# Cross-Pacific AI Chessboard: Concepts MOC Concept notes behind [[The Cross-Pacific AI Chessboard]], grouped by the part of the argument they support. ## 1. The loop: capability crosses the Pacific twice - [[Frontier Models]] - [[API Distillation]] (builds on [[Distillation]]) - [[Open Weights]] - [[Pretraining, Post-Training and Fine-Tuning]] (and [[Parameter-Efficient Fine-Tuning (PEFT)]]) ## 2. Export controls and the efficiency culture - [[AI Chip Export Controls]] - [[Nvidia H800]] (with [[NVLink]], [[InfiniBand]]) - [[Below CUDA - GPU Kernels, PTX and Streaming Multiprocessors]] - [[Systems and Compiler Depth]] - [[DeepSeek Technical Differentiators]] ## 3. Free weights and procurement - [[Open Model Adoption Signals - OpenRouter and Hugging Face]] - [[Token Economics of Inference]] (and [[Inference]]) - [[Neoclouds]] and [[Hyperscalers]] - [[Chinese Frontier Labs]] ## 4. Power vs silicon - [[Grid Interconnection Queues]] - [[Power Infrastructure]] ## 5. Revenue and valuation - [[Run Rate, ARR and Revenue Multiples]] ## 6. What caps self-improvement - [[Evaluation and Verification Infrastructure]] - [[recursive self-improvement]] - [[AI Verification]] ## 7. Decoupling and sovereignty - [[Agent Harnesses and Agentic Search]] - [[Chokepoints in the AI Stack]] - [[Jurisdiction and Control Planes]] - [[Sovereign AI Positioning]]