NVIDIA Sees Open AI Models Driving a Multi-Model Enterprise Future

NVIDIA Director of Developer Tech Nader Khalil stated at The Six Five Summit 2026 that enterprise artificial intelligence is heading toward a multi-model future where open-source and closed frontier models operate together. Khalil emphasized that enterprise value resides primarily in proprietary data rather than model weights, enabling companies to post-train and fine-tune smaller, open-source models for domain-specific tasks while reserving frontier models for open-ended or complex workloads. Smaller localized models offer reduced latency, lower operating costs, and enhanced control over sensitive intellectual property. To facilitate this shift, NVIDIA continues to release open tooling, datasets, and its proprietary Nemotron model family to resolve deployment bottlenecks. Khalil underscored the growing necessity of model routers such as NeMo Switchyard and agent harnesses to safely route requests across DGX Stations, private enterprise infrastructure, and cloud platforms. Looking ahead, NVIDIA anticipates token consumption from open models will exceed closed frontier models within a year, driving widespread enterprise compute adoption across hybrid architectures.