Nvidia Bets $6B on Open-Weight Models

Nvidia is trying to own both sides of the open-model bargain: make weights cheaper and more portable while monetizing the scarce compute needed to train and run them. The Poolside deal combines a $6B technology license, a $1B investment, and a major talent transfer—not an outright acquisition. Nemotron gives developers open models, training recipes, and deployment paths across NVIDIA’s ecosystem. Open weights reduce API lock-in, but they do not eliminate infrastructure costs; large models still demand expensive GPUs, memory, networking, and power. The timing exposes the contradiction: model access is commoditizing while the hardware layer remains concentrated and capable of passing costs to customers. For developers, self-hosting becomes more attractive—but mainly for teams with enough workload volume and capital to justify operating the stack.