Today on Closelook

1. 1 Korea: the second leg Structure

   What

   Seoul crashed again — KOSPI down as much as 8% intraday to a close around −7%, another market-wide circuit breaker, the eighth this year. SK Hynix lost another ~9% after its print, Samsung ~7%. That follows Tuesday's −10.8% close at 6,024, where Hynix fell 14.7% and Samsung 13.4%.

   If

   Seoul opens Thursday without a third leg — the two-day repricing reads as an event, not a regime.

   Why

   Yesterday's edition asked Seoul to stabilise as the first external witness of a bottom. It printed the second leg instead — Monday's bought-flush was the anomaly, not the signal.

   Then

   The question crosses the Pacific: does the US session discriminate, or import the block-sale? Points 6 and 10 carry the answer.

2. 2 The print that did it: SK Hynix Structure

   What

   A record quarter that missed the bar: Q2 revenue ₩79.3T and operating profit ₩60.5T — up 557% y/y — still came in ~6% below consensus on both lines. And the capex line: Hynix held FY26 spending at the high end of its already-guided ₩40T range, while the street had modelled a raise.

   If

   Thursday's Samsung print repeats the shape — record levels, missed expectations, no capex raise — the Korea repricing becomes a memory-sector verdict, not a Hynix story.

   Why

   The punishment regime has widened: raised capex was the week's question, and now unraised capex gets sold too. Expectations, not results, are the asset being repriced.

   Then

   Phrase it precisely: Hynix did not cut capex — it declined to raise it. The market treated that as the same thing.

3. 3 Outside view — "ammo for THE TOP" Outside view

   What

   Dan Niles reads the Hynix miss and the sub-consensus capex guide as ammunition for his call that an AI spending top is forming — his framing: a near-term "speedbump" that can get as ugly as late-1995/1997 did for the internet buildout, inside a build that continued for years afterwards.

   If

   N/A

   Why

   The analogy cuts both ways and he says so himself: 1995–97 corrections were vicious and the buildout still ran to 2000. Our forward map says the build re-arms through the next silicon generations; the tape says the reaction function has flipped now. Both can be true — that is the position.

   Then

   Read him directly, weigh it against the duopoly evidence in point 6, and hold probability, not prophecy.

   Dan Niles · Niles Investment Management ↗

The full Morning 10 →