Every AI-levered field trades above its own historical multiple band

Claim

Nine tech fields swept across three rounds in July 2026. Every AI-levered one showed the same shape: −15% to −60% off highs, but simultaneously +110% to +582% off lows. The 2024–26 melt-up was larger than the 2026 correction, so the correction returned prices to elevated, not to cheap.

The practical consequence: a screen built on drawdown will surface these names and they will all be false positives. In AI-adjacent sectors the drawdown filter has near-zero precision and must be replaced by an own-band filter.

Evidence

Field	Result
Semi test / ATE	11 of 11 above own band. TER at 76% above its own FY2025 high multiple
Optical / photonics	10 of 13 above own band. LITE at 2.6x its own recent-high P/B
Semi materials	Above own EV/EBITDA medians across the group
Robotics / machine vision	Quality names at 32–35x earnings
Japan AI	Zero discount. Advantest 60x P/E, 28x P/B

Fields where the pattern did not hold — and where the survivors came from — were the non-AI-levered ones: health tech (DOCS), industrial scanning (ZBRA), China internet (NTES).

What would falsify this

A further leg down that takes a major AI-levered field below its own 5-year median multiple without a corresponding collapse in the earnings base. That is the condition this note exists to detect — it converts the whole cohort from avoid to shopping list. recheck is set accordingly rather than left to drift.

Related

Mechanism: [[principle-down-a-lot-is-not-cheap]]. This note is that principle observed at field scale rather than name scale.

History

• 2026-07-29 — established across three sweep rounds covering nine fields.