Playbook › Playbook
In a tech selloff, value is deepest in application software and shallowest in hardware
Claim
Four independent March 2026 screens produced the same layer ordering without being designed to test it:
| Layer | Best multiples found | Assessment at the time |
|---|---|---|
| Application software | ADBE 9.3x fwd, WDAY 10.7x | deepest value |
| Cybersecurity / infra software | CHKP 13.4x, FFIV 17.3x | strong, a tier below |
| Hardware / IT infrastructure | NTAP 12.1x, CDW 10.6x, ZBRA 10.2x | "comparatively weaker — moats narrower, FCF profiles less compelling" |
Hardware multiples were not higher — the discount was similar. The gap was in moat durability and FCF quality, which is why the ordering is about risk-adjusted value rather than headline cheapness.
Mechanism (proposed, not established)
The selloff's trigger was AI-displacement fear aimed squarely at software. Fear concentrated where the narrative pointed, so the mispricing concentrated there too. Hardware fell in sympathy — less fear, less overshoot, less opportunity.
If that mechanism is right, the ordering is specific to narrative-driven selloffs and should invert in a demand-driven one, where hardware takes the earnings hit and software's recurring revenue cushions it. Untested.
⚠️ Confidence is medium and the reason matters
This is four observations inside a two-week window in one selloff. It is a pattern, not
a law, and it has never been scored against outcomes — nobody checked what those
names did afterward. Marked fast deliberately: it should not be relied on again without
that retrospective.
What would falsify this
A hardware-led selloff producing better risk-adjusted value in hardware — or, more cheaply, the March-cohort retrospective showing the application-software picks underperformed the hardware ones.
Related
[[regime]] Act I is the context. [[principle-down-a-lot-is-not-cheap]] is the filter that should be applied on top of this ordering, not replaced by it.
History
- 2026-03-28 — ordering observed across four screens.
- 2026-07-30 — recorded, with confidence explicitly capped pending the retrospective.