US Pre-Market Brief | 2026-09-16 (ET) Wednesday

Coverage window: 2026-09-15 16:00 ET regular-session close → 2026-09-16 08:15 ET. Includes Tuesday's after-hours, the overnight Asia/Europe sessions, and this morning's pre-market. Methodology notes:

• Single-stock pre-market prices and pre-market volumes come from the Nasdaq official quote API, benchmarked to the 09-15 16:00 ET close, with pre-market price − net change = previous close reverse-computed and verified name by name; key names were then independently cross-checked once more against the CNBC quote API, and the two sources agree.
• Indices/futures/commodities come from the CNBC quote API; yesterday's Treasury yield closes were checked against the official US Treasury curve (daily_treasury_yield_curve, 15:00 ET fixing): 09-15 official 2Y 4.67 / 10Y 5.00 / 30Y 5.36, consistent with the vendor's previous close.
• Pre-market quotes are cumulative, not point-in-time facts. This piece gives two read moments: the main table is 07:35–08:06 ET, with a re-read of key names at 08:14–08:16 ET.
• Sector/theme baskets are computed by me, name by name, equal-weighted, not third-party sector indicators; each basket lists its constituents and its count. The sample is the 575 names whose pre-market quotes passed validation out of 738 large caps (18 names with isRealTime=false indicative fake quotes were removed).
• ✅ August retail sales (08:30 ET) have been released, and this piece already includes the print and the market's reaction to it (supplementary reads 08:32–08:34 ET, see §0 item 6). The main table is still the pre-release read; wherever a re-read was taken after the release, both numbers are marked in the text.
• 🔴 Next time gates: 10:30 ET EIA crude inventories, 14:00 ET FOMC, 14:30 ET press conference, 16:45 ET Lennar earnings — all four come after this piece.

---

0. Today in one line

1. There is only one real protagonist in this pre-market, and it does not come on stage until two in the afternoon. The FOMC decision at 14:00 ET and the press conference by Powell's successor Kevin Warsh at 14:30 ET, with the market pricing roughly 92%–95% odds of a 25bp hike to 3.75%–4.00% — this would be the first hike since 2023. The four and a half hours between the open and 14:00 are essentially position adjustment, not directional trading.

2. But the structure of this pre-market table is nothing like yesterday's, and that must be said clearly — because yesterday's piece got exactly this wrong. Yesterday's pre-market core structure was "mean reversion of Monday's moves" (corr −0.55), and it turned out to have fully played out 100% before the 09:30 open, with the direction reversing after the open (open→close corr +0.20). Today I re-ran the same method: corr(9/15 daily return, this morning's pre-market return) = −0.184 (n=573), and it does not strengthen but weakens once liquidity filters are applied (notional ≥$1 million takes it to −0.131, ≥$10 million to −0.119). Sorting into quintiles by 9/15 return, this morning's pre-market means are +0.75% / +0.49% / +0.43% / +0.45% / +0.32% — all five groups are nearly equally strong, with a maximum spread between groups of only 0.43 percentage points. ⚠️ The wording has to be precise; it cannot be written as "no relationship": corr=−0.184 at n=573 gives t=−4.47 (p<1e-5), which is statistically significant, and the quintiles do lean monotonically toward "low first, high after." But R² is only 3.4% — it is real, yet it explains almost none of the dispersion. Correct statement: 1-day mean reversion still exists today, but its strength is only one-third of yesterday's (−0.55), and its economic magnitude is too small to serve as a ranker. Today is a broad-advance day, not a reversal day.

3. The real dispersion is not in "yesterday's move" but in "which theme line a name belongs to." Of the 575 names, 425 are up and 148 are down, but the baskets are torn apart (all equal-weighted and computed by me; constituents in Section 2): Semiconductor equipment +2.09% (6/6 all up), neocloud compute +2.23% (8/8), gold and silver miners +2.08% (15/15), AI hardware/optical modules +1.84% (17 of 18 up) — versus — energy E&P −1.82% (8/8 all down), pure-play cybersecurity −0.38%, software SaaS −0.13%.

4. The single strongest catalyst is Intel, and on volume it is not "one of the leaders" — it is first by a wide gap. Reuters reported on 09-16 that SK Hynix is in talks with Intel to produce memory chips in the US, with possible structures including leasing Intel's repeatedly delayed Ohio fab, or a joint venture with Intel and cloud providers. INTC pre-market $99.99, +2.93% (08:04 ET), re-read $100.35, +3.30% (08:16 ET); pre-market notional value was $577 million, the highest of the 575 names I scanned, with MU second at $453 million. 🔴 Before filing, a development surfaced that weakens this catalyst, and it has to go right at the front: SK Hynix publicly played the report down the same day. The company told Korean media that "the scenario described in the report runs far ahead of where we are; we are still at the evaluation stage," and declined to confirm whether it is in talks with Intel, acknowledging only that it "intends to add overseas plants and is evaluating multiple countries and sites"; another Korean outlet's headline flatly states it denied acquiring the Ohio plant ("We have no plans for an acquisition"). Published 09-16 16:49 KST = 03:49 ET, ahead of the US pre-market. Aju Press¹ So this is not merely "neither side has confirmed"; it is "one of the parties has actively cooled it down." What is being priced is option value, not cash flow. 📌 The Seoul single-stock read has been obtained, and it is far more persuasive than the index: SK Hynix (000660.KS) closed in Seoul +2.7% at ₩1,735,000, having been up nearly 4% intraday before giving it back once the company played the story down. Against INTC pre-market +3.64% — Intel is stronger than SK Hynix, consistent with the reading that "Intel selling idle assets is the beneficiary, SK Hynix is only the buyer." ⚠️ At the same time, correcting an error in my own argument: KOSPI (+1.37%) and EWY (+1.78%) are not two independent confirmations — EWY is mechanically a function of KOSPI, and SK Hynix is itself a KOSPI heavyweight, so the index's strength may simply be the stock itself. The only genuinely independent external read is the Seoul single-stock line. ⚠️ One more methodology correction: overnight media (the Saxo morning note) wrote that the Nikkei was "roughly flat, −0.07%" and KOSPI "+0.10%" — those were reads taken while the Asian session was still running, not closes. On a close basis, the Nikkei was +0.69% and KOSPI +1.37%. This piece uses closing prices throughout; read moment 08:20 ET.

5. Two lines that "look macro but are not" must be separated out:

   • This morning's AI hardware move is not rate-driven. The 10Y has come back from 5.00% to 4.977% — it is indeed falling. But software SaaS, equally long-duration, is down this morning: SMH +1.32% vs IGV −0.32%, a 1.64pp scissor gap. If this really were "duration repair," software — which should rally most — would not be at the bottom. This is a repair of the AI capex narrative itself, unrelated to rates.
   • Energy's decline today is not "geopolitical de-escalation." The Saudi East-West pipeline is still shut, over 4 million barrels per day of export capacity is at risk, and Saudi Arabia has also cancelled some September deliveries to Europe. The reason for the fall is inventories: API reported a US crude build of 7.14 million barrels for the week of 9/11, against a market expectation of a draw of 1.6 million barrels. The official EIA data only comes at 10:30 ET — that is today's second time gate, and it falls before the FOMC.

6. The 08:30 retail sales print flipped from negative to positive month over month, and the bond market barely moved — but I am actively downgrading this observation, because I did not get a firm grip on either end of it. August retail sales came in at +1.2% MoM (July revised up from −0.5%), ex-autos +1.4%, +6.0% YoY. ⚠️ Downgrade reason one: I did not obtain a consensus estimate. Census does not publish a consensus itself, and I was unable to obtain a time-stamped median from the Reuters / Dow Jones / Bloomberg surveys. So this piece only says "flipped from negative to positive MoM, absolute level not weak" — it does not say "beat." Without a consensus there is no expectation gap; that is discipline. Also: +1.2% comes directly after July's −0.5%, so it contains a low-base rebound; and retail sales are a nominal series, so the +6.0% YoY is not price-adjusted. ⚠️ Downgrade reason two: that "zero reaction" in the 2-year may be a bad read, not market behavior. I read CNBC's US2Y three times, at 08:15 / 08:33 / 08:56, and all three came back as exactly the same 4.636%, identical to the third decimal place. For the most liquid market in the world to be motionless for 41 minutes, the prior probability of "field caching/lag" is no lower than that of "the market truly did not react" — especially since the 2Y net-change field on that same API is already broken today (see the methodology note below). This piece therefore does not treat "the bond market did not react" as a conclusion, only as an observation pending verification. The actual reads (listed as taken, with the caveats above):

   	08:15 ET (pre-release)	08:33 ET (post-release)	Change
   2-year Treasury	4.636%	4.636%	0.0bp
   10-year Treasury	4.977%	4.979%	+0.2bp
   Nasdaq futures	+0.59%	+0.63%	+0.04pp
   VIX	16.77	16.82	+0.05
   INTC pre-market	+3.30%	+3.64%	Still strengthening

   Table scrolls horizontally

   A +1.2% retail sales beat, and the 2-year Treasury did not budge. That says the market has handed the entire pricing weight of today over to 14:00, and treats the morning's macro data as noise. Put the other way: every price this morning is a "price while waiting," and has no property of being inherited.

Pre-market board at a glance (reads 08:05–08:23 ET)

⚠️ Those three Treasury rows I computed myself as "current level − official Treasury close," without using the vendor's net-change field. The reason is that CNBC's 2-year change_pct field is broken again today: it returns +0.0547%, while within the same record last (4.636) is plainly below prev (4.663). Taking that field at face value would turn "front end down, curve bull steepening" into "front end up, curve bear flattening," and the entire rates narrative would come out backwards.

---

1. News overview

⚠️ On "today's analyst ratings" I ran a falsification test, and the conclusion is not to include them. The "Wednesday's Top Wall Street Analyst Research Calls" item returned by search looks same-day, but checking the URL shows the actual publication date is 09-02 (this recurring column's headline is word-for-word identical week after week). Apart from the two JBHT price-target cuts (separately verified), this piece carries no "today's rating changes" category at all — because I cannot confirm the publication dates.

---

2. Strongest themes, descending

---

3. Overall single-stock strength table (descending by stock; split into positive and negative zones)

Reads: the main table is the Nasdaq official API at 07:35–08:06 ET; the "re-read" column is a second read at 08:14–08:16 ET. Maximum drift between the two reads was 0.51pp (TCOM); the structure is stable. "Distance from 52-week high" is benchmarked to the 09-15 close, not the pre-market price (a pre-market price is not a confirmed transaction price). Notional value = pre-market volume × pre-market price. ⚠️ Share count cannot be used as liquidity: LEN's 4,013 pre-market shares look like a decent number, but the notional is only $320,000 and constitutes no signal.

3A. Positive zone

Scoring methodology: 100 points total; component definitions are in §4. Scores have been aligned with the fundamental checks in §5 — the draft ranking was changed in three places after those checks: COHR was cut from 4th to watch only (cash flow), TCOM was raised from "TBD" to 3rd (confirmed as a genuine earnings report), and BE was moved from 17th in the positive zone to avoid (all three "today's catalysts" turned out to be old news).

3B. Negative zone

---

4. Single-stock scoring model (100 points)

Using INTC as the worked example of the component breakdown (all other names are scored the same way; scores are listed in the "Total score" column of §3A):

---

5. Detailed analysis of top names

The ①–⑪ in this section are deep dives organized by theme grouping, not by the score ranking in §3A (§3A is the descending master list). Group order: single-stock catalyst (①) → the three AI hardware names (②③④) → semiconductor equipment (⑤–⑦) → precious metals (⑧) → the two "huge volume but questionable logic" names (⑨⑩) → China ADR earnings (⑪).

① INTC — Intel | priority deep-dive | total score 64

Related news: Reuters exclusive, 09-16 at about 01:31 ET (reporters Heekyong Yang / Alexandra Alper / Hyunjoo Jin, three people familiar with the matter). Two conceived structures: ① SK Hynix leases part of Intel's Ohio fab; ② a joint venture with Intel and a large cloud provider. The deal structure is undecided — even which type of memory chip would be produced is undecided. SK Hynix responded that it is "evaluating multiple options" and that "nothing has been decided at this stage"; Intel declined to comment, calling the reports speculation. CNBC³ · Korea Herald⁴

🔴 The single most important verification result in this piece: this is not the first disclosure; it is round two. The Korea JoongAng Daily already ran an exclusive on the same thread on 2026-07-22 — in that version SK Hynix was in talks to acquire the entire Ohio campus, targeting "front-end memory production starting within five years"; SK's response at the time was that it had "neither advanced nor decided" anything. Comparing the two versions: the structure has retreated from "acquisition" to "lease/JV" (the scale is smaller, not larger). And after that July round, INTC fell from $139.63 to $90.20 over the course of July (−35%). So "the market had never priced this thread" is a fact, but "this is new news" is not. Today's increment is a source upgrade (Reuters plus three sources) and a structural addition (a cloud-provider JV), not the thread itself.

Catalyst logic: what this affects is valuation and narrative, not revenue or EPS. Three hard constraints make the near-term financial impact roughly zero:

1. The current schedule for Ohio One is Mod 1 completed in 2030 and production in 2030–2031 (primary source: Intel's own Newsroom, "Ohio One Construction Timeline Update"), with construction currently only at "the first fab's below-grade level complete, above-grade structure under construction."
2. ⚠️ The wording must be corrected: Ohio is not "idle capacity," it is "an unfinished construction site." What SK Hynix would get is not a cleanroom it can tool up immediately, but a half-built site that still needs tens of billions of dollars of further investment and a redesign for DRAM/HBM processes. That directly determines how fast this can be realized.
3. Ohio is currently not in the foundry segment's income statement (capitalized construction in progress, no depreciation charged). So "monetizing Ohio" will not directly narrow that foundry TTM operating loss of −$9.36 billion in the near term — this is the most easily confused link in the market narrative.

Fundamental verification (done by a sub-agent using SEC EDGAR primary sources; yfinance was rate-limited this morning, see the internal record at the end):

The foundry (Intel Foundry) segment — where this news lands, so it must be broken out:

⚠️ Two facts the sub-agents dug up that change the conclusion:

• Foundry revenue is almost entirely internal transfer. Q2 inter-segment eliminations were −$5.477 billion against foundry segment revenue of $5.765 billion — the two are nearly equal. The only external revenue the company explicitly discloses is $293 million per quarter from advanced packaging (year-ago $22 million, a 13-fold increase, but only 5% of foundry revenue). So the driver of the narrowing foundry loss is "internal product volume plus yield improvement," not external customers. The narrative that "external customers arrive → foundry turns profitable" has, so far, no magnitude support in the financials.
• INTC's GAAP net income is negatively correlated with its own share price. Within Q2's −$11.033 billion GAAP net loss, operating income was actually +$1.796 billion; the entire problem sits in "interest and other, net −$12.576 billion," of which $12.529 billion is a mark-to-market loss on the "Escrowed Shares" under the agreement with the US Department of Commerce — that item is classified as a derivative liability, whose fair value rose from $2.654 billion at 2025-12-27 to $15.570 billion at 2026-06-27 (10-Q notes 4/13, original text). Share price up → a huge non-cash loss is booked. Any attempt to compare INTC across peers using GAAP P/E or ROE is invalid under this structure.

Moat/market position: the only advanced-node IDM on US soil, and the strategic scarcity is real; but earnings quality is still weak — on the company's own non-GAAP basis, annualized ROE is about 8.0%, while the market pays 4.8x P/B.

Pre-market and technicals: $99.99 +2.93% (08:04) → $100.35 +3.30% (08:16) → $100.68 +3.64% (08:33, still strengthening after retail sales); cumulative pre-market volume about 6.5 million shares, notional roughly $600–650 million, first in the field. It is −31.8% from the 52-week high of $142.35, and 3.97x the 52-week low of $24.45.

⚠️ Three risks that must be stated (all from primary sources):

1. A $23 billion offering was just completed in August 2026 (8-K, 2026-08-12): 210,526,315 shares at $95.00, upsized from an original plan of $15 billion to $20 billion, with underwriters exercising the greenshoe in full for 31,578,947 shares, for a total of 242,105,262 shares. Today's $100.68 is only 6.0% above the offering price.
2. 14 months of dilution, +20.8% (shares outstanding 4.377 billion → 5.286 billion). Every "capital raised" is simultaneously a dilution, so per-share metrics improve markedly more slowly than company-level metrics.
3. The magnitudes do not match: +3.64% ≈ about $19 billion of added market value, more than 2x Intel's entire TTM foundry loss, and more than the present value of any conceivable rental income.

Final judgment: priority deep-dive, but track it as a "narrative repricing," not an "earnings inflection." It is the highest-volume name in the field with the most direct catalyst and deserves the top slot; but it simultaneously carries four giveback conditions: rumor-grade + round two + just-completed offering + already up nearly 4x. The cleanest verification point (falsifiable): the gap between the Seoul line for SK Hynix (000660.KS) and INTC. If the Seoul line is clearly weaker than INTC, it says the market reads this as "Intel benefits from selling an asset" rather than a mutual positive, or that the Korean market is more relaxed about news it already heard in July — either reading points to limited incremental information. Korea's KOSPI closing +1.37% today is index-level evidence, but it cannot substitute for a single-stock read.

---

② DELL — Dell Technologies | watch closely | total score 59

Related news: no stock-specific news. A sub-agent went through every SEC filing after 8/11 one by one; DELL's most recent fundamental disclosure is the FQ2 earnings on 9/1, after which there is only the 9/10 8-K (Item 8.01) and the $5 billion senior notes offering on 9/15 (8-K Item 1.01, settled 2026-09-15: 5.100%/2029 $1.25 billion, 5.400%/2031 $1.25 billion, 5.600%/2033 $1.5 billion, 5.900%/2037 $1 billion).

Catalyst logic: this morning's move is a refill of the 9/14 "AI capex worry" rotation, not new information.

Fundamental verification (SEC EDGAR primary source, FQ2 FY2027, ended 2026-07-31):

⚠️ Three points of methodological discipline:

1. The $60.9 billion is orders booked in the quarter, not a cumulative figure (the company's own words: "we booked a record $60.9 billion in orders," directly comparable QoQ with the prior quarter's $24.4 billion). This is genuinely new, not old news. ✅
2. The $95 billion is a point-in-time stock, and cannot be added to the $60.9 billion nor treated as this quarter's new bookings.
3. 🔴 The "Adjusted FCF +224%" is manufactured by the definition. True FCF is $986 million, −47% YoY. The largest single item in the gap is an increase in financing receivables of $6.667 billion (year-ago only $592 million, 11x), equal to 14.2% of the quarter's revenue. The balance sheet confirms it: financing receivables rose from $14.280 billion at end-January to $20.430 billion at end-July. This appears only in footnote (b) of the reconciliation table; it is not in the body of the 8-K. Judgment: Dell is financing AI server sales with its own on-balance-sheet credit — the cash-conversion risk in that $95 billion backlog may be systematically underestimated.

Pre-market and technicals: $557.96 +2.66% (08:03) → +2.63% (08:16) → +2.43% (08:33, the only key name to fade after retail sales). 🔴 Its location is the most expensive in the group: only −4.4% from the 52-week high; yesterday it touched $568.67 intraday, exactly equal to the 52-week high, and closed at $543.51, giving back −4.42% from that high. This morning's pre-market $557.96 is precisely a retest of the level that was rejected yesterday. Relative to its own history, EV/Sales and EV/EBITDA are both at 5-year highs, and forward P/E is in the 95th percentile, 1.74x the 5-year median (11.1x).

Final judgment: watch closely. The order data is the only hard evidence among the three names of a step-change in magnitude, but the price is standing at its most expensive level in five years while retesting yesterday's high. The fundamentals can support the story; they cannot support "chase it today."

---

③ LITE — Lumentum | watch closely | total score 61

Related news: no stock-specific news; the most recent disclosure was the FQ4 earnings after the close on 8/11.

Fundamental verification (SEC EDGAR, FQ4 FY2026, ended 2026-06-27) — the most solid fundamentals of the three:

⚠️ Two methodological traps that must be called out:

1. GAAP EPS of −$84.65 has to be adjusted back and cannot be used directly. It comes from a one-off non-cash debt-extinguishment loss of $7.7566 billion created when part of the convertible notes were converted into equity in FQ4. LITE's GAAP P/E is completely meaningless; only non-GAAP can be used.
2. Market cap is understated by about 12.6%: third parties use 89.7 million common shares, but the 2.9 million Class A convertible preferred shares held by Nvidia convert 1:1 into common (2026-03-02, $695.31 per share, $2 billion in total). On the non-GAAP diluted count of 101 million shares, true P/S is 28.1x rather than 25.0x. 🔴 28.1x is LITE's own 5-year P/S high, and 6.6x its median. Forward P/E of 38.7x is only in the 75th percentile — the entire difference comes from operating margin going from 9.7% to 29.8%. Margin mean reversion is the ruler that actually binds. Short interest is 7.94% of the float (the highest of the three), which shows the disagreement is public.

Pre-market and technicals: $858.00 +2.27% (08:04) → +2.33% → +2.20% (08:33). It is −22.7% from the 52-week high, above both the 50-day and 200-day moving averages. Pre-market $858.00 vs yesterday's intraday high of $856.00 — it too is retesting yesterday's high. It rose 13.63% the day after earnings, and is +2.2% from the pre-release close — the only one of the three where the guidance itself produced a positive expectation gap.

Final judgment: watch closely. The fundamentals are the cleanest of the three (growth, margin and cash flow all hold at once); the risk is entirely in the price, not in the company.

---

4Coherent | watch onlyCOHRI have written the sub-agent's ranking view into this piece and cut the position my own draft had given it

Related news: no stock-specific news; the most recent disclosure was the FQ4 earnings after the close on 8/12.

Fundamental verification (SEC EDGAR, FQ4 FY2026, ended 2026-06-30):

Item	Value
Quarterly revenue	$2.0455 billion, +33.8% YoY, +13.3% QoQ
non-GAAP gross margin	40.2% (+215bp)
non-GAAP EPS	$1.74 (+74% YoY)
Datacenter & Communications segment (quarter)	$1.615 billion, +58.6% YoY, +18.6% QoQ, 79.0% of the total
Industrial segment (quarter)	$430.5 million, −15.8% YoY, −3.0% QoQ
FY26 operating cash flow	$79.5 million (FY25 was $633.6 million, −87%) 🔴
FY26 capex	$1.1029 billion (FY25 $440.8 million, 2.5x)
FY26 free cash flow	−$1.0234 billion 🔴
Net-income-to-cash conversion	10% 🔴
Inventory days	133 days → 187 days; inventory +79.5% against revenue +22.5%
ROIC vs WACC	6.74% vs 15.14% 🔴 (the only one of the three destroying capital on this basis)
Share count YoY	+26.3% 🔴
Valuation percentile	EV/Sales 95%, EV/EBITDA 90%

Table scrolls horizontally

🔴 This is the hardest negative finding in the whole piece: COHR's GAAP net income jumped from $49.4 million in FY25 to $805 million in FY26 (+1531%), while operating cash flow over the same period fell from $633.6 million to $79.5 million. Broken down: inventory absorbed −$1.183 billion, receivables −$367.6 million, payables +$748.5 million → a net working capital drag of about −$802 million. DSO barely changed (57→60 days); the problem is precisely concentrated in inventory.

Verified equity matters: on 2026-03-02 Nvidia subscribed for 7,788,161 shares at $256.80 per share, $2 billion in total (the same date and the same amount as its LITE stake), with a 6-month lockup (expiring around 2026-09-02). The current price of $271.17 is only 5.6% above Nvidia's cost.

⚠️ The most important comparison: COHR fell on earnings day (−7.99%), and from the pre-release close of $355.64 to $271.17 now is −23.8%. A report with revenue +33.8%, non-GAAP EPS +74% and gross margin +215bp, and the market spent a month voting against it — voting against the guided deceleration and the cash flow, not the quarter itself. This morning's +2.93% has not overturned a single one of those objections. Technically: it is below both the 50-day and the 200-day moving averages (the only one of the three in a downtrend), −38.4% from the 52-week high, RSI 43.9, beta 2.10 (LITE 1.54, DELL 1.41).

Final judgment: watch only. Under the "basket refill" explanation, the name that fell the most and has the highest beta bouncing the most is a mechanical result and needs no fundamental reason at all — COHR was −12.73% on 9/14, the worst in the group. But the opposite possibility must be written alongside it (direction and holding value are two orthogonal conclusions): COHR's segment growth is in fact the fastest of the three (datacom +58.6% YoY and +18.6% QoQ for the quarter, faster than DELL's +2% QoQ AI server revenue). Whether its cash flow problem is directionally "prepaying for growth" or "stuffing inventory against inflated demand" cannot be distinguished from the current data — the only test is whether inventory days (currently 187) come down in FQ1. "The fundamentals cannot support this morning's move" is an attribution of the move, and does not constitute a prediction of the subsequent price direction.

---

⑤–⑦ ASML / AMAT / LRCX — semiconductor equipment | watch closely

The 6-name equipment basket is 6/6 all up, equal-weighted +2.09%, the most perfect breadth of any line this morning. ASML +2.95% ($90.19 million), LRCX +2.23%, AMAT +2.09%, KLAC +1.90%, ENTG +1.82%, TER +1.55%.

⚠️ One point of attribution discipline that must be flagged: equipment orders/revenue are a "supply-side" indicator, not confirmation of "demand-side" strength. Equipment rallying reflects fabs' willingness to spend capex — which is a different thing from "how strong end AI demand is," with 1–2 years of capacity release in between. Today this line is logically consistent with the Intel/SK Hynix news (if Ohio really is activated it will need a lot of tooling), but that is a 2030 matter.

Final judgment: watch closely, with ASML as the preferred expression (largest notional, hardest moat). But do not read the equipment rally as "AI demand beating expectations."

---

⑧ Precious metals basket (GDX / NEM / WPM) | watch closely

Gold spot +1.09% to $4,339.41, silver spot +1.41%; all 15 gold and silver miners up, 15/15, equal-weighted +2.08%; GDX +1.87% ($36.53 million). ⚠️ This is not an all-time high: spot at $4,339.41 is still 22.4% below the 2026-01-28 record of $5,589.38. Final judgment: watch closely, expressed through GDX (the single names AEM/WPM/NEM have pre-market notionals of only $920,000–$4.13 million, far too thin). The direction depends entirely on the real-rate path after 14:00; this is a purely event-dependent position.

---

9Micron | watch onlyMUthe transmission may run the other way

Pre-market +1.01%, notional $453 million (second in the field). 🔴 The market is buying Intel/SK Hynix as a "memory chain positive," but the direction of that transmission deserves doubt: SK Hynix adding memory capacity in the US = future supply increase, which is medium-to-long-term bearish rather than bullish for MU. There is no near-term impact (production only in 2030), and the far end is intensified competition. MU reports on 9/30, which is a harder catalyst. Final judgment: watch only. The $453 million of pre-market volume says people are trading this logic seriously, but the logic itself has never been verified for direction.

---

10Nvidia | watch onlyNVDAincluded here as counter-evidence

Pre-market +0.81%, notional $292 million. 🔴 NVDA's gain is less than half that of the whole AI hardware line (basket +1.84%, with COHR/AAOI/CIEN all near +2.9%), while it is only −10.3% from its 52-week high. The statement that "the leader is not leading" is itself a fact and holds.

⚠️ But my draft took one step beyond that, and that step does not survive a full-sample test, so it must be withdrawn. The draft said: "the gains across the whole line are highly aligned with the distance from the 52-week high — AAOI, −59.2% from its high, rose 2.92%, while NVDA, −10.3% from its high, rose only 0.81%," and used it as a test for "oversold beta repair." That conclusion rested on only 2 points. Feeding in all 18 constituents:

Test	Result
Pearson corr (distance from 52-week high %, this morning's pre-market %)	−0.175
t-value / n	−0.71 / 18 → not significant
R²	0.030
Spearman	−0.243

Table scrolls horizontally

And the two cleanest counterexamples are in my own table: MOD is −44.7% from its high (deeply fallen) yet is the only name in the group that is down (−0.04%); DELL is only −4.4% from its high (the most expensive in the group) yet is up +2.66%, the 4th strongest in the group. So "the deeper the fall, the bigger the bounce" does not hold across these 18 names. I keep the single-point fact that "the leader NVDA is not leading," and withdraw the regularity claim that "the whole line is sorted by degree of oversold." Correspondingly, the closing sentence of §9 ④ has been rewritten accordingly.

---

11携程TCOMwatch closely | total score 60 (the only positive name in this piece driven by its own earnings) · Trip.com

Related news: primary company disclosure. The call was at 09-15 20:00 ET, and the 6-K was filed at 09-16 02:02 ET (accession 0001193125-26-392473). SEC 6-K Ex-99.1¹¹

Earnings highlights (Q2/H1 2026):

Item	Value
Net revenue	RMB 15.7 billion (US$2.3 billion), +6% YoY, −3% QoQ
Accommodation reservation	RMB 6.6 billion, +6%
Transportation ticketing	RMB 5.4 billion, −1%
Packaged tours	RMB 1.2 billion, +8%
Corporate travel	RMB 771 million, +11%
International platform revenue	more than +50% YoY, with inbound travel growing at a high double-digit rate
non-GAAP net profit attributable to parent	RMB 4.8 billion (US$706 million)
non-GAAP diluted per ADS	RMB 7.27 = US$1.07 (year-ago RMB 7.20)
vs consensus	US$1.07 vs US$0.98, a 9.2% beat; revenue 0.8% above

Table scrolls horizontally

🔴 The GAAP loss is a known item, not a new landmine, and this must be spelled out: the GAAP operating loss of RMB 1.462 billion and net loss attributable to parent of RMB 2.5 billion come entirely from the RMB 5.2 billion (US$763 million) SAMR antitrust fine booked into G&A (which is why G&A rose 477% YoY). That fine was already public in late July 2026, and is not new information today. Excluding it, net profit is RMB 2.7 billion. Looking only at the GAAP headline would turn a beat into a blow-up — this is the textbook case of "GAAP and non-GAAP cannot be mixed."

The bull case: the fine landing = regulatory uncertainty cleared + a non-GAAP beat + the international line up more than 50%.

⚠️ The counter-evidence that must be placed alongside it: Mizuho cut its price target from $65 to $60 this morning (09-16), maintaining Outperform. The sell-side reaction is not aligned — this is not a report that turned everyone bullish. Also: the release contains no guidance and no buyback announcement.

Pre-market and technicals: $41.05 +4.59% (08:05) → $41.25 +5.10% (08:16). Pre-market notional is only $4.97 million, the thinnest of any name in this piece's positive zone. The previous close was $39.25 against a 52-week range of $38.04–$78.99 — it is basically sitting 3% above its 52-week low, and −50.3% from its 52-week high.

Final judgment: watch closely. This is the only name in this piece's positive zone whose catalyst is the company's own earnings — it depends neither on theme transmission nor on a rumor. But two hard constraints keep it out of the top tier: ① pre-market volume of only $4.97 million, so absorption is questionable; ② sell-side disagreement is openly on display (a price-target cut this very morning).

---

6. Negative/avoid list

Two methodological reminders specific to this section:

1. Avoid ≠ short. Of the 7 groups in the table above, I give a "short-watch" only to CRWD, and with a time condition attached. "JBHT is down 11%" and "JBHT should be shorted" are two different things — for a position whose gap has already completed, the pre-market price is the worst possible entry.
2. ⚠️ The FANG line is the biggest honesty cost in this piece. See the box below.

📌 Box A: FANG — keeping "could not find it" separate from "it did not happen"

A large cap with $26.26 million of pre-market notional that is down about 3pp more than its sector, and I ran through every channel I could think of and still did not find the cause. Hypotheses ruled out by evidence (not "not found," but "checked and rejected"):

There is also a counterintuitive finding that overturns a very natural explanation: yesterday (9/15) FANG's +2.82% actually underperformed the sector — the same day APA was +5.29%, EOG +3.50%, COP +3.33%, XOP +3.22%. So the "it rose too much yesterday and is giving it back today" line does not hold; it did not overshoot at all yesterday.

Possibilities neither ruled out nor confirmed (not written as conclusions): ① a sell-side downgrade issued this morning that has not yet been indexed; ② an overnight agented block trade done by a legacy holder (that produces no same-day SEC filing and requires no company release).

📌 Box B: WING's +5.84% — a number I have to knock down myself

This is today's cleanest live demonstration that "pre-market quotes are cumulative; a single snapshot is not a fact":

1,138 more shares traded, and the gain collapsed from +5.84% to +1.45%. The +5.84% in my main table was built on a single-digit number of trades; it is not "the market thinks WING has stopped falling," it is just a few scattered buy orders hitting an empty book. Treatment: this piece voids WING's pre-market move entirely and does not use it as an input to any judgment. It is kept in the table to display the trap itself.

📌 Box C: BE — no news today; what is rallying is the forced buying at the 9/18 close

I originally had BE listed 17th in the positive zone and marked "catalyst pending verification." The result of that verification is that it has to be moved to the avoid zone, on the grounds that all three "today's catalysts" are old news:

What is actually happening: the S&P 500 inclusion is effective before the open on 09-21, which means passive funds' forced buying must execute at the 09-18 (this Friday) close — and 9/18 is also quarterly quadruple witching. This morning's 300,000 pre-market shares ($80.8 million) is the heaviest of the five unusual movers I checked, and this is a trade front-running an index rebalance, not a news trade.

⚠️ By our own test, whether front-running is already priced is judged by relative volume, not by the percentage move, and this line has already been slapped in both directions: BE was down as much as 7.0% pre-market to $256.37 on 09-14, precisely because the early front-running buyers were giving it back. There is also a pending securities class action (class period 2025-02-27 to 2026-07-08, lead-plaintiff deadline 2026-09-28). Conclusion: avoid. Not because I am bearish on Bloom Energy, but because what is being traded at this price is mechanical buying at the 9/18 close, not fundamentals, and the front-running has been proven to cut both ways.

📌 Box D: JBHT is falling on costs, not demand — and that is why it belongs in "avoid" rather than "short"

drayage refers to the short-haul trucking of containers between rail ramps and shippers/receivers. JBHT outsources a great deal of this to third-party drayage carriers, so it lands on the purchased transportation line, which is an externally sourced cost — not the wages of JBHT's own drivers.

The cost side is being squeezed by two things at once: ① the drayage market is tightening (driver wages rising), and the company already warned on the Q2 call that "we need to add drayage capacity, and that will be a headwind"; ② diesel at roughly $6 per gallon, while fuel-surcharge recovery lags — a fast run-up in diesel compresses intermodal margins directly within the quarter. Even though WTI is down 2.1% this morning, refined products transmit far more slowly than crude.

🔴 And the demand side is good: Q2 2026 intermodal volumes +10% (Eastern network +16%), intermodal segment operating income +58%, improving throughout.

This explains something that looks contradictory at first glance: BofA cut its price target from $340 to $302 while keeping the Buy rating. The two banks' actions point the same way but are different in kind — BofA maintains Buy (viewing this as a cost-timing issue), Barclays maintains Equal Weight (both dated 09-16 around 06:30 ET).

Implication for trading: margin problems usually recover faster than demand problems (fuel surcharges catch up, drayage rates peak), so shorting at −11% is betting on direction in a cost-timing problem, and the risk/reward is asymmetric. That is why this piece gives "avoid" rather than "short watch."

7. Intra-theme rankings

7.1 AI hardware/optical modules (18 names, equal-weighted +1.84%, 17 up 1 down)

There is one structural flaw inside this theme: the gains are highly aligned with "distance from the 52-week high" — the deeper the fall and the lower the quality, the bigger the gain (AAOI, −59.2% from its high, rose 2.92%; NVDA, −10.3% from its high, rose only 0.81%). That is the shape of beta repair in oversold names, not the shape of an industry move led by the bellwethers.

7.2 Precious metals (15 names, equal-weighted +2.08%, 15/15 all up)

Silver is stronger than gold, and the silver miners (AG +2.74% / HL +2.36% / CDE +2.64%) are stronger than the gold miners — in a pure risk-off move silver is usually weaker than gold, so this looks more like "risk appetite rising across precious metals" than "hedging ahead of the FOMC." ⚠️ But this evidence is weaker than it looks and must be self-downgraded: on a futures basis silver (+2.04%) beats gold (+1.20%) by 0.84pp, while on a spot basis the edge is only 0.32pp (silver +1.41% vs gold +1.09%). The same conclusion differs in strength by 2.6x depending on which ruler is used, so "risk appetite rising" can only be called a lean, not a test.

---

---

8. Opening verification signals

Pre-market (before 09:30)

• The one thing in this piece that most needs verifying: after 09:30, is SMH a gap-and-go or a gap-fill? Yesterday this leg said everything before the open and then fully reversed after it (semiconductor equipment, 8 names, open→close, 8/8 all negative). If SMH falls back to $542.11 (the 9/15 close) within 30 minutes of the open today, the AI hardware theme is void for the day, no matter how pretty the pre-market looked.
• Whether INTC's volume holds: $577 million of pre-market notional is first in the field, but this is "rumor-driven" volume. If turnover in the first half hour after the open does not scale up to a reasonable proportion of average daily volume, it says this was overnight speculative flow, not allocation flow.

Intraday

• 10:30 ET EIA crude inventories (before the FOMC): this is the only hard data today that can falsify or confirm the energy line. API reported a 7.14 million barrel build; if EIA comes in clearly below API or flips to a draw, this morning's energy decline gets pushed back, and the pre-market declines in FANG/COP/APA cannot be extrapolated.
• 14:00 ET FOMC + 14:30 ET press conference: the real variable is not the 25bp (already 92–95% priced), but ① the dot plot — the June dots showed 3.8% at end-2026 (i.e., only one hike), with 9 members expecting at least one and 8 expecting no change; if today's dots shift up to imply "there is another one coming," that is a completely different scenario from current pricing. ② whether Warsh's wording frames this as "a one-off adjustment" or "the start of a cycle."
• Sector ETF confirmation: for the AI hardware line to hold, the SMH–IGV scissor gap needs to keep widening (currently 1.64pp). If the gap narrows or inverts after the open, this morning was just overnight noise.

Options sentiment

• VIX 16.77 (−2.50%, 08:15 ET) — on the morning of a decision day where a hike is 92% priced and would be the first in three years, VIX is below 17 and still falling. That either says the market is certain the outcome is known, or it says not enough protection has been bought. I lean toward reading the latter as the risk more worth watching.
• 9/18 (this Friday) is quarterly quadruple witching, and gamma near expiry will amplify directional moves after the FOMC.

Risks

• Reversal after a gap: this is what killed yesterday's piece, and it applies equally today — plus today there is an extra event gate at 14:00 — any position built pre-market has to pass through it.
• A lone name rising with no sector follow-through: INTC is a single-stock catalyst, and the simultaneous gains in MU/SNDK/STX/WDC are not necessarily the same-direction logic (see row 6 of §3A: SK Hynix adding US capacity is a supply increase for MU over the medium to long term).
• Index futures diverging from Treasuries: this morning futures are up, yields are down, VIX is down and gold is up — gold and equities rising together is usually not a combination that can coexist for long, and after 14:00 one direction will most likely have to break out.
• Waiting on macro data still to come: this piece's reads were completed at 08:06/08:16 ET, and the three gates — 08:30 retail sales, 10:30 EIA, 14:00 FOMC — all come after it.

---

---

9. Final conclusions

① The 5 names most worth watching today

② The 3 strongest themes today

⚠️ One self-imposed constraint regarding theme 3: a sub-agent went through every SEC filing for COHR / LITE / DELL after 8/11 one by one and confirmed that none of the three has made any new fundamental disclosure since earnings. So the new fundamental information content behind this morning's +1.84% is zero; it is a repricing of the same set of old facts. That does not mean it will not rise, but it does not have the persistence of something "driven by new information."

③ What to avoid today, and why

④ Final one-line judgment

Today's pre-market is a table with excellent breadth and very little depth — 425 of 575 names are up and four theme baskets are up almost across the board, but the biggest gains belong precisely to the names that fell the furthest, carry the highest beta and have the weakest fundamentals (AAOI, −59.2% from its 52-week high, rose 2.92%, while the leader NVDA, −10.3% from its high, rose only 0.81%); the only names driven by "new information from the company itself" are INTC (rumor-grade, and round two of that July thread) and TCOM (a genuine earnings report, but with only $4.97 million of volume); and a +1.2% retail sales beat leaving the 2-year Treasury motionless already says the market has put the entire pricing weight of today on 14:00 — so any price formed between the open and two in the afternoon is a "price while waiting" and has no property of being inherited, and the lesson of yesterday's piece is exactly this: the pre-market table can be entirely true and still have fully played out before the opening bell.

---

⚠️ Risk disclaimer: this list is pre-market information gathering and observation only and does not constitute investment advice. US equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals occur; automatically generated content may contain timeliness gaps or factual errors. Please rely on company disclosures/SEC filings, and do not use this directly as a basis for trading.

🔧 Internal record: data-pull failures and channel status for this run

1. yfinance was rate-limited repeatedly, independently confirmed by two sub-agents.

• .info / .history / .financials returned YFRateLimitError on the first call (read at 09-16 12:13 UTC); only .fast_info worked.
• Direct calls to query1 / query2 for /v8/finance/chart and /v1/test/getcrumb both returned Edge: Too Many Requests.
• ps aux ruled out leftover local processes → this is an IP/endpoint-level block, and retrying does not help, a different root cause from the "leftover process" class.
• Remedy: all financial data was switched to SEC EDGAR primary sources (XBRL companyfacts + 10-Q/10-K R-files + 8-K EX-99.1). This run's INTC / COHR / LITE / DELL financial figures all come from EDGAR originals, and the quality is actually higher than yfinance's.

2. Scrape channels that failed in this run (so they can be skipped next time):

3. Broken quote fields (the Nth time this has come up; already written into the methodology notes in the body):

• CNBC's US2Y change_pct field returned +0.0547% while within the same record last (4.636) < prev (4.663). This run worked around it by reverse-computing from the Treasury's official CSV. This trap is recurring; recommend permanently distrusting the net-change fields for Treasuries.
• DXY= returns all null on CNBC; used EUR= / JPY= to express the dollar instead.

4. Nasdaq fake pre-market quotes: 18 of 738 names had isRealTime=false (a date only, no hours and minutes) and were all removed; the list is in the output of work_uspre0916/an.py. Without removing them, ADTN +0.57%, CALX +2.16%, OTLY −3.53% and others would contaminate the gainers/losers tables.

5. FANG cause not identified — the list of channels already exhausted (next time, start after this list): EDGAR filing list + full-text search (only 3 Form 144s), the Nasdaq dividend API, the company press-release RSS (latest stops at 08-03), stockanalysis (latest 08-30), MarketBeat (latest 09-15 19:36), finviz (latest 09-11), investing.com (latest 09-14), TradingView news, the company IR site (Akamai refusal), StockTitan (DNS), Benzinga (403), CNBC live (403); also confirmed there was no energy conference featuring FANG on 09-15/16. Channels not covered: Bloomberg / Reuters terminal copy, and brokerage morning-meeting notes. Those two are the first additions to try next time.

6. Third-party data errors (found and worked around):

• The FANG previous close of $206.29 given by stockmarketwatch is wrong; Nasdaq / CNBC / finviz all agree on $211.525.
• stockanalysis puts COHR's FY26 FCF at −$650 million, while Table 4 of the 10-K works out to −$1.0234 billion (OCF 79.5 − capex 1,102.9). Switched to the EDGAR original.
• The Saxo morning note's Nikkei/KOSPI figures were intraday reads treated as closes; corrected using CNBC closing prices (see §0 item 4).

7. "Fake today's catalysts" blocked during this run's verification (four items, none used):

8. Items pending verification (flagged in the body, collected here for recap reconciliation):

• JBHT's "Q3 −5% to −10% QoQ" rests only on a paraphrase of the CNBC live broadcast, with no second independent source; the webcast replay address has been recorded in the sub-agent's conclusions and can be listened back to for verification at recap time.
• INTC's cumulative capex already invested in Ohio: neither the 10-K nor the 10-Q breaks it out by site, confirmed not disclosed. The widely circulated "$28 billion" is a planned commitment, not an amount already spent, and the two may differ by an order of magnitude; they must never be used interchangeably.
• COHR / LITE non-GAAP EPS for Q3'25–Q1'26 was not fully obtained, so the denominator basis for forward P/E is incomplete.