPU kernels it runs on, and in June Microsoft unveiled a topological quantum chip designed with its own agent. Kimi K3 autonomously designed a chip in July. August closes the loop: Architect Labs’ Redwood was designed, verified and deployed from a specification alone, OpenAI’s Jalapeño inference chip beat every incumbent part tested, and the model running on Redwood found optimizations for its own operations, two days before the first chip designed end to end by AI was announced. It is the hardware sub-thread of recursive self-improvement.