NVIDIA announced Cosmos, a suite of open-source generative world foundation models and tools optimized for NVIDIA GPUs, designed to accelerate synthetic-data generation and model training for physical AI applications, particularly robotics and autonomous vehicles. The platform includes state-of-the-art video tokenizers (8x compression, 12x faster processing), accelerated video curation (20 million hours in 14 days vs. three years), and a customizable fine-tuning framework available immediately via Hugging Face and NGC. Leading robotics and AV firms including 1X, Agility, Figure AI, Waabi, Wayve, Uber and XPENG are adopting it immediately, positioning NVIDIA as the primary infrastructure and software stack provider for the physical AI category. This consolidates NVIDIA's moat in compute-intensive AI workloads and extends its platform lock-in from language models to sensorimotor and simulation domains, potentially expanding enterprise GPU attachment rates beyond current LLM-focused budgets.
435 other tracked developments touching the same claims.