“At CES in January, NVIDIA released a "model router" that can intelligently route queries between an open source model and a foundational model.”