Samsung HBM4E Lifts Kospi to Record as Kosdaq Drops 3.17%

South Korea’s Kospi jumped more than 3% to a fresh intraday record on Friday, Japan’s Topix set a new all-time high, and Samsung Electronics surged as much as 6.51% after shipping the industry’s first 12-layer HBM4E sample chips. On the same screens, in the same trading day, South Korea’s small-cap Kosdaq fell 3.17%.

That split is the story. Asia’s headline indexes were not shrugging off Iran. They were riding one specific trade, the AI-memory cycle, and everything outside that trade got hit by exactly the geopolitical risk reporters said had been waved away.

Samsung Did the Heavy Lifting on Korea’s Record

Samsung Electronics moved first. Before the open in Seoul, the company said it had begun shipping samples of its 12-layer HBM4E (high-bandwidth memory 4 extended, the next-generation stacked DRAM used in AI accelerators) to global customers, becoming the first memory maker to do so. The stock opened higher and ran to a 6.51% intraday gain.

Samsung is the single biggest weighting in the Kospi. When it moves 6%, the index can hold a 3% gain even if most of the names underneath are falling. That is roughly what happened Friday: the headline index printed a new high while the broader market underneath it sold off.

The press release was unusually pointed about positioning. “Following the successful mass production of HBM4, Samsung has once again demonstrated its distinct technological edge with HBM4E,” Sang Joon Hwang, executive vice president and head of memory development, said in the company’s HBM4E shipment announcement. The phrase “industry-first” is the part that mattered for the tape.

For two years, SK Hynix has been the default supplier to Nvidia for top-bin HBM. Samsung playing catch-up was the consensus trade. A first-to-sample claim on HBM4E rewrites that script, at least for one session.

The 3% Up, 3% Down Split Inside Korea

The Kosdaq does not carry Samsung or SK Hynix. It carries the smaller Korean tech, biotech, gaming, and battery names that should, in theory, also benefit from a risk-on Asia session. They did the opposite on Friday.

Foreign and institutional investors were net sellers of small caps even as they bid up the large-cap chip complex. That is not random. It is a concentration trade: when the AI-memory story is the only story global money managers want to own in Korea, everything else gets funded by selling the everything-else.

The same pattern showed up across the region. Hong Kong’s Hang Seng added 1.1% on the day, but the broader CSI 300 in mainland China was unchanged. India’s Nifty 50 traded near the flatline. Australia’s S&P/ASX 200, which has the lowest direct exposure to the AI-memory cycle of the major regional benchmarks, rose 0.72%.

Read the table from the top and the bottom at the same time. The benchmarks that hit records are exactly the ones with the highest semiconductor weighting. The benchmarks that did nothing are the ones with the lowest. There is no Iran narrative that explains that split. There is a chip narrative that does.

What HBM4E Buys Samsung Against SK Hynix

The specs are what make Friday’s pop more than a news bounce. Samsung’s 12-layer HBM4E ships at speeds the rest of the market is still chasing.

The Performance Step-Up

• 16 Gbps pin speed, with a stable 14 Gbps floor, more than 20% faster than HBM4
• 3.6 TB/s of memory bandwidth per stack, the highest publicly disclosed for any shipping HBM sample
• 48 GB capacity in the 12-layer base SKU, a 30%-plus jump over the previous generation
• 16% better energy efficiency and 14% better thermal resistance versus HBM4, the two metrics hyperscale buyers care most about

The Roadmap and the Process Node

Samsung says the 12-layer SKU is the first of a wider lineup, with 32 GB 8-layer and 64 GB 16-layer variants planned. The part is built on Samsung’s 1c DRAM node, its sixth-generation 10-nanometer-class process, paired with a 4nm logic die. Both are leading-edge for memory.

That is the technical case. The commercial case is colder. SK Hynix has been Nvidia’s lead HBM supplier through the entire current AI build-out and is preparing its own HBM4E samples for the second half of 2026. Micron is on a similar timeline. Samsung’s first-to-sample claim is real for now. It survives only if the qualification cycle with one or more US hyperscalers actually closes, which usually takes another two to three quarters.

Why the Sample Beat Matters Even If the Order Book Is Slower

Memory buyers move in long cycles. Once a hyperscaler qualifies a part for a 12-month design-in, the supply slot is locked. A first-sample lead translates directly into the negotiating room for the next two annual contract rounds. That is the math the buyside is pricing, not Friday’s trade alone.

Tokyo Rode the Same Trade, With Better Cover

Japan’s record was quieter and arguably more durable. The Topix closed up 1.86% at a new all-time high and the Nikkei 225 added 2.49%, both lifted by the semiconductor-equipment names that supply the Korean and Taiwanese fabs running the AI memory cycle.

Tokyo Electron, Advantest, Disco, and Screen Holdings all move on HBM and advanced-node capacity expansion. A Samsung HBM4E shipping milestone, validating a new process node, is a direct read-through to their order books. So the Tokyo move was a chip move too, just routed through the equipment supply chain rather than the chip itself.

The yen helped at the margin. Friday’s weaker open against the dollar gave exporters a small lift before the Bank of Japan’s policy calendar takes over next week. None of that changes the underlying point: the Topix high and the Kospi high were the same trade in two different jurisdictions.

Iran Did Not Get Ignored, the Kosdaq Got Hit

The wire framing said Asia “shrugged off” Iran. The Kosdaq tape says otherwise. Friday’s risk-off flow was real and concentrated in the parts of the market that lack a chip story to hide behind.

The Pentagon confirmed earlier in the week that Iran had fired a ballistic missile toward Kuwait and deployed five one-way attack drones in and around the Strait of Hormuz, with US forces intercepting the missile and the drones. United States Central Command called it an “egregious ceasefire violation.” A White House official then told Axios that Washington and Tehran had “mostly agreed” on terms of a temporary halt, a framing the president had not personally signed off on by Friday’s Asia close.

That is a setup, not a resolution. The visible Asia damage shows up in three places this week:

• A 5.22% Kosdaq drop on the prior session, the steepest single-day fall on any major Asia-Pacific index, covered in our prior read on the Asia oil and equity reaction to the US strikes
• Brent crude holding near $111 a barrel, with the second-order effects on Asian importers and central bank policy outlined in our analysis of the Hormuz oil premium and Fed pressure
• Foreign net selling in Korean small caps that continued straight through Friday’s headline rally

Investors did not decide Iran did not matter. They decided that if you have to own Asia risk, you own the names that have a buyer no matter what oil does.

When 82% of Stocks Fall and the Index Rises

The longer-cycle view is the part of Friday that should make institutional readers uncomfortable. The Kospi is up roughly 24% over the past month. Over the same window, 82.34% of the 2,764 listed Korean stocks have actually fallen, according to data compiled by Seoul Economic Daily.

SK Hynix is up 77.17% on the month. Samsung Electronics is up 33.41%. Take those two names out and the index does not have a rally; it has a drawdown.

Toward the end of a bubble, concentration tends to intensify rather than ease, and this time will be no different. When the concentration begins to unwind, it is not a signal of a welcome broadening but a precursor to a bubble burst.

That is Lee Eun-taek, equity strategist at KB Securities, quoted in Seoul Economic Daily. The opposite reading also has a serious advocate: Lee Kyoung-min at Daishin Securities frames the same concentration as a structural AI-memory premium that may persist as long as the hyperscaler capex cycle does. Both readings are consistent with Friday’s tape. Only one of them is consistent with the Kosdaq number.

Either way, the broad-market signal is unusually thin under the record. A Kospi top driven by two stocks is not a Kospi top that survives those two stocks repricing.

If the Iran ceasefire framework holds into next week and Brent gives back the Hormuz premium, the concentration may broaden naturally as the Kosdaq finds a bid. If the missile-and-drone exchanges continue and the deal stalls, the same large-cap trade will need to absorb both the geopolitical risk and the small-cap drawdown alone, and Friday’s split will look less like a record and more like a warning that took a week to arrive.