Why AI stocks are down — classify before you react

Bucket 1 — Macro/rates: higher real yields compress duration-sensitive multiples even if sales are fine.

Bucket 2 — Capex digestion: cloud buyers pause GPU/NIC spend; backlog stories reprice.

Bucket 3 — Supply/inventory: channel fill, HBM or GPU inventory, lead-time normalization.

Bucket 4 — Policy/export: licensing, bans, customer geography mix.

Bucket 5 — Thesis break: competition (ASIC), product miss, customer loss, accounting red flag.

Write which bucket you believe — and what evidence would move you to another bucket — before you trade.

SMH and QQQ can be red while a diversified platform (MSFT/GOOGL) is mostly “multiple.” NVDA can be red while memory (MU) has a different cycle clock. “AI stocks” in search language is a mashup — unmash it on paper.

Open the specific /stocks page. Ask debate seats. Demand the footnote that supports your bucket.

1) Re-read Mandate theme caps. 2) Always-on Scan for position breaches. 3) Ask red-team: “Argue we should not add.” 4) Update kill-conditions in Build. 5) Paper a small add only if rules allow — then decide at the broker.

Courage is not a strategy. Classification is.

Do not double size because a YouTube thumbnail said “crash.” Do not ignore taxes on short-term churn. Do not treat Agents61 paper P&L as a promise. Do not paste NVDA logic onto every semiconductor.

If you came from /ai-stocks-to-invest-in, tighten caps rather than expand the shopping list during panic.

Inconclusive is a valid desk output. Wait for the next 10-Q, capex guide, or export clarification. Idle cash with a written question beats a forced narrative.

Prediction Markets can sometimes frame event odds — still not a substitute for filings. See /predictions.