AI Model Reimagines Spacecraft Procurement for the Pentagon

A new artificial intelligence model aims to trim development and acquisition timelines for space systems by reducing risks in the development and testing phases of engineering, tech leaders shared Tuesday at the NVIDIA GTC event in Washington, D.C.

The physics AI foundation model, developed by Luminary Cloud in partnership with Northrop Grumman and powered by NVIDIA’s PhysicsNeMo framework, integrates physics into machine learning to address early-stage risk, a fundamental problem in defense and aerospace acquisition.

“The implications on the design of aerospace systems, but also on the procurement of these systems, is profound,” Luminary Cloud CTO and Co-Founder Juan Alonso told GovCIO Media & Research in an interview. “Physics AI is one of the key technologies that’s going to emerge to help us sort of shorten those design cycles and reduce risk early in the early stages of the design process.”

Alonso added that physics AI is the “revolution that’s still to come” because it is fundamentally different from traditional large language models. These models predict the behavior of complex physical systems such as aerodynamics, heat transfer and structural analysis by incorporating physical laws and symmetries into the training process.

“We are using advanced AI and machine-learning techniques with physical data,” Alonso said. “That may be simulation, it may be experimental data, operational data to build models of the physical world that can be used to design better systems more quickly, compress timelines and look at more alternatives.”

The model reduces the analysis time for a single simulation case from hours to mere seconds while preserving the necessary engineering-grade accuracy, according to Alonso. This acceleration allows engineers the freedom to explore entire design spaces interactively and in real time.

Addressing Risk in Aerospace Engineering

In the traditional engineering process, 90% of a system’s future performance, capability and cost freeze during the conceptual design phase, he said. Because engineers often use low-fidelity tools at this stage, reliance on incomplete information leads to “costly redesigns, delays, [and] additional money” later in the program.

“More information earlier on is always better,” he said. “Physics AI models give you the ability to have that information of very high quality as early as possible in the design process.

For Northrop Grumman, this technology represents a major leap forward in spacecraft development, enabling faster delivery with greater performance and more robust safety margins.

“Using Al to make something small, like a spacecraft thruster, puts us on a path to do much bigger things, like using Al to design larger components or even an entire spacecraft,” said Han Park, vice president of AI integration at Northrop Grumman Space Systems.

The leaders said the model’s speed and precision also support the government’s push to move away from overly prescriptive requirements toward outcome-based problem solving. That shift allows agencies to work with contractors to iterate faster and explore more options.

“This announcement signals more than just a technical achievement — it is an invitation to reimagine the very process of spacecraft design to be faster, smarter and more adaptable,” Alonso said.