Mag Pulse · daily board · the Six plus the Torque · as of Sep 17, 2026
Mag Pulse
One question, answered daily: is the hyperscaler cohort turning? The Mag complex spent ~19 months going sideways on capex fear while nearly every benchmark ran. This week's cloud prints put booked demand against that fear — this board keeps the score. Six names in three cohorts, one torque row, and the popular instrument (MAGS) as the shared price frame. Tesla is excluded by methodology.
The regime-turn line Hyperscalers ÷ Consumer-AI · equal-weight · indexed 1y
Rising = the market pays the build-out cohort over the consumer-AI cohort. The line is the board's headline read: a durable turn here is what "the discount is closing" looks like in price.
The MAGS map popular-instrument foil · standing levels
Ten-month compression: flat top at the 69.50 supply shelf (two rejections), rising floor on the multi-year trendline (~56, hand-set monthly), pivot 62.56 = the Dec-2024 top on price. Levels first, story second.
The cohorts
MSFT · AMZN · GOOGL — the thesis longs: does the cohort close the gap?
META · AAPL — not yet in the build-out cycle; edge-AI optionality
NVDA — the anchor between build-out and deployment
ORCL — highest-beta hyperscaler pivot; thesis amplifier both ways
The Six absolute + relative, Sep 17, 2026 close
| Name | Cohort | Close | 1D | 1W | 1M | 1Y | 1Y vs QQQ | 1Y vs MAGS | Record |
|---|---|---|---|---|---|---|---|---|---|
| MSFT | Hyperscalers | 497.75 | +1.52% | +1.08% | +3.54% | -1.81% | -23.66% | -13.30% | print record |
| AMZN | Hyperscalers | 251.19 | +2.13% | -0.28% | -3.18% | +7.32% | -14.53% | -4.17% | print record |
| GOOGL | Hyperscalers | 347.33 | +1.30% | +4.43% | +0.91% | +38.48% | +16.63% | +26.99% | print record |
| META | Consumer-AI | 682.31 | +1.34% | +5.89% | +25.50% | -12.21% | -34.06% | -23.70% | print record |
| AAPL | Consumer-AI | 337.00 | +1.38% | +3.19% | +8.70% | +41.78% | +19.93% | +30.29% | print record |
| NVDA | Supplier | 219.34 | +2.54% | +0.45% | -0.18% | +25.58% | +3.73% | +14.09% | print record |
The Torque · ORCL separated by design — the thesis amplifier, both ways
The torque: hyperscaler-since-recently, declined the most, most volatile in both directions. If the thesis holds, the biggest move lives here.
Benchmarks the field the Six are measured against
| Instrument | Close | 1D | 1W | 1M | 1Y |
|---|---|---|---|---|---|
| QQQ | 716.92 | +1.73% | +1.16% | -0.08% | +21.85% |
| SPY | 762.60 | +1.13% | +0.63% | -0.63% | +16.83% |
| SOXX | 519.10 | +3.39% | +0.39% | -2.25% | +101.40% |
| XLK | 188.06 | +2.25% | +1.53% | +1.31% | +38.82% |
| IGV | 105.78 | +0.78% | +4.53% | +3.75% | -6.14% |
| MAGS · the foil | 70.78 | +1.81% | +2.31% | +6.05% | +11.49% |
The thesis layer the standing read per name — editorial, updated on evidence
The capex-fear discount, refuted where it was priced deepest: Azure +43%, $678B backlog — the market paid the print +15% in a day.
AWS re-accelerated to +37% — the demand witness. The biggest print reaction in its record parked it directly under the all-time-high line.
The straddle: cloud yes, search disruption already visible — and AI replacement monetizes less well. Re-rated in sympathy without printing.
Excellent at AI-for-ads — but that is not genAI. Charged hardest of the Six for strategy-push capex without a booked-demand anchor.
Waiting to monetize reach, with the tail risk that a new AI interface takes reach away from mobile. The cost repricing hit its margin story first.
The supplier both cohorts pay either way — priced on workloads, not orders; the boundary name between build-out and deployment.
Methodology
Universe: six Mag names in three cohorts — Hyperscalers (MSFT, AMZN, GOOGL), Consumer-AI (META, AAPL), Supplier (NVDA) — plus Oracle as a visually separated torque row. Tesla is excluded by stated methodology: not a Mag company for this monitor — its price is set by autonomy and robotics narratives, not by the cloud/AI build-out economics this board measures.
Computation: cohort composites are equal-weight within cohort. All windows (1D/1W/1M/1Y) are trading-day windows on adjusted closes. Relatives are simple return differences against each benchmark (QQQ, SPY, SOXX, XLK, IGV) plus MAGS as the popular-instrument foil. The headline line is the hyperscaler composite divided by the consumer-AI composite, indexed over one year — the regime-turn detector.
The MAGS map is a standing frame, not a computed one: the 69.50 supply shelf and 62.56 pivot (the Dec-17-2024 top on price basis) are fixed levels; the multi-year trendline (~56) is hand-set and updated monthly.
Data: equities from the Closelook data lake; benchmark ETFs from EODHD. Refreshed nightly after the US close (22:30 UTC, Mon–Fri).
This board is a diary entry on our process. Closelook publishes research, not investment advice.