This report tracks the hyperscaler, chip and foundry basket — Amazon, Nvidia, ASML, TransAlta, TSMC, Palantir, Marvell, Arm, Everpure, and the two AI/semiconductor funds — not the separate physical-buildout universe that shares a similar name. Last cycle's headline was the tightest, most uniform band this basket has carried in the stretch, with every name between RSI 48 and 58; this week Arm broke through it entirely, surging 36.31% to RSI 71.1 — its largest move on record for this desk and now the basket's only overbought name — while Amazon's cushion kept thinning for a third straight cycle. No dated catalyst was found for Arm's move despite a full check of its news receipt and the scheduled-event record. Eight of eleven names still sit above their 200-day with a positive year.

LONG-HORIZON READ: a name whose short-window trend flag reads "down" but sits ABOVE its 200-day with a positive 1-year change is a deep pullback in an intact trend, never a breakdown. Only price below the 200-day paired with a weak or negative year is real structural damage.

This basket's defining feature last cycle was uniformity — every name inside a tight ten-point RSI band with nothing oversold and nothing overbought. That ended this week. Arm surged 36.31% in a single week, its RSI jumping from 57.7 to 71.1 and its 200-day cushion nearly doubling from +33.0% to +58.2%, with its trailing year climbing from +88.08% to +135.87%. That is the largest single-week move this desk has recorded for any name in this basket, and no dated item explains it: Arm's own news receipt carries only generic stock-quote aggregator pages, and no earnings print falls inside this window in the scheduled-event record. The same semiconductor complex carried Marvell (+13.58% on the week, RSI 63.1) and the semiconductor fund (RSI 61.4, +10.24%) higher without a specific dated cause either, though the broader tape this week did carry a sourced CPU-stock rally — CNBC reported Intel surging 12% on September 21 — that plausibly lifted sentiment across chip names generally even where this basket's own individual names lack a specific receipt.

Palantir posted the second-largest move, RSI +9.9 to 66.9 on a +7.97% week, extending the strength flagged last cycle; its trailing year is still only +5.06%, the weakest long-horizon profile among this basket's intact names despite three consecutive strong-to-flat cycles. Everpure (P) rose again, +5.29% on the week with its cushion widening to +42.6%, but its news receipt was last pulled September 1 — nothing on file speaks to why. TSMC and the AI-fund (AIQ) both firmed modestly and cleanly. Nvidia added another 1.46% with its cushion widening slightly to +13.3%.

Two names need their headline numbers read carefully in opposite directions. Amazon is now on its third straight cycle of a thinning cushion — 7.1% → 5.6% → 3.6% — while its RSI fell another 3.9 points to 44.1; it is still the basket's only name moving the wrong way on both the week and the long-horizon check, and if the cushion crosses zero next cycle it becomes the basket's first confirmed-damage name. TransAlta is the opposite kind of concern: its RSI cooled this week rather than firming (45.1, down 4.3 points) while its year deteriorated further to -12.26% and its 200-day gap widened to -8.0% — a name still deteriorating on both measures with a news pull that has now failed since April. The desk has no receipt-based evidence to explain either name's direction; both are tape-only reads for now.

1. The basket's uniform band broke: last cycle's spread was RSI 48-58 with nothing above or below; this cycle Arm alone jumped to RSI 71.1, an 18-point break from the top of that range.
2. ARM posted the basket's largest move on record for this desk: +36.31% on the week (price $275.61→$332.56), RSI +13.4 (57.7→71.1), 200-day cushion +33.0%→+58.2%, year +88.08%→+135.87% — no dated catalyst found.
3. PLTR posted the second-largest move: RSI +9.9 (57.0→66.9), +10.01% on the week, 200-day cushion +17.1%→+26.2%, though its year (+5.06%) is still the weakest intact profile in the basket.
4. MRVL extended hard: RSI +5.1 (58.0→63.1), +13.58% on the week, cushion +54.2%→+62.2%, year approaching +250%.
5. AMZN's cushion thinned for a third straight cycle: 7.1%→5.6%→3.6%, RSI -3.9 (48.0→44.1) — still the basket's only decliner and the name closest to a 200-day crossing if the trend continues.
6. TAC deteriorated on both RSI and year for the first time in two cycles: RSI -4.3 (49.4→45.1), year -9.28%→-12.26%, 200-day gap -6.1%→-8.0%. Its news pull remains failed since April.
7. TSM and AIQ both firmed cleanly: TSM RSI +3.7 (56.9→60.6), AIQ +6.2 (54.6→60.8).
8. ASML's headline year continued climbing on a rolling window: +81.00%→+81.07%, essentially flat, while price rose 3.89% on the week.
9. SMH firmed alongside the broader semis complex: RSI +7.1 (54.3→61.4), cushion +17.6%→+22.5%.
10. NVDA extended modestly: RSI +1.5 (54.2→55.7), cushion +12.0%→+13.3%.

RSI below 25 is capitulation. RSI 25-40 is oversold — check the trend, the death-cross flag and the long-horizon numbers before treating it as a buy. RSI 40-70 is neutral-to-healthy, above 70 is extended. "vs SMA" is distance from the 20-day average and is short-term positioning only. "1Y%" and "vs SMA200" are the long-horizon check that decides whether a trend is intact: a name below its 200-day with a weak or negative year is real weakness; a name above its 200-day with a strong year is a deep pullback even when its short-window reading looks soft. Everything here is raw screen evidence from settled prices. Where a cause is named it comes from a dated item in that name's news receipt or a dated capture; where the receipt is stale or failed the scan says so plainly.

None.

None. The basket's lowest reading is Amazon at 44.1, still comfortably above the oversold line.

Ten names inside a 44-to-61 band: PLTR (66.9), MRVL (63.1), SMH (61.4), AIQ (60.8), TSM (60.6), P (60.3), NVDA (55.7), ASML (55.0), TAC (45.1), AMZN (44.1).

One name: ARM at 71.1 — the basket's first overbought print in this stretch, up from a 44.0-to-70.1 spread two cycles ago and last cycle's tight 48-to-58 band.

• Whether Arm holds its overbought print or gives back a meaningful share of the 36.31% week. The stock is now 26.7% above its own 20-day average, a wide gap that leaves room for a sharp pullback without breaking the underlying trend.
• Whether a dated catalyst for Arm's move ever surfaces. Checked this cycle: its own news receipt (generic quote pages only) and the scheduled-event record (no earnings print in this window). If nothing surfaces next cycle either, this stands as an unexplained move inside an otherwise well-covered basket.
• Whether Amazon's three-cycle thinning cushion (7.1%→5.6%→3.6%) crosses zero next cycle, which would make it the basket's first confirmed-damage name.
• Whether TransAlta's divergence — RSI and year now both deteriorating together, rather than moving in opposite directions as in prior cycles — resolves, and whether its failed news pull (since April) is ever restored so the desk has evidence to weigh against the tape.
• Whether Palantir's trailing year, still just +5.06%, finally breaks out of its multi-cycle flat pattern now that its cushion has widened to +26.2%.
• Whether the basket's sudden dispersion — from a ten-point RSI band to a 27-point one in a single cycle — is the start of a broader re-rating or an isolated Arm-driven event that reverts next cycle.

• Tape data (price, RSI, moving averages, returns): full-scan pre-computed summaries as of the 2026-09-23 close — ai-infra.
• News receipts checked for cause attribution: stocks{ARM,NVDA,ASML,TAC,P,PLTR,MRVL}/news/latest.json`.
• Scheduled-event check for Arm and Marvell: research/stock-events/.
• Broader CPU-stock rally context: 2026-09-21-cnbc-intel-surges-12-as-cpu-stocks-rally-here-s-what-s-driving-th-108365666.
• Prior cycle for comparison: 2026-09-18-ai-infrastructure.