Tensor parallelism is ill-suited for local AI setups because it requires a high-bandwidth interconnect like NVLink, which consumer hardware lacks.