Open-weight AI just caught the usable frontier. NVIDIA's Nemotron 3 Ultra: 550B params, but only 55B fire per token — near-frontier reasoning without frontier compute bills. Weights and data are public. The humbling part? Best US open model, and still behind China's open releases. If open weights are 'good enough,' the moat was never the model — it's distribution. www.contentbuffer.com Open or closed — what would you build on? #AI #OpenSource #NVIDIA #LLM #TechNews #AIengineering