Data note first: this report currently reads the original 9-name hyperscaler/chip/foundry basket (ai-infra.json — AMZN, NVDA, ASML, TAC, TSM, PLTR, MRVL, ARM, P), not the 52-name physical-buildout universe (power/nuclear, cooling/electrical, networking/optical, materials, data-center REITs, compute/memory) that the 2026-07-11 edition of this scan used. There is no valid ticker-for-ticker comparison against that 52-name report; the comparisons below instead use the same 8 overlapping names as read in the 2026-07-11 ai-scan report. The clearest move: MRVL and ARM both cratered -39% over 30 days, while AMZN stayed the steadiest name in the basket and PLTR posted the only clean positive week.

FROTH READ: vs VWAP measures distance from the average buyer's cost basis across the instrument's full history.

The two clearest breakdowns in this basket are MRVL and ARM. MRVL fell -19.99% on the week, pushing its 30-day change from -15.97% to -39.23% and its RSI from 45.2 to 35.7 — into the oversold zone. ARM moved almost identically: -17.38% on the week, 30-day change from -5.51% to -39.2%, RSI down to 38.4. TSM also cooled into the oversold zone (RSI 49.8→39.4, 30D flipping from +3.1% to -13.8%), and TAC weakened sharply (RSI 58.6→42.7, 7D flipping from +2.67% to -6.75%).

AMZN and NVDA held up better by comparison. AMZN's RSI is essentially unchanged (51.0→51.4) and its thesis flag still reads "intact," with a 52-week gap that actually narrowed slightly (-11.93%→-11.25%). NVDA cooled (RSI 57.2→48.1, 7D flipping from +7.32% to -3.86%) but remains thesis-flagged intact and only modestly further from its 52-week high. ASML cooled slightly (RSI 51.1→47.9) while its 30-day change worsened from -5.38% to -9.44%.

PLTR was the one name in this basket to post a genuine reversal — RSI ticked up from 48.4 to 52.9, and both its 7-day (-3.08%→+4.41%) and 30-day (-3.27%→+3.04%) changes flipped positive, even though it remains death-crossed and -36.21% off its 52-week high. P (Everpure, formerly Pure Storage) has no comparable prior read in this cycle's data; its current numbers — RSI 42.8, -12.54% on the week, -31.03% off its 52-week high, in a confirmed downtrend — establish the baseline going forward.

No valid comparison exists against the 2026-07-11 ai-infrastructure report — that report's 52-name universe (power/nuclear, cooling/electrical, networking/optical, materials, data-center REITs, compute/memory) has zero ticker overlap with this cycle's 9-name basket. The deltas below instead compare against the same 8 tickers as read in the 2026-07-11 ai-scan report (P is not in that report either, and has no baseline this cycle).

1. MRVL cratered: RSI 45.2→35.7, 30D -15.97%→-39.23%, 52wkHi gap -28.52%→-42.8%.
2. ARM cratered in tandem: RSI 47.9→38.4, 30D -5.51%→-39.2%, 52wkHi gap -28.56%→-40.98%.
3. TSM cooled into oversold: RSI 49.8→39.4, 30D +3.1%→-13.8%, 52wkHi gap -9.37%→-16.83%.
4. TAC weakened sharply: RSI 58.6→42.7, 7D +2.67%→-6.75%, 52wkHi gap -20.47%→-25.84%.
5. NVDA cooled from a strong week to a weak one: RSI 57.2→48.1, 7D +7.32%→-3.86%, 30D +2.97%→-3.74%.
6. ASML cooled modestly: RSI 51.1→47.9, 30D -5.38%→-9.44%.
7. AMZN stayed the steadiest: RSI 51.0→51.4, 30D +1.59%→1.16%, 52wkHi gap narrowed -11.93%→-11.25%.
8. PLTR reversed positive: RSI 48.4→52.9, 7D -3.08%→+4.41%, 30D -3.27%→+3.04% — still death-crossed, but the only clean improvement in the basket.
9. P has no prior read this cycle — first appearance in this data window.

RSI < 30 = capitulation zone (check trend/death-cross before treating as a buy). RSI 30-40 = oversold, often still structurally weak in this group. RSI 40-70 = neutral-to-healthy. RSI > 70 = extended. "vs VWAP" measures distance from the average buyer's cost basis — the further above 0%, the more "paid up" the current price relative to its own trading history.

None.

MRVL (35.7), ARM (38.4), TSM (39.4). All three in confirmed weak-down/pullback regimes with real 30-day damage — not constructive dip-buys on the tape alone.

AMZN (51.4), NVDA (48.1), ASML (47.9), PLTR (52.9), P (42.8), TAC (42.7).

• Whether the ai-infrastructure scan slot reverts to the 52-name physical-buildout universe next cycle, or stays on this 9-name basket — worth confirming with the watchlist owner, since the two scans currently answer different questions.
• Whether MRVL/ARM's -39% 30-day drawdowns are a bottoming process or still unfolding (also flagged in the parallel ai-scan report).
• Whether TSM/TAC's slide into oversold territory finds a floor or keeps deteriorating.
• Whether PLTR's reversal has legs given it remains death-crossed and well below VWAP.

• Tape data (price/RSI/SMA/returns): full-scan pre-computed summaries as of the 2026-07-17 close — *.