NVIDIA officially revealed at SC25 on November 17 2025 its NVIDIA Apollo. According to reports, the models will be delivered via NVIDIA’s developer portals (e.g., build.nvidia.com), the Hugging Face model hub, and NVIDIA’s NIM microservices.
NVIDIA Apollo is a collection of domain-specific AI models optimised for physics-based simulation tasks: think structural mechanics, computational fluid dynamics (CFD), electromagnetics, semiconductor device modelling and weather/climate forecasting.
The idea is to provide pre-trained checkpoints and reference workflows that developers and engineers can use for training, inference and benchmarking — then fine-tune them for their particular simulation workloads. Apollo leverages advanced neural-operator, transformer and diffusion-based architectures with domain knowledge baked in, enabling high-fidelity, GPU-accelerated modeling.
Potential Users
The users for Apollo are engineers and researchers across industries where complex simulation and modelling are core to product development or scientific discovery. Some key user groups include:
Semiconductor manufacturers and device-design teams (for defect detection, computational lithography, electro-thermal modelling) — e.g., Applied Materials is cited as an early adopter.
Aerospace, automotive and consumer electronics firms (for structural mechanics, fluid dynamics, multi-physics) — firms such as Siemens, Northrop Grumman are mentioned in this regard.
Weather and climate researchers (for global/regional forecasting, downscaling, data assimilation) — these users benefit from large-scale physics modelling.
Electromagnetics and communications system developers (for radar, optical systems, wireless design) — anything requiring accurate EM simulation.
Benefits
The Apollo model suite offers several tangible advantages:
Speed and efficiency: Traditional simulation workflows can be time- and resource-intensive. Apollo’s AI-surrogate models can dramatically accelerate these processes — NVIDIA touts speedups such as “orders of magnitude faster” while preserving fidelity.
Cost reduction: Faster turnaround means fewer compute-hours and potentially less specialised hardware for some tasks. Resources previously tied up in long simulations become available for design-iteration cycles.
Broader exploration of design space: With faster simulation, engineers can explore many more scenarios, variants and “what-if” cases. This leads to better optimisation, innovation and reduced time-to-market. For example, semiconductor firms can evaluate more process variations or device configurations in less time.
Integration and flexibility: Because Apollo offers reference workflows and checkpoints, organisations don’t have to build their simulation stack from scratch. They get a foundation, then customise it. That makes adoption easier than starting from zero.
Future-proofing with AI physics: As simulation meets AI, users gain access to capabilities that meld first-principles physics with machine learning. This convergence means simulation tools of the future may depend increasingly on models like Apollo, so early adoption gives a competitive edge.
Conclusion
NVIDIA Apollo is more than just another AI model bundle — it represents a strategic push to bring AI-acceleration into core engineering workflows across science and industry. By opening up a family of physics-specific, high-performance models, NVIDIA is positioning itself at the heart of simulation and design ecosystems. For organisations working with complex physical systems, Apollo offers an opportunity to go faster, iterate smarter and build more optimised products or systems in less time.
Compiled using AI
