Lockheed Martin

Solution

Lockheed Martin built an internal AI factory on NVIDIA DGX SuperPOD to centralize training, inference, MLOps, and generative AI for high-assurance workloads. The platform gives developers secure GPU environments in minutes instead of weeks, hosts open-source LLMs internally, and uses retrieval-augmented generation against proposal repositories and other sensitive data without moving that information to public APIs. The factory combines DGX SuperPOD infrastructure with Lockheed Martin’s customized MLOps platform, NVIDIA Triton Inference Server, NVIDIA NeMo Framework, CUDA, and NVIDIA AI Enterprise so teams can fine-tune models, run enterprise knowledge tools such as Text Navigator, and support internal chatbots and coding assistants. Lockheed Martin also keeps burst capacity for non-classified cloud work, but the on-prem platform handles tokenization, training, and deployment in-house, which reduces external token fees and creates a repeatable path from prototype to production across many programs.

Data flow

Sensitive internal data stays on approved on-prem infrastructure, DGX SuperPOD-backed models are trained or fine-tuned through the AI factory, RAG connects proposal and other repositories to internal LLMs, and tools like Text Navigator return secure answers or generated content to engineers and proposal teams.