PILLAR GUIDE · AI INFRASTRUCTURE
AI Infrastructure Bottleneck Map
Signal & Flow’s AI infrastructure research in one reading order: from GPUs to networking and memory, then power and data centers, and finally whether the capital spent comes back as cash.
- GPU
- Networking·Memory
- Power·Grid
- Data center
- Payback
1. The big picture: bottlenecks are profit pools
Rather than guessing the next single winner after GPUs, first identify which bottlenecks get scarcer as demand grows.
2. Inside the compute: networking, memory, tokens per watt
After the GPU count race, the KPIs become networking, memory, training-loop efficiency and tokens per watt.
3. Physical infrastructure: power, sites, data centers
The slowest steps today are grid connection and data-center completion. The dashboard tracks the stage of 12 major projects.
4. Payback: does CapEx come back as cash?
As bottlenecks ease, the question becomes who recovers the money. We track payback speed by layer and the credit risk behind it.
Last updated 2026-10-31. New AI infrastructure notes are added to this map.