NVIDIA扩展开放模型体系以覆盖代理式、物理与医疗AI三大领域，形成跨行业规模化应用：Nemotron模型支持企业级部署，已被多家公司采用；Kosmos系统服务超过50,000名研究人员，并可并行执行数百项任务，将“数月”研究压缩至“一天”。该体系同时支持“数十亿用户”级本地化模型开发，体现全球化扩展能力。

性能与效率指标显著提升：Nemotron 3 Ultra在Blackwell平台上实现5倍吞吐效率提升；机器人模型GR00T N2在新任务成功率上超过主流模型2倍以上；药物研发中nvQSP实现最高77倍加速，使研究者可在原本仅能模拟少数情景的时间内分析“数百”剂量与人群组合。物理AI模型（Cosmos、Alpamayo）强化多模态推理与实时决策能力。

数据规模与科研能力同步扩张：蛋白质数据库新增约3000万结构预测，其中170万为高置信度结果，显著扩大生物计算基础。Proteina-Complexa模型推动结构设计与实验验证一体化。整体趋势显示，AI从语言处理扩展至“推理+行动+生物模拟”，性能提升范围从5倍至77倍，任务周期从月级压缩至日级，支撑跨行业高复杂度应用。

NVIDIA is expanding open model families across agentic, physical, and healthcare AI, enabling cross-industry scale: Nemotron models support enterprise deployment and are adopted by multiple firms; the Kosmos system serves over 50,000 researchers and can run hundreds of tasks in parallel, compressing months of research into a single day. The framework also supports localized models for billions of users, indicating global scalability.

Performance metrics show major gains: Nemotron 3 Ultra delivers 5x throughput efficiency on the Blackwell platform; the GR00T N2 robot model achieves more than 2x higher success rates on new tasks than leading models; in drug discovery, nvQSP achieves up to 77x acceleration, allowing analysis of hundreds of dose and population scenarios in the time previously needed for only a few. Physical AI models like Cosmos and Alpamayo enhance multimodal reasoning and real-time decision-making.

Data scale and research capacity are expanding simultaneously: about 30 million protein structure predictions have been generated, including 1.7 million high-confidence additions, significantly enlarging biological datasets. The Proteina-Complexa model integrates design and experimental validation workflows. The overall trend shows AI evolving from language processing to combined reasoning, action, and biological simulation, with efficiency gains ranging from 5x to 77x and task timelines compressed from months to days.