NVIDIA, SK Telecom and Rebellions agree to split inference between NVIDIA GPUs and Korean NPUs in sovereign AI racks · Polylog