Nvidia's expanded open model families are an infrastructure play dressed as an open source commitment. The $26 billion in financial filings across DGX Cloud and model R&D — a multi-year commitment that its executives are now characterizing as seeding the ecosystem. The models themselves are genuinely open: weights downloadable, architectures published, safety validation pipelines shared with partners like Hirundo running on GB200 NVL72 hardware. What is not open is the assumption embedded in the infrastructure: that running these models at production scale requires Nvidia's…