ai-infra-bottlenecks

AI Infrastructure Bottleneck → Equity Mapper

Output language: match the user's language. Body/reference files are in English.

Goal: turn "where is the AI buildout constrained right now" into a structured, evidence-backed list of public-equity candidates, organized by bottleneck category — not a single stock pick, not a portfolio, not a DCF.

1. Scope the request

Infer scope from the request; ask only what changes the deliverable (one grouped askUser pause, skip if already answered or safely inferable):

• Market scope: US-only, or include HK/Japan/Korea/Taiwan/China listings that are core to this supply chain (foundry, HBM, optical components are mostly non-US). Default to global-if-relevant since this supply chain is not US-centric — state the assumption if you proceed without asking.
• Depth: all four categories in the trigger (semis, photonics, networking, robotics) plus power/ memory as bottleneck context, or a narrower subset the user named.
• Horizon: near-term capacity constraints (next 2-4 quarters) vs structural multi-year bottleneck — affects which evidence (lead times/capex guides vs long-term design wins) to prioritize.

2. Identify current bottlenecks (evidence, not memory)

The specific binding constraint (which component, whose capacity, what lead time) changes quarter to quarter and must come from current sourcing, not model memory. Read references/bottleneck-map.md for the category framework and illustrative candidate tiers — treat it strictly as a starting point for names to check, not as current fact.

For each in-scope category, est