CAPEX-IQ reads earnings-call language, filed backlogs, and customer concentration to find the narrowest link in the AI supply chain — then grades its own signals against the market.
“…we remain on allocation for indium-phosphide substrates through year end…”
Search-grounded guidance per hyperscaler — tracked and revised weekly, not a training-data guess. Ribbons fan out to six physical tracks, widths proportional to dollars. The guidance trend is the actual signal: the first derivative of hyperscaler capex.
NLP over every earnings call in the chain — “on allocation”, “sold out through”, “lead times extended” — quarters before it shows up in consensus. Each hit is classified: bottleneck owner, or merely input-constrained.
“…we remain on allocation for indium-phosphide substrates through year end…”
“…lead times on optical engines have extended to fifty-two weeks…”
The Composite Bottleneck Score blends everything the system knows about a name’s own bottleneck evidence into one 0–100 number — inputs stored per row, so every score is auditable.
A curated supplier → customer graph radiates each bottleneck downstream with criticality-weighted decay — an InP constraint at AXT flags Lumentum and Coherent as input-risk and traces all the way to the hyperscalers. Where a 10-K discloses customer concentration, the edge upgrades to the filed percentage.
A weekly AI scout reads bottleneck news per track — shortages, allocation, sole-source language — and proposes names not yet on the map. Every ticker is identity-verified and enriched with the same transcript + XBRL snapshot before a human approves it.
Enter your email to join the members list. Sign-in uses a one-time PIN sent to your inbox — no password to remember.