The Coffee Grind by Provokative AI
The Selloff Deepens, and Tranche 11 Closes the Day With Three More Pairs
Final Edition, settled close · semiconductors extend Monday's decline through the full session · Tranche 11 takes its worst day yet, then adds three pairs at the close · the June 30 harvest of P1 and P2 keeps paying for itself
- Active book: 34 pairs · Realized register: +$487,933.25 (frozen) · Active unrealized: −$107,873.37 · ITD: $380,059.88
- SOXX −4.80% · GLW −12.10% (worst single name in the book) · DELL −8.15% · MU −8.85%
- Tranche 11 (eight pairs, P42–P49): −$5,499.86 on the day, −$12,908.92 unrealized since inception
- Three new pairs added at the close: P47 (GNRC/META), P48 (MU/NVDA), P49 (CAT/META), $100K/side each
- Still ahead tonight: Microsoft and Meta both report after-market — not yet known as of this edition's close
Monday asked whether the market could still verify the demand number underneath the artificial-intelligence trade. Tuesday's session answered with a second question: if the demand number is unverifiable, how much of the physical build-out around it — the glass, the memory, the foundry capacity — was priced on the same unverifiable basis. The tape spent the full day working through the second question, and by the close it had not resolved the first.
I. The Session, Settled
| Instrument | 7/27 Close | 7/28 Close | Change |
|---|---|---|---|
| SPY | 739.09 | 740.86 | +0.24% |
| QQQ | 682.12 | 675.49 | −0.97% |
| Semiconductors (SOXX) | 516.23 | 491.46 | −4.80% |
| Corning (GLW) | 143.36 | 126.01 | −12.10% |
| Micron (MU) | 900.20 | 820.53 | −8.85% |
| Intel (INTC) | 91.67 | 86.30 | −5.86% |
| Advanced Micro Devices (AMD) | 494.95 | 454.62 | −8.15% |
| Broadcom (AVGO) | 383.22 | 380.91 | −0.60% |
| Nvidia (NVDA) | 196.51 | 197.01 | +0.25% |
| Microsoft (MSFT) | 389.10 | 393.35 | +1.09% |
| Apple (AAPL) | 336.91 | 340.08 | +0.94% |
| Alphabet (GOOGL) | 326.56 | 333.71 | +2.19% |
| Tesla (TSLA) | 309.22 | 307.44 | −0.58% |
| Dell (DELL) | 426.91 | 392.10 | −8.15% |
| CBOE Volatility Index (VIX) | 18.67 | 18.21 | −2.46% |
All figures are 7/28 settled closes, two-source confirmed to the cent (yfinance batch + Yahoo Finance v8 chart API). Index futures, WTI crude, and gold have been dropped from this table as not directly relevant to the day's equity story; VIX's modest decline despite the semiconductor rout is itself notable — the damage was concentrated, not systemic.
II. Corning Cracks, and Tranche 11 Absorbs It
The semiconductor complex that fell two percent Monday closed down nearly five percent Tuesday. Micron, Intel, Advanced Micro Devices, and Dell all finished the session off six to nine percent. That is a continuation of Monday's story — the market still cannot verify the demand number the whole complex is priced against, and the second-day reaction produced a wider distribution of outcomes across the complex than Monday's did.
The name that stands out is Corning. GLW closed down 12.10%, worse than any semiconductor proper, on no company-specific news the desk has identified beyond the broader AI-infrastructure repricing. Corning is the anchor long leg of two active pairs, P21 and P42, and it traded the full session as an artificial-intelligence capital-expenditure derivative rather than as a specialty-glass manufacturer — precisely the pattern flagged in Monday's edition for the merchant-gas majors. A twelve-percent single-session move in a company that makes optical fiber and glass substrates is not a glass-market event.
Tranche 11 at the settled close: eight pairs, five opened at the July 24 or July 27 close and three added at today's close, closing out the tranche's roughest day since inception. Each ticker links to a TradingView candlestick chart; each pair links to a long/short ratio chart on the same platform.
| Pair | Structure | Since Inception | Day (vs 7/27) | Pair Chart |
|---|---|---|---|---|
| P42 | long GLW / short MSFT | −$17,128.78 | −$12,946.20 | ratio |
| P43 | long BLK / short MSFT | +$926.30 | +$2,250.45 | ratio |
| P44 | long GNRC / short AMZN | −$2,634.20 | −$731.87 | ratio |
| P45 | long GNRC / short NVDA | −$1,236.14 | −$1,236.14 | ratio |
| P46 | long GNRC / short DELL | +$7,163.90 | +$7,163.90 | ratio |
| P47 | long GNRC / short META | $0.00 | new | ratio |
| P48 | long MU / short NVDA | $0.00 | new | ratio |
| P49 | long CAT / short META | $0.00 | new | ratio |
| Tranche 11 Total (8 pairs) | −$12,908.92 | −$5,499.86 | ||
The tranche's design did most of what it was built to do: P46's DELL short and P43's MSFT-vs-BLK structure both worked, while the two GLW-anchored and NVDA-adjacent legs (P42, P45) absorbed the bulk of the damage. MSFT is short on 524 combined shares across P42 and P43 — net positive on the day for that leg specifically, since GLW's collapse dominated P42's loss far more than MSFT's own 1.09% gain did.
See Section IV for the rationale behind P47, P48, and P49, added at today's close.
| Ticker | RSI-14 | vs. SMA20 / SMA50 | Read |
|---|---|---|---|
| GLW | 30.7 | below both | approaching oversold, no reversal signal yet |
| GNRC | 27.4 | below both | most oversold of the four, pre-earnings |
| MU | 40.3 | below both | weak but not extreme; the P48 entry point |
| NVDA | 42.9 | below both | mildest weakness of the four |
On tranche- and portfolio-level charts: no single tradable instrument exists for "Tranche 11" or the OEF book as a whole, so neither can be linked to an external chart provider the way individual tickers and pair ratios can above. Producing either would mean generating an internal equity-curve chart from the book's own daily P&L history — which the desk does not yet persist as a time series; each session currently captures only point-in-time marks. Flagging this as a build item rather than fabricating a placeholder link.
Concentration flags, updated with today's additions: GNRC is now the long leg on four separate active pairs — P44 (vs. AMZN), P45 (vs. NVDA), P46 (vs. DELL), and P47 (vs. META) — 2,018 combined shares, four distinct theses, not netted. META is now short on two pairs, P47 and P49, 338 combined shares — both legs face the same after-market report tonight, which is concentration, not diversification. NVDA is now short on two pairs, P45 and P48, 1,017 combined shares.
III. The June 30 Harvest, Revisited Again
Monday's edition ran the exercise of pricing the ten June 30 close-outs as if they had never been harvested. Tuesday's settled close sharpens two of those comparisons considerably — both involving Corning and Nvidia, the two names that drove the day's damage. These figures also anchor the standing Harvest Shadow book maintained alongside the daily register.
P1 (GLW/MSFT, harvested June 30). Realized at the June 30 close: +$245,728.34 — GLW 1,246 shares exited at $255.43, MSFT 194 shares covered at $373.02. Priced at today's settled close (GLW $126.01, MSFT $393.35) as if the position had simply been held from original inception, the same structure would show +$80,527.00. The harvest banked $165,201.34 more than holding would show as of tonight's close — and that gap has widened every session since June 30, as GLW has fallen from the $255.43 exit to $126.01, more than halving.
P2 (GNRC/NVDA, harvested June 30). Realized at the June 30 close: +$66,542.97 — GNRC 603 shares exited at $292.81, NVDA 550 shares covered at $200.09. Priced at today's settled close (GNRC $195.60, NVDA $197.01), the same structure would show +$9,619.34 held. The harvest banked $56,923.63 more than holding would show — GNRC alone has fallen more than a third from its exit price.
Combined, the two harvests preserved $222,124.97 of value relative to the counterfactual of simply holding through tonight's close. That is not a coincidence of good luck on two names; it is what the June 30 harvest was designed to do — the publication's own house rule has been to bank the scarcity rent while it was unrealized rather than let a demand-sink thesis ride indefinitely. GLW's role as the long leg of P1 and the short-Mag-8-adjacent structure of P2 both routed through the same artificial-intelligence demand number Monday's edition identified as the excluded variable. Two sessions later, that variable is visibly reasserting itself, and the pairs that took the money off the table before it did are looking considerably better than the pairs — P42 among them — that carry closely related structures today. The full ten-pair Harvest Shadow book, updated alongside this edition, shows the same discipline paying off across the broader June 30 sweep, not just these two names.
IV. Tranche 11 Expands — Three New Pairs at the Close
At the close. Three new pairs were entered at today's settled close, all joining Tranche 11 and all sized at a flat $100,000 per side by design — the desk is expressing three related but distinct legs of the same capital-expenditure thesis and chose equal weighting over conviction-weighting so that the tranche's aggregate result reflects the thesis as a whole, not a bet on which single leg is strongest. Exact share counts, zero notional approximation. Tranche 11 now runs eight pairs, P42 through P49.
The strongest piece of evidence behind all three trades is five days old and did not involve any of these six names directly: Alphabet reported a quarter by any reasonable measure a blowout — 24% revenue growth, EPS far above consensus — and its stock fell roughly 7% anyway, purely because management raised 2026 capital-expenditure guidance to $195–205 billion and flagged free cash flow already negative. A sell-side analyst's comment following that print, that a similar guidance raise from Microsoft or Amazon would likely produce the same selling pressure, is the working thesis behind shorting META here: the market has just demonstrated it will punish capital-expenditure escalation even from a company beating every headline number.
P47 — long GNRC 511 sh @ $195.60 ($99,951.60) / short META 169 sh @ $593.41 ($100,286.29). Generac is scheduled to report its own second-quarter results tomorrow morning, before the open, with the Street looking for a data-center-driven beat on the back of a rapidly building commercial-and-industrial backlog. The rationale is straightforward capital-expenditure arithmetic: the same hyperscaler build-out under scrutiny on the demand side is unambiguously good for the physical suppliers standing behind it, and Generac's data-center backlog is a direct, verifiable claim on that spend rather than a promise about future model economics. What would prove this wrong: a soft print or a cautious guide from Generac tomorrow would undercut the "verifiable claim on spend" half of the thesis directly, independent of whatever META does tonight.
P48 — long MU 122 sh @ $820.53 ($100,104.66) / short NVDA 508 sh @ $197.01 ($100,081.08). This is the more contrarian of the three. Micron has fallen nearly 40% this month on a combination of a disclosed short position from investor Michael Burry (per public 13F/options disclosures), record insider selling, and renewed competitive anxiety following ChangXin Memory Technologies' market debut in China. None of that changes the structural argument for memory as an architecture-agnostic beneficiary of the capital-expenditure cycle: every hyperscaler accelerator design now in volume production or ramping — Nvidia's own roadmap included, alongside Google's TPU, Amazon's Trainium, and Microsoft's Maia — still requires high-bandwidth memory, while multiple sell-side estimates place Nvidia's own share of total accelerator spending on a declining path over the next two years as hyperscaler-designed silicon takes a growing slice of inference workloads specifically. The position is a deliberate value entry into a name still under active selling pressure; the desk is not asserting the selloff is finished. What would prove this wrong: confirmation that CXMT's output is displacing Micron's own HBM order book, rather than just pressuring sentiment, would break the thesis at its foundation.
P49 — long CAT 119 sh @ $840.85 ($100,061.15) / short META 169 sh @ $593.41 ($100,286.29). Caterpillar's Power Generation segment is one of the cleanest direct expressions of data-center buildout available in a name with an independent, diversified underlying business — segment revenue was up sharply again last quarter with a record enterprise backlog, and the company has both committed fresh capital to expand generator manufacturing capacity and landed a two-gigawatt natural-gas generator order tied to a named hyperscale compute campus. What would prove this wrong: Caterpillar has already fallen two consecutive sessions into this entry on its own, independent of tonight's META print — if that slide continues on Caterpillar-specific weakness (margin compression in its core construction segment has already been flagged this week) rather than reversing, the long leg fails regardless of what META does.
Correlation risk, entered with eyes open: Caterpillar's data-center-power thesis and META's capital-expenditure decisions are not cleanly offsetting, since both names are ultimately exposed to the same underlying question of whether the AI buildout continues at its current pace — a shared risk-off shock could hit both legs of P49 in the same direction rather than offsetting. The desk discussed this explicitly before sizing the position rather than treating the two legs as a clean hedge.
META reports after tonight's close with options pricing an implied move of roughly seven percent; Microsoft, short on P42 and P43, also reports tonight, with its own options market pricing a comparably elevated implied move. Neither result was known to the desk as of this edition's close.
The Synthesis — At the Close
Tuesday's settled numbers confirm the direction of Monday's thesis: the artificial-intelligence demand-verification problem was not confined to Nvidia, and it was not confined to one session. Corning's twelve-percent decline, in a company whose product is glass, was the clearest illustration yet that the market spent the day repricing an entire physical supply chain on the same unverifiable number — and the June 30 decision to harvest P1 and P2 while that number still looked solid paid for itself again today. The three pairs added to Tranche 11 at the close are a direct extension of the same logic in the opposite direction: if capital-expenditure escalation is what the market is punishing in the payer stocks, the physical suppliers standing behind that same spend are where the desk chose to lean.