2026-10-16 AI Scan

Evidence as of 2026-10-16

Seagate closed under its 200-day average on October 16: 639.01, 5.3% under the 674.45 average, after an 18.39% five-session drop, on a year still up 182.71% (it was 2.7% above the line on October 15). The long trend is under test, not broken. The week was cohort-wide (the chip ETF SMH fell 10.53% over seven days), and the scan's cause check found only quote pages in Seagate's Friday news pull. Price evidence only; it says nothing about HDD pricing, lead times or customer orders.

The AI Trade Sold as One; Oil and Staples Took the Money

Evidence as of 2026-10-16

On October 16 storage sold with every adjacent AI-hardware lane: the nearline storage list fell 6.9% with none of its three names up, the AI storage and data-pipeline list fell 5.6% at the median with none of 17 up, memory fell 6.0% with none of 10 up, and optical and neoclouds also had no gainers. A lane-wide sale is not evidence on HDD pricing, lead times or orders.

Everything Bounced but the Chips

Evidence as of 2026-10-15

On October 15 Seagate fell 3.19% on media reports that Toshiba will expand hard-drive production and on CXMT pressure across memory stocks (StockAnalysis; a media report, not a Toshiba release), and still closed 2.7% above its 200-day average. It repeats the October 2 competitor-supply fear already linked as a challenge and adds no primary capacity figure, price concession or lead-time change, so it is not counted as a second challenge.

OpenAI's Smaller Number Hits Every Layer of AI Hardware; SpaceX Hits the Carriers After the Bell

Evidence as of 2026-10-08

The October 8 slide was cohort-wide: AI hardware sold as one on the Financial Times report, confirmed by CNBC, that OpenAI's annualized revenue was about 50 billion dollars, not the 68 to 70 billion widely cited. No storage-specific cause is named. It took Western Digital 2.8% under its 200-day average; it is not evidence on HDD pricing, lead times or orders.

2026-10-02 AI Scan

Evidence as of 2026-10-02

Carries the counterpoint on timing: an X post relayed a Rosenblatt view that Toshiba's planned Philippine expansion materializes closer to 2030, preserving Seagate's position. Attributed via X and not checked against the note; it bounds the dispute to when competitor units arrive, not whether.

Soft Jobs, Higher Yields, Chips Keep Running

Evidence as of 2026-10-02

MarketWatch (October 2) tied same-day drops of about 10% in Seagate and Western Digital, on 3.6 times normal volume, to worry that Toshiba will raise output of a key AI storage product. The case's named competitor-supply test has arrived as a market fear and challenges the supply-discipline leg. No price concession, shorter lead time or margin compression is reported, so neither failure condition is met.

SemiAnalysis — ruthenium, a hard-drive input, went from about $560 an ounce in early 2025 to near $1,950 in September

Evidence as of 2026-10-02

A cost question, not a supply test: SemiAnalysis puts ruthenium, a hard-drive material, at about $560 an ounce in early 2025 against dealer quotes near $1,950 in September. Per-drive usage and pass-through are unread (only the thread's first post is captured). Check gross-margin commentary in the next two reports; this is not evidence of compression.

Terry — Citi Global TMT readout: NAND TAM $50B to $500B by 2027, AMD CPU TAM $60B to $220B, optical testing capacity about 100x short

Campaign 1: INTACT, provisional

Evidence as of 2026-09-04

Earlier same-day thesis review retained as context. Its prior classification is historical; the current scope review supplies the live disposition and preserves its unresolved evidence gaps.

Seagate: cloud storage and the density roadmap

Evidence as of 2026-07-28

Commercial HDD growth and operating economics support a separate storage case. Competitor unit supply and allocation durability remain unresolved.

WD: storage demand and operating margins

Evidence as of 2026-08-05

Commercial HDD growth and operating economics support a separate storage case. Competitor unit supply and allocation durability remain unresolved.

SemiAnalysis — 3D NAND is switching word lines from tungsten to molybdenum above ~300 layers

Memory Splits: the Shortage Story Is Intact, the Single Names Are Not

Memory rebuilt for four sessions, gave it back on Friday, and walked into the only two days that matter