Modernizing Amdahl's Law: How AI Scaling Laws Shape Computer Architecture

This project is a theoretical academic paper (arXiv:2603.20654) rather than a software product. Its 'moat' consists purely of intellectual contribution and potential academic citation. With 0 stars and a very recent upload date, it has no community traction or implementation ecosystem yet. The defensibility is low (2) because mathematical frameworks are not 'proprietary' in a software sense; they are meant to be adopted and potentially superseded by more accurate models as hardware evolves. The primary 'competitors' are established performance models like the Roofline Model or the original Amdahl's Law itself, as well as internal scaling papers from labs like OpenAI or DeepMind (e.g., Chinchilla). While frontier labs define the empirical reality of scaling, they are unlikely to 'compete' with a theoretical framework, making frontier risk low. The risk of displacement is medium-term (1-2 years) as the rapid evolution of AI architectures (e.g., from Transformers to SSMs or new MoE variants) may require further revisions of the math presented here.