~12k for 80 core gpu with 256gb, 14k in October for 512gb. Seems like that could make for a very descent on prem inference server.