llm-d 1.0: Red Hat, IBM, Google and Nvidia bet that AI’s real bill is inference, not training Martin Cid Magazine