Weekly Macro Note — Week of March 23–27, 2026
March 27, 2026Week in Review: Distribution Into Quarter-End
The final week of Q1 2026 confirmed what the prior two weeks had suggested: the AI infrastructure complex is in a corrective regime with no near-term technical catalyst for reversal. QQQ lost 4.2% on the week (Monday $588.00 → Friday $563.27), closing at its lowest level since last fall and extending its distance below the 50-day SMA to 7.1%. SPY dropped 3.2% ($655.38 → $634.31, -6.7% vs SMA50). SMH shed 4.2% ($391.36 → $374.81, -6.4% vs SMA50). All three ended the week with multiple closes below SMA50, a bearish regime signal that did not resolve.
The week's price pattern was particularly informative: a brief midweek stabilization (Tuesday-Wednesday) gave way to renewed selling Thursday-Friday, eliminating any hope of a quarter-end recovery. Quarter-end rebalancing flows, which in prior quarters had supported prices, appear to have been insufficient or absent this cycle.
Key Regime Shift: Breakout System Clearing
The most operationally significant development this week was the sequential triggering of exit signals across multiple held positions:
- ANET (exit Friday): 50-day stop triggered after a 50-day hold. The stock had been recovering from a prior breakdown, and this exit signals the recovery attempt has failed.
- TSM (exit Friday): 50-day stop triggered after a 205-day hold — a long-duration position forced out by persistent index-level selling. TSM's breakdown has broader implications as the world's leading foundry.
- DDOG (exit Friday): 10% hard stop hit after a 14-day hold. The fastest exit of the three, reflecting the severity of software selling pressure.
Three exit triggers in a single day, two of which (ANET, TSM) are bellwether AI infrastructure names, is a regime-level signal. The system is correctly de-risking into deteriorating conditions.
Layer Rotation: Memory Holds, Software Breaks
The week's clearest trend was the widening RS divergence between layers:
| Layer | Avg RS (Mon Mar 24) | Change Direction |
|---|---|---|
| Memory & Storage | +116.2 → ~95.4 | Leading (some compression but still dominant) |
| Networking | +90.2 → ~74.3 | Second, held well |
| Semiconductor Equipment | +68.2 → ~55.2 | Third, some fade |
| Data Centers | +35.4 | Moderate |
| Energy Infrastructure | +30.9 | Moderate |
| Foundries | +29.1 | Weakened (TSM exit confirms) |
| Processors | +8.4 | Low |
| Software & Models | -4.5 → -4.7 | Only negative layer |
SNDK hit the 97th percentile in relative strength (Mansfield RS: 206.9) — an extreme reading that signals memory-specific strength relative to the broad market. LITE (94th, RS 197.3) and CIEN (91st, RS 144.2) are similarly elevated. This group is being bid even as the index sells off, which is a notable divergence.
The social intelligence layer reinforced this theme: HBM demand commentary around latency bottlenecks and per-GPU memory requirements circulated repeatedly across tracked accounts, pointing to a structural rather than cyclical demand driver for SNDK, MU, and related names.
Macro Themes That Defined the Week
-
Quarter-end without relief: The usual window-dressing and rebalancing bids did not materialize in sufficient size to halt the decline, suggesting institutional selling pressure is real and ongoing.
-
Software capitulation accelerating: DDOG, PLTR, and ARM all took outsized losses this week. Software names are at the tail end of a multi-week de-rating that started when macro uncertainty (tariffs, Fed path uncertainty) began weighing on growth multiples.
-
Memory as a flight to quality within semis: Counterintuitively, memory — normally a more cyclical semiconductor category — is acting as a relative safe haven within AI infrastructure. The HBM/AI training narrative is providing a demand floor that pure-play software does not have.
-
No breakout signals generated this week: Zero new breakout setups fired, confirming the scanner is correctly reading a low-quality environment for new entries.
Forward Look: What Matters Next Week
- Q1 earnings season begins mid-April: No major AI infrastructure earnings in the next 5-10 days, but pre-earnings positioning will begin. Watch for guide-down risk at software names.
- QQQ $550 level: A close below $550 would represent a 9.3% breakdown below SMA50 — territory that has historically required macro-level policy intervention (Fed pivot language, trade truce) to reverse.
- Memory RS durability: Can SNDK, LITE, CIEN, and WDC maintain outperformance if QQQ accelerates lower? A correlation spike (everything selling together) is the key risk to the current rotation thesis.
- April 2 tariff deadline: Residual risk from potential tariff escalation. Markets appear to have partly priced this, but a surprise announcement could catalyze another leg down.