TL;DR
Luma AI raises $900M led by Saudi Arabia's HUMAIN to build Project Halo, a two-gigawatt AI supercluster for training multimodal world models at peta-scale.
Key Points
- Total funding now exceeds $1 billion; current valuation reported at ~$4 billion
- Project Halo will support training on peta-scale multimodal datasets—10,000x larger than current LLM datasets
- AMD and Amazon provide significant participation alongside HUMAIN, Andreessen Horowitz, and existing investors
- Infrastructure designed for real-time global inference on multimodal AI tasks at frontier scale
Why It Matters
This represents a significant infrastructure play for multimodal AI development. The peta-scale compute deployment enables training on datasets orders of magnitude larger than current LLMs, potentially accelerating progress on world models that can perceive and reason about physical environments—critical for robotics, embodied AI, and real-world applications.
Source: thelettertwo.com