NVIDIA Unveils Apollo: Open AI Physics Models for Real-Time Simulation

Details

• NVIDIA launched Apollo, a suite of open AI physics models, at the SC25 conference on November 17, 2025, aiming to advance real-time industrial and scientific simulations.
• The initiative features collaborations with nine key industry partners, including Applied Materials, Cadence, LAM Research, Siemens, Synopsys, Northrop Grumman, Rescale, PhysicsX, and KLA, who are embedding Apollo into their platforms.
• Apollo leverages neural operators, transformers, and diffusion models, applying these machine learning techniques to domains like semiconductors, structural mechanics, weather forecasting, computational fluid dynamics, electromagnetics, and multiphysics.
• Initial deployments show strong results, with Applied Materials reporting a 35x acceleration in semiconductor simulations and Synopsys achieving up to 500x speed improvements in computational workflows using AI-based digital twins.
• The models are accessible via build.nvidia.com, HuggingFace, and as NIM microservices, making them readily available for developers to adapt and integrate into custom engineering and scientific use cases.

Impact

With Apollo, NVIDIA accelerates its evolution from GPU giant to core AI infrastructure provider, targeting critical sectors that depend on computational simulation. Swift industry adoption and major performance gains point to wide-reaching impact, potentially reshaping R&D across manufacturing, aerospace, and energy. This launch, alongside other AI model families revealed at SC25, cements NVIDIA's central role in the future of scientific and engineering innovation.