The AI Infrastructure Constraint Map
A practical map of the bottlenecks shaping the next phase of AI capital spending.
The map covers the full constraint stack and what to watch in each layer:
- Compute — GPUs and accelerators, custom silicon programs
- Memory — HBM and DRAM pricing, the memory-cycle position
- Interconnect — networking and storage
- Physical plant — cooling, data-centre power, grid infrastructure
- The watchlist — key companies and categories to monitor in each layer
- The risk overlay — the macro and geopolitical scenarios relevant to the sector, and the indicators that would challenge the thesis
Download the Constraint Map (PDF)
Free download · July 2026 edition · 5 pages. For the weekly follow-through on every indicator in the map, get the M3I Weekly Briefing.
What's inside
- How to read the map — the constraint → evidence → challenge framework
- Layer 1: Compute — accelerators, custom silicon and the export-control overlay
- Layer 2: Memory — the current binding constraint, with the June–July reporting evidence
- Layer 3: Interconnect — networking and data-centre storage
- Layer 4: Physical plant — power, cooling and the grid (the next likely constraint)
- The watchlist — categories and published coverage examples per layer
- The risk overlay — five weighted geopolitical flashpoints and the two principles that matter for the stack
- The invalidation checklist — six indicators that would challenge the thesis