Local AI perhaps represented the first step towards making AI use private, confidential, and independent of cloud infrastructure, but there was always a catch. In the earlier days, running capable models locally either meant owning flagship-class GPUs or accepting outputs that couldn’t compete with cloud alternatives. That doesn’t seem to be the case anymore.