Market Insights

Test an A-share AI research tool against a real annual report: identifiers, dates, sources, profit definitions, uncertainty and reproducible records.

A 100 MW facility supports about 45,000 GPUs under explicit rack-power and overhead assumptions. See why IT capacity, annual PUE and commissioned capacity differ.

Micron FY2026 gross cash spending was $30.712 billion versus $27.367 billion of net capex. We reconcile the $3.345 billion gap and use a hypothetical receipt delay to show why net spending changes cannot directly measure capacity changes.

As of October 14, 2026 US information, distinguish a media-reported bond yield from fixed coupons and new funding costs, and trace the conditional link to AI capital spending and supplier orders.

Samsung’s October 12, 2026 buyback results show 6.70% more shares than estimated but KRW 92,250 less spending. Reconciling 29 daily records and the weighted price explains the quantity difference, without treating compensation-purpose treasury shares as cancelled.

At the October 12, 2026 Korean close, foreign investors sold both SK hynix and Samsung Electronics, while institutional flows diverged. Volume-normalized flows and closing-price value proxies separate a common foreign-selling signal from an unproven sector-wide exit.

At the October 8, 2026 U.S. close, an equal-weighted Micron and SanDisk basket fell 4.85%, 2.01 percentage points more than the SMH semiconductor ETF. The gap supports a two-layer description—semiconductors were weak and memory was weaker—but does not identify a one-day cause. We use broad-market, growth, and non-technology sector ETFs as checks and set out what to verify next.

Using Samsung’s preliminary October 8, 2026 disclosure, consolidated Q3 operating margin rose about 2.90 percentage points from Q2. A revenue-first bridge assigns KRW 5.65 trillion of the KRW 17.91 trillion profit increase to margin expansion, or 31.53%; this is not HBM attribution or proof of durable growth.

October 7, 2026 review: Samsung's buyback reportedly ended on October 6. Approximately KRW 15 trillion over 29 sessions implies historical average buying of KRW 517.24 billion, not next-day selling or other corporations' net buying. Conditional scenarios separate past purchase rates from hypothetical missing demand and unverified price effects.

For the quarter ended July 31, 2026, Palo Alto Networks grew adjusted net income 26.7% but adjusted diluted EPS only 7.4%. We reconstruct the impact of a 17.7% increase in the matching share count, retain rounding differences, and explain why this alone cannot establish acquisition value creation or destruction.

Micron HBM4 and HBM3E 12-high products both hold 36GB, so higher bandwidth is not a revenue-growth multiplier. Separating volume, capacity and realized price shows that, at unchanged capacity and an assumed 30% price increase, doubling revenue requires approximately 53.85% more volume. This is a reproducible scenario, not a company forecast.

An October 1, 2026 intraday decomposition: MU and SNDK lag SMH by 1.37 and 0.50 points after the open. Testing whether opening gaps explain their relative weakness.

On September 30, 2026, the intraday MU/SNDK average gained 0.52%: +0.58 pp versus SMH but -0.08 pp versus QQQ. A same-snapshot decomposition tests what semiconductor outperformance really establishes.

In the 2026-09-24 session Micron's (MU) at-the-money implied volatility jumped from 59.9% to 90.9% as the nearest expiry rolled from 09-28 to 10-02, the first contract to span the 09-30 earnings date. Isolating earnings-day variance, options price Micron's post-earnings move at roughly ±10.5% to ±11.7% (one standard deviation). Controls confirm it: SanDisk and Western Digital did not roll, fell 3.5%–4.9%, and saw IV decline; AMD and NVIDIA rolled into non-earnings contracts and IV moved only 1–3 points. Protection demand sits in indexes (SPY and QQQ put/call above 1), with every single stock except Western Digital (1.02) below 1. Verdict: event pricing, not fear pricing, with falsification conditions.

In the 2026-09-23 session both U.S. and Hong Kong equities fell across the board, with declines ordered strictly monotonically by risk exposure: the U.S. ran from S&P 500 at −0.72% down to SanDisk at −3.73%, while Hong Kong ran from Jiangxi Copper at −0.41% down to Alibaba at −4.36%. Yet implied-volatility percentiles for all nine U.S. tickers sat in the bottom two deciles of a self-collected sample (high 19.7, low 0.8; n=99–128 trading days, series start 2026-03-20), and NVIDIA's IV was 10.70 points below its own 20-day realized volatility. Genuine hedging bids up protection costs; that did not happen, so the character is de-risking rather than hedging. More counterintuitive still is the inverted protection bid: the Spearman rank correlation between change and put/call ratio is ρ=+0.65 (n=9, a descriptive cross-sectional statistic rather than a significance test), meaning the more a name fell the less protection was bought on it. Micron fell 2.22% yet carries the table's lowest put/call ratio at 0.369, while TSMC at 1.344 and the Nasdaq 100 at 1.276 — the tier that fell least — carry the highest. Micron reports FQ4 earnings after the close on 2026-09-30, just four trading days out, with both its IV percentile of 8.7 and put/call ratio of 0.369 at the bottom of the table. The article also documents a CCASS 'net change' data trap with full-table evidence: of four Hong Kong tickers showing positive headline net change, two flip negative once newly-entered participants are excluded, including Tencent.