inference

AI agents designed the chip that runs their inference

openTPU is an AI-written inference accelerator whose FPGA card reproduces its reference simulator's tokens bit for bit, checked by an executable spec.