NVIDIA Forms a Nemotron Coalition of Open-Model Labs Across Twelve Countries

NVIDIA has positioned its Nemotron open models as the customizable stack for enterprises and governments that want control over the weights they run, and paired that pitch with a new Nemotron Coalition of research institutions across twelve countries that share compute, data and expertise to train frontier models under permissive licenses.

The commercial argument rests on customization economics. NVIDIA says the legal-AI company Harvey post-trained Nemotron 3 Ultra on its own benchmark and reached frontier-class accuracy on complex legal tasks while cutting cost per run by at least a factor of ten against leading closed models. YTL AI Labs post-trained a Nemotron model for the Malaysian language, and Perplexity, Amdocs, Palantir, Cadence and Siemens are named as users.

The move is a deliberate response to two pressures at once. It counters Chinese open-weight suppliers with a US-based open alternative, and it counters Western labs that are restricting access to their weights behind metered APIs. As Constellation Research notes, a chipmaker sponsoring open frontier models also sells more of its own hardware, so the coalition's openness and NVIDIA's commercial interest serve the same goal.