The VLA Foundation Models Landscape: Vision + Language + Action

The Vision-Language-Action (VLA) model represents AI’s most significant architectural expansion since transformers. Traditional foundation models map input → output (text → text, image → text). VLAs add a third modality: physical action.

What Is a VLA?

Vision + Language + Action = VLA Model

VLAs enable robots to understand visual scenes, process natural language instructions, and generate physical manipulation actions—all in one integrated model.

The VLA Landscape

The Training Data Challenge

VLA models require fundamentally different training data than LLM — as explored in the intelligence factory race between AI labs — s:

• LLMs: Train on internet text (effectively unlimited)
• VLAs: Need embodied trajectories—demonstrations of physical manipulation

This has created an entirely new data economy, with synthetic data generation becoming a competitive moat.

Platform Insight

Unlike LLMs (3-player oligopoly), VLA models remain competitive with 8+ contenders. NVIDIA — as explored in the economics of AI compute infrastructure — ‘s strategy: Enable ALL models via Isaac Sim + Cosmos.

This analysis is part of a comprehensive report. Read the full analysis: Physical AI Is Crossing the Manufacturing Chasm on The Business Engineer.