Playbook › Market
Market regime — the 2026 two-act rout
Claim
The single most important context for reading any 2026 report in this repo: the year contained two routs, in opposite sectors, in opposite directions, and a report's date determines which one it is talking about. A March screen and a July screen using the same words ("tech is cheap") mean incompatible things.
Act I — the SaaSpocalypse (Feb–Apr 2026)
Trigger: AI-displacement fear following Claude Code's early-2026 release. The market re-rated the entire B2B software complex on the thesis that AI commoditizes software, collapsing switching costs and lowering the barrier to entry.
| Marker | Value | As-of |
|---|---|---|
| Software market cap erased | ~$1T | 2026-03-18 |
| IGV drawdown | −20%+ YTD; worst stretch since 2008 | 2026-03-27 |
| Avg public tech name off 52wk high | −41.5% | 2026-03-27 |
| SaaS forward P/E | ~22x — a discount to the S&P 500 for the first time in the modern era | 2026-03-27 |
Sell-side dissent was loud and early: JPMorgan called the selloff "broken logic"; Deutsche Bank declared "peak fear" on March 10. The distinguishing feature was indiscriminacy — the AI-displacement multiple was applied uniformly, without regard to whether a given moat was actually exposed. That indiscriminacy is what made it a screening opportunity rather than just a decline.
Act II — the AI-capex rout (Jun–Jul 2026)
The mirror image. Capital that fled software had crowded into AI infrastructure; in late June that trade broke.
| Marker | Value | As-of |
|---|---|---|
| SOX peak close | 14,634.72 | 2026-06-22 |
| SOX close | 10,447.49 | 2026-07-29 |
| SOX drawdown | −28.6% | 2026-07-29 |
| Final-week acceleration | −4.9% (7/27), −4.5% (7/28), −5.3% (7/29) | 2026-07-29 |
Critically, Act II did not create value the way Act I did. Act I took names below their own historical bands. Act II only took them back to elevated — the preceding run-up was larger than the drawdown. Every AI-levered field swept in July still traded above its own band. See [[pattern-ai-levered-fields-trade-above-own-band]].
Act II reversed — Jul 30 to Aug 4 2026
Act II lasted five weeks and unwound in four sessions.
| Marker | 2026-07-29 | 2026-08-04 |
|---|---|---|
| SOX | 10,447.49 | 12,166.22 (+16.5%) |
| SOX off June high | −28.6% | −17.0% |
| VIX | 20.66 | 16.19 (−21.6%) |
| QQQ | $682.12 | $720.53 (+5.6%) |
| SCHD | $33.43 | $33.77 (+1.0%) |
| S&P 500 | −1.52% on the day | all-time high, 7,600.50 |
Three drivers, none of them monetary: (1) broad earnings — all 11 S&P sectors showing YoY growth, three triple-digit, with AMZN +15.3% on a $200.6B print; (2) geopolitical relief as a US–Iran strike was called off to negotiate the Strait of Hormuz, dropping oil and the risk premium together; (3) the AI-capex fear simply unwinding.
This is the confirming evidence for the Act II diagnosis. The July 29 AMAT analysis called the −39% drawdown "~100% multiple de-rating, 0% earnings revision." A de-rating with no earnings revision behind it can re-rate on sentiment alone — and did, at a speed no fundamental process can trade against. That is the practical lesson, not the direction: Act II drawdowns were never priced-in value, so they were never durable entries either.
Rates are the live risk, and they did not participate
The rally is not a rate-cut rally. The 7/29 FOMC held 9–3 with all three dissents hawkish, Chair Warsh gave no forward guidance, the 10yr sits ~4.70% just off an 18-month high, and the market now prices ~63% odds of a 25bp HIKE in September.
The rate-duration cluster confirms it by omission — VICI +1.2%, O −4.2%, ARCC +1.4%, NLCP −0.2%, MAIN +2.6% against a +5.6% QQQ. Every rate-sensitive name sat out the best four days of the year. Anything underwritten on a 2026 cut needs re-checking against a hike as the base case.
What this means for reading old reports
- Pre-May software screens — the drawdown was real and the multiples genuinely compressed. Names are stale but the mechanism was sound. Worth re-checking prices.
- June–July AI/semi screens — a large drawdown here is not evidence of value, and any report leaning on "% off 52-week high" as its value case should be discounted.
- The two acts are related, not independent. The same capital that left software arrived in AI infrastructure and then left that too. Treating them as separate events misses that this was one rotation with two victims.
What would falsify this
A third leg that hits both simultaneously — a broad multiple compression driven by rates or growth rather than sector narrative — would make the two-act framing obsolete and require reading the year as one continuous de-rating instead.
🚩 Open question
Act I's resolution is untested. The March thesis was "AI-displacement fear is indiscriminate, so moated names are mispriced." Nobody has gone back to check what the March shortlists actually did between March and July. That retrospective is the single highest-value unrun analysis in this repo — it would score the framework itself, not just a name. See [[Themes/software-saas]].
History
- 2026-03-18 — Act I first recorded across the March screen series.
- 2026-07-29 — Act II recorded across the three July sweep rounds.
- 2026-07-30 — unified into one regime note; the two-act structure and its consequence for reading dated reports made explicit.
- 2026-08-04 — Act II reversed in four sessions. Added the reversal section and the rates finding (Sept prices a hike, not a cut). The claim itself is unchanged and was confirmed: Act II was a sentiment de-rating, which is why it undid itself so fast. Execution consequence recorded separately in [[pattern-shortlist-runs-when-unexecuted]].