# SK Hynix Stock Before Q3 2026 Earnings: CXMT, Nvidia's NVHBM and the Part of HBM a Supplier Still Owns — InvestMoat Research

_SK hynix sold off in October on a Chinese DRAM claim, but I think the bigger change is Nvidia designing the base die of its custom NVHBM for all three suppliers, which takes away the custom-memory upside more than the share lead._

_Published: October 20, 2026 · Last reviewed: October 20, 2026 · Canonical page: https://www.investmoat.com/research/sk-hynix-and-the-layer-nvidia-is-standardizing_
_Tags: Semiconductors, Memory, AI Infrastructure, Catalyst_
_Covers: 000660.KS, 005930.KS, MU, TSM, NVDA, MRVL_

## Summary

SK hynix reports third-quarter results on October 28 in Seoul after a month with two pieces of news about memory. CXMT said on October 10 that it will launch server DRAM on a denser 4F² cell by year-end, and SK hynix fell about 9% intraday on October 13, but the claim was about commodity DRAM and came with no specifications. The change I think matters more is NVHBM: Nvidia designs the base die of its custom HBM, and on October 18 it was reported to have asked SK hynix, Samsung and Micron to make that die on TSMC's 3nm process and the memory dies on 1c DRAM. That takes the custom base die, the layer memory makers hoped would stop HBM being a commodity, away from all three. I argue it costs SK hynix less than its rivals, because it already buys its HBM4 base die from TSMC, and leaves the suppliers competing on DRAM yield, stacking and contracts, where SK hynix's scale lead sits, though a common drawing also makes rivals cheaper to qualify.

## What would prove this wrong

If, by the time SK hynix reports its fourth quarter of 2027, TrendForce, IDC or Counterpoint report SK hynix's share of HBM revenue below 45% for two quarters in a row, or a named NVHBM program is reported as sourced mainly from Samsung or Micron, then the yield and stacking lead I say the standard leaves to SK hynix isn't holding.

_Status as of October 20, 2026: **holding**._

---

SK hynix price: (live) · SK hynix composite: 72 · Share of HBM revenue: 56.4% · Q3 2026 results: Oct 28

[SK hynix](/stocks/skhynix) reports its third quarter on October 28 in Seoul, after a month with two pieces of news about memory. On October 10 the Chinese DRAM maker CXMT said it had a denser chip coming, and SK hynix fell about 9% intraday on October 13. Seoul Economic Daily put its fall over the week at 24.3%. On October 18 the same paper reported that Nvidia wants all three HBM makers to build the base die of its custom memory on TSMC's 3nm process. SK hynix closed slightly higher on October 19, the first session after that report, and MoneyToday's market story put the morning drop down to the US AI selloff without mentioning Nvidia. I think the second story is the one that changes what SK hynix sells, and I'm writing this before the print so the claims below can be checked against it.

## Who would design and build each part of a custom HBM stack

An HBM stack is a pile of DRAM dies sitting on a base die, the logic chip at the bottom that connects the stack to the accelerator. On August 26 Nvidia announced NVHBM, which moves Nvidia's memory controller from the accelerator into that base die. It's offered to companies building their own accelerators through NVLink Fusion, with Amazon's Annapurna Labs named first, and Nvidia says it's built on the same technology it will use for future GPUs. Tom's Hardware notes it isn't in the Vera Rubin systems already shipping and isn't a replacement for standard HBM. Counterpoint's summary is that Nvidia designs the base die and the memory maker builds it to order. Then on October 18 came the report about how Nvidia wants it built.

**How the three suppliers make an HBM stack today, and what Nvidia has asked for NVHBM. As reported by the companies and the trade press on the dates in the sources.**

| Layer | SK hynix today | Samsung today | Micron today | NVHBM, as requested by Nvidia |
| --- | --- | --- | --- | --- |
| Base die design | SK hynix | Samsung, sold turnkey with the memory | Micron | Nvidia |
| Base die manufacture | TSMC 12nm for HBM4, under a 2024 MOU; TSMC 3nm prioritised for HBM4E and custom HBM | Samsung Foundry 4nm for standard HBM4 and HBM4E | In-house for HBM4; an external foundry for HBM4E | TSMC 3nm. Samsung has agreed; SK hynix and Micron have been asked |
| DRAM core dies | SK hynix's own node | Samsung's own node | 1β for HBM4, 1γ for HBM4E | 1c DRAM, made by each supplier |
| Stacking and packaging | SK hynix | Samsung | Micron | Each supplier |
| Other custom base-die designs | Named collaborator on Marvell's custom HBM (Dec 2024) | Named collaborator on Marvell's custom HBM; turnkey custom HBM for AMD | Named collaborator on Marvell's custom HBM | Nvidia's, for NVLink Fusion partners |

_Figures as of October 19, 2026. Source: [TrendForce report on SK hynix's MOU with TSMC for the HBM4 base die](https://www.trendforce.com/news/2024/04/19/news-sk-hynix-partners-with-tsmc-to-strengthen-hbm-technological-leadership/); [NVIDIA's NVHBM Reshapes Base Die Strategies: How Micron, Samsung Are Bringing TSMC Into the Fold](https://www.trendforce.com/news/2026/10/09/news-nvidias-nvhbm-reshapes-memory-makers-base-die-strategies-as-micron-samsung-turn-to-external-foundries/); [Samsung to Use TSMC for Base Die in Nvidia's Custom HBM](https://en.sedaily.com/finance/2026/10/18/samsung-to-use-tsmc-for-base-die-in-nvidias-custom-hbm); [The Trillion-Dollar Bottleneck: NVIDIA's NVHBM and the Fight for the Base Die](https://counterpointresearch.com/en/insights/the-trillion-dollar-bottleneck-nvidias-nvhbm-fight-for-base-die-blog); [NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory](https://blogs.nvidia.com/blog/nvlink-fusion-nvhbm-custom-high-bandwidth-memory/); [Nvidia custom NVHBM promises 30% higher bandwidth, 15% lower power than commodity HBM4e](https://www.tomshardware.com/pc-components/dram/nvidia-custom-nvhbm-promises-30-percent-higher-bandwidth-15-percent-lower-power-than-commodity-hbm4e-custom-base-die-and-phy-will-be-available-to-nvlink-fusion-partners); [Nvidia Co-Designs Custom HBM With Samsung, TSMC to Widen Bandwidth 30%](https://en.sedaily.com/finance/2026/10/19/nvidia-co-designs-custom-hbm-with-samsung-tsmc-to-widen); [Marvell Announces Breakthrough Custom HBM Compute Architecture to Optimize Cloud AI Accelerators](https://www.marvell.com/company/newsroom/marvell-announces-breakthrough-custom-hbm-compute-architecture.html)._

Read the last column against the first three. Nvidia draws the base die, TSMC would make it, and Nvidia has named the DRAM node the core dies should use. What stays with the supplier is making 1c DRAM dies at a good yield, stacking them, and selling the stack. The industry sources Seoul Economic Daily quoted gave two reasons for the common process: getting the same performance from one design whichever supplier builds it, and not having to validate each supplier's chips separately.

The custom base die was the memory makers' plan for getting out of the commodity business. If each big accelerator buyer wanted its own base die, the memory maker that designed it would be tied into that buyer's roadmap, the way a foundry is tied in by a tape-out. SK hynix's stock page builds its bull case partly on that, with custom-base-die HBM keeping it first to qualify at Nvidia and the hyperscalers. It was contested before NVHBM. In December 2024 Marvell announced its own custom HBM architecture, with the HBM support logic moved onto the base die and SK hynix, Samsung and Micron named as collaborators. NVHBM puts Nvidia behind the same idea. TrendForce, citing The Elec, reports that Micron's COO still expects NVHBM to carry a higher selling price and better margins, so I don't read this as a margin hit.

I think it costs SK hynix less than its two rivals, because it's the closest to already working this way. It agreed with TSMC in 2024 to move its HBM4 base die off its own process, ships it on TSMC 12nm now, and TrendForce reports it was already prioritising TSMC 3nm for HBM4E and custom HBM. Samsung, which has made both the memory and the base die for all its HBM, is giving that up for Nvidia's part and keeping it for AMD and for its own standard HBM. Micron makes its HBM4 base die in-house and, again per TrendForce citing The Elec, is moving the HBM4E and NVHBM base die to an outside foundry. Under NVHBM the two challengers end up on a supply chain close to the one SK hynix already runs.

> **The test I'd apply** — When a big buyer standardizes a layer across its suppliers, I'd stop asking who designs that layer best and ask what's left that the buyer can't specify. Here that's DRAM yield on the named node, stacking, and how much of next year's output is sold under contract. A supplier's lead now has to live in those.

## What CXMT actually claimed

CXMT's claim was about commodity DRAM. Its president said in Shanghai that the company had made key progress on a 4F² cell and would launch server memory on it by the end of 2026, with DDR5 server modules expected to follow. A 4F² cell takes about a third less area than the 6F² design the industry uses now. CXMT hasn't published a speed, a yield, a die size or a customer, and a former SK hynix researcher quoted by JoongAng Daily said it remains questionable whether CXMT can produce enough chips to have a meaningful impact on the market. CXMT does make HBM. JoongAng reported in September that it had started low-volume HBM3E production, a generation behind the HBM4 the three incumbents are shipping, with Alibaba's T-Head and Cambricon testing it. That's a real entrant for Chinese accelerators, but it isn't what the October announcement was about.

## Who gains when the base die is common

**The three HBM makers, the foundry that would make every NVHBM base die, and the two companies offering a custom HBM base die. Scores are live and recompute against the current price.**

| Ticker | Name | Moat | Growth | Valuation | Composite | Note |
| --- | --- | --- | --- | --- | --- | --- |
| **Build the stack** |  |  |  |  |  |  |
| [000660.KS](https://www.investmoat.com/stocks/skhynix) | SK hynix | 71 | 73 | 71 | 72 | HBM4 base die already on TSMC 12nm; about ten long-term agreements, typically five years with deposits; no backlog or coverage figure |
| [005930.KS](https://www.investmoat.com/stocks/samsung) | Samsung Electronics | 55 | 68 | 84 | 68 | First to agree to TSMC 3nm for NVHBM; keeps its own 4nm base die for standard HBM4 and HBM4E and turnkey custom HBM for AMD |
| [MU](https://www.investmoat.com/stocks/micron) | Micron Technology | 67 | 73 | 76 | 73 | Moving the HBM4E and NVHBM base die to an outside foundry; take-or-pay agreements with disclosed revenue coverage |
| **Makes the base die** |  |  |  |  |  |  |
| [TSM](https://www.investmoat.com/stocks/tsm) | Taiwan Semiconductor Manufacturing | 94 | 86 | 72 | 87 | Would make the base die of every NVHBM stack, whichever memory maker wins the order |
| **Draw a custom base die** |  |  |  |  |  |  |
| [NVDA](https://www.investmoat.com/stocks/nvda) | NVIDIA Corp. | 81 | 91 | 74 | 85 | NVHBM through NVLink Fusion; Amazon's Annapurna Labs named first |
| [MRVL](https://www.investmoat.com/stocks/mrvl) | Marvell Technology | 70 | 83 | 76 | 78 | Custom HBM architecture announced December 2024 with all three memory makers as collaborators |

_Valuation and composite above use each asset's static valuation score; the live site recomputes them against the current market price._

The clearest winner is [TSMC](/stocks/tsm). If SK hynix and Micron follow Samsung, every NVHBM stack carries a TSMC die whichever memory maker wins the order. That's one more layer of the accelerator package running through TSMC, next to the logic die and CoWoS. [Nvidia](/stocks/nvda) gets something different. Counterpoint's argument is that NVHBM is less a memory product than a way to sell Nvidia's memory architecture to the companies building alternatives to its GPUs. [Marvell](/stocks/mrvl) makes the same offer to its own custom-chip customers, so the companies competing to draw the base die are the accelerator designers, and the memory makers build to whichever drawing their customer picks.

**Scale, proprietary data and transaction embedding across the cohort. For the three memory makers these are the pillars that carry whatever lead survives a common base die. Statuses are read live from each stock's analysis.**

| Ticker | Scale Economics | Proprietary Data | Transaction Embedding |
| --- | --- | --- | --- |
| **Build the stack** |  |  |  |
| [000660.KS](https://www.investmoat.com/stocks/skhynix) | strong | intact | intact |
| [005930.KS](https://www.investmoat.com/stocks/samsung) | intact | intact | intact |
| [MU](https://www.investmoat.com/stocks/micron) | intact | intact | strong |
| **Makes the base die** |  |  |  |
| [TSM](https://www.investmoat.com/stocks/tsm) | strong | strong | strong |
| **Draw a custom base die** |  |  |  |
| [NVDA](https://www.investmoat.com/stocks/nvda) | strong | intact | strong |
| [MRVL](https://www.investmoat.com/stocks/mrvl) | intact | na | strong |

SK hynix has one pillar rated strong, scale economics, and the note behind it is about volume: more HBM shipped means more yield learning per wafer. That's the layer NVHBM leaves with the supplier. Proprietary data is intact, and what it rates, MR-MUF packaging know-how and HBM yield-learning data, is on the same side of the line. Transaction embedding is intact rather than strong. SK hynix said on its July call that it has long-term agreements with about ten customers, typically around five years with deposits and pricing that varies by customer, but unlike [Micron](/stocks/micron) it hasn't put a figure on the backlog or on how much revenue the agreements cover. I argued in the [Micron piece](/research/micron-and-the-contract-at-the-top-of-the-memory-cycle) that the contract is the moat a memory maker can actually own. NVHBM takes away one of the other places a lead could live, which makes that point matter more for SK hynix.

## Where I could be wrong

The strongest objection is that NVHBM is aimed at exactly the edge SK hynix has. The reason Seoul Economic Daily's sources gave for a common base die and a common DRAM node is to stop validating each supplier separately. Being Nvidia's longest-standing HBM partner is how SK hynix built its share, and TrendForce wrote in February that it was likely to keep an edge in bit allocation because of that partnership. A standard that makes the second and third supplier cheap to qualify lets Nvidia move volume between them on price. The TSMC point cuts the same way. SK hynix's 2024 deal with TSMC was a head start, and NVHBM hands the same TSMC access to Samsung and Micron. It also moves SK hynix from 12nm to 3nm, so it isn't simply already working this way. Read like that, NVHBM costs SK hynix the most, because it had the most of the advantage being standardized away.

Samsung is where that objection looks strongest. According to its [stock page](/stocks/samsung), it cleared Nvidia's HBM4 qualification in the first quarter of 2026, is ramping it, and is the largest memory maker by capacity. It was also the first to agree to Nvidia's TSMC request. It also keeps its turnkey route, memory and base die from one company, for AMD and others, so it can sell to both kinds of buyer. If a common drawing turns HBM into a contest of who can supply the most qualified bits fastest, Samsung's capacity matters more than SK hynix's head start, and SK hynix's stock page already calls its scale lead narrowing as Samsung ramps HBM4.

The second objection is that I'm dismissing CXMT without data, the same way the market reacted to it without data. Commodity DRAM sets the price of every wafer a memory maker could turn into HBM instead, so a denser Chinese DRAM cell would press on SK hynix's margins even if CXMT never sells a stack outside China. SK hynix held 29.1% of DRAM revenue in the first quarter of 2026, second in the market, according to IDC as cited in its F-1, so it has plenty of commodity exposure. CXMT is also already making HBM3E, and a Chinese accelerator market that buys Chinese memory is volume the Korean makers don't get back. If CXMT's server modules ship with competitive specifications, the October 13 drop will look early rather than wrong.

The third is that neither story matters for years. NVHBM has no announced ship date, Annapurna is the only named customer, and Nvidia's own Vera Rubin systems use standard HBM. Seoul Economic Daily cites a UBS forecast of 2027 HBM demand up 91% against supply that falls about 10% short. In a shortage that size every supplier sells everything it makes, and who draws the base die decides very little about the next several quarters. If that's right, October 28 is a pricing story and this article is early.

## Why I still think the yield gap matters more

I take the qualification objection most seriously, and my answer to it is the 1c node. Nvidia can name the node, but each supplier still makes those DRAM dies in its own fabs and stacks them on its own lines, and a common drawing lowers the cost of qualifying a supplier without lowering the cost of yielding one. That gap is what SK hynix's scale pillar rates, though I'm assuming its lead carries over to 1c HBM, and nothing public measures yield on it by supplier yet. The same assumption is my answer on Samsung: its capacity only turns into qualified HBM at the rate it yields. So I'd watch share and margin, not who draws the base die. If SK hynix keeps around half of HBM revenue as Samsung ramps and NVHBM arrives, the standard took away the custom-memory upside but not the lead. If its share slides while Samsung and Micron build to the same drawing, the objection was right. On CXMT, I'd treat the October claim as unproven until there's a product with specifications, and its HBM3E as a China-only volume question for now. The contract question is the one SK hynix can answer itself.

## What I'll check in the results

**What I'm checking, written down before the third-quarter results on October 28 (call date per SK hynix's October 14 filing). Prior-quarter figures as reported by SK hynix on July 29, 2026, unless noted.**

| What I'll look at | Last time | What it would tell me |
| --- | --- | --- |
| Long-term agreements | About ten customers, typically around five years with deposits; no backlog, coverage or deposit amount disclosed | Whether SK hynix will put a number on the moat that matters more after NVHBM |
| HBM share | 56.4% of HBM revenue in Q1 2026 (IDC, cited in the F-1); TrendForce expected SK hynix to keep an edge in bit allocation (Feb 2026) | Whether the lead holds as Samsung ramps HBM4 |
| HBM4 ramp | Mass shipments began in Q2, ramping in the second half | Whether the shipping lead, and the yield learning behind it, is holding |
| Base die and NVHBM | Not mentioned in the Q2 release; TSMC 3nm reported as prioritised for HBM4E | Whether SK hynix confirms TSMC 3nm for NVHBM and names any custom HBM customer of its own |
| Commodity DRAM and China | Q2 operating margin 76%; CXMT reported on Sept 1 to have begun small-scale HBM3E production | Whether management describes pricing pressure from Chinese DRAM or treats CXMT as noise |
| New capacity | Yongin Phase 1 cleanroom opening early 2027 | Whether capacity arrives in time to keep share as Samsung ramps |
| Share price reaction | Slightly higher on Oct 19, the first session after the NVHBM base-die report, with no mention of it in the market coverage | Whether investors price NVHBM at all once SK hynix is asked about it |

_Figures as of October 19, 2026. Source: [SK hynix Announces 2Q26 Financial Results](https://news.skhynix.com/en/q2-2026-business-results/); [SK hynix FY26 Q2 earnings call summary (Dolphin Research)](https://longportapp.cn/news/294135299); [SK hynix Inc. Form F-1 registration statement](https://www.sec.gov/Archives/edgar/data/0002120882/000119312526280172/d32785df1.htm); [SK hynix Form 6-K: third-quarter 2026 earnings conference call](https://www.sec.gov/Archives/edgar/data/0002120882/000119312526420934/d413172d6k.htm); [NVIDIA's NVHBM Reshapes Base Die Strategies: How Micron, Samsung Are Bringing TSMC Into the Fold](https://www.trendforce.com/news/2026/10/09/news-nvidias-nvhbm-reshapes-memory-makers-base-die-strategies-as-micron-samsung-turn-to-external-foundries/); [HBM4 Validation Expected in 2Q26; Three Major Suppliers Poised to Shape NVIDIA Supply Landscape](https://www.trendforce.com/presscenter/news/20260213-12929.html); [CXMT's next-gen DRAM plans send Samsung, SK hynix and Micron shares lower](https://www.koreajoongangdaily.com/business/global-chip-stocks-tumble-as-chinas-cxmt-unveils-nextgeneration-dram-plans/12917865); [China's CXMT begins HBM3E production as chip battle with Korean powerhouses widens](https://www.koreajoongangdaily.com/business/chinas-cxmt-begins-hbm3e-production-as-chip-battle-with-korean-powerhouses-widens/12854260); [Samsung Electronics, SK hynix rebound on foreign, institutional buying; Kospi closes up 1.8%](https://www.mt.co.kr/en/stock/2026/10/19/2026101915563310038)._

> **What would break this** — In order of how likely I think it is. Samsung or Micron is reported as the main supplier of a named NVHBM program, or SK hynix's HBM revenue share drops below 45%, which would mean a common base die did make the stacks interchangeable. SK hynix keeps its contracted backlog and revenue coverage undisclosed through 2027 while Micron's coverage grows, which would leave its one contract moat unproven. CXMT's 4F² server DRAM is qualified at a large cloud buyer in 2027 with competitive specifications, which would mean the October selloff was right about the threat and I was wrong to treat it as unproven.

I don't think either October story means SK hynix's lead is over. CXMT's October claim wasn't about HBM, and its HBM3E is a small-volume product for Chinese accelerators so far. NVHBM moves Samsung and Micron onto a supply chain close to the one SK hynix already uses. What NVHBM does end is the hope that the memory makers would own the custom layer of the stack, which leaves SK hynix competing on yield, stacking and contracts. The first two are what its scale lead rests on. It hasn't put a number on the third, and October 28 is the next chance for it to.

## Sources

1. [SK hynix Announces 2Q26 Financial Results](https://news.skhynix.com/en/q2-2026-business-results/) — SK hynix, July 29, 2026 (press-release)
2. [SK hynix FY26 Q2 earnings call summary (Dolphin Research)](https://longportapp.cn/news/294135299) — Longbridge, July 29, 2026 (third-party)
3. [SK hynix Inc. Form F-1 registration statement](https://www.sec.gov/Archives/edgar/data/0002120882/000119312526280172/d32785df1.htm) — SK hynix (SEC filing), June 24, 2026 (filing)
4. [SK hynix Form 6-K: third-quarter 2026 earnings conference call](https://www.sec.gov/Archives/edgar/data/0002120882/000119312526420934/d413172d6k.htm) — SK hynix (SEC filing), October 14, 2026 (filing)
5. [TrendForce report on SK hynix's MOU with TSMC for the HBM4 base die](https://www.trendforce.com/news/2024/04/19/news-sk-hynix-partners-with-tsmc-to-strengthen-hbm-technological-leadership/) — TrendForce, April 19, 2024 (third-party)
6. [NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory](https://blogs.nvidia.com/blog/nvlink-fusion-nvhbm-custom-high-bandwidth-memory/) — NVIDIA, August 26, 2026 (company-site)
7. [Nvidia custom NVHBM promises 30% higher bandwidth, 15% lower power than commodity HBM4e](https://www.tomshardware.com/pc-components/dram/nvidia-custom-nvhbm-promises-30-percent-higher-bandwidth-15-percent-lower-power-than-commodity-hbm4e-custom-base-die-and-phy-will-be-available-to-nvlink-fusion-partners) — Tom's Hardware, August 26, 2026 (third-party)
8. [The Trillion-Dollar Bottleneck: NVIDIA's NVHBM and the Fight for the Base Die](https://counterpointresearch.com/en/insights/the-trillion-dollar-bottleneck-nvidias-nvhbm-fight-for-base-die-blog) — Counterpoint Research, September 1, 2026 (third-party)
9. [Marvell Announces Breakthrough Custom HBM Compute Architecture to Optimize Cloud AI Accelerators](https://www.marvell.com/company/newsroom/marvell-announces-breakthrough-custom-hbm-compute-architecture.html) — Marvell, December 10, 2024 (press-release)
10. [NVIDIA's NVHBM Reshapes Base Die Strategies: How Micron, Samsung Are Bringing TSMC Into the Fold](https://www.trendforce.com/news/2026/10/09/news-nvidias-nvhbm-reshapes-memory-makers-base-die-strategies-as-micron-samsung-turn-to-external-foundries/) — TrendForce, October 9, 2026 (third-party)
11. [Samsung to Use TSMC for Base Die in Nvidia's Custom HBM](https://en.sedaily.com/finance/2026/10/18/samsung-to-use-tsmc-for-base-die-in-nvidias-custom-hbm) — Seoul Economic Daily, October 18, 2026 (third-party)
12. [Nvidia Co-Designs Custom HBM With Samsung, TSMC to Widen Bandwidth 30%](https://en.sedaily.com/finance/2026/10/19/nvidia-co-designs-custom-hbm-with-samsung-tsmc-to-widen) — Seoul Economic Daily, October 19, 2026 (third-party)
13. [CXMT's next-gen DRAM plans send Samsung, SK hynix and Micron shares lower](https://www.koreajoongangdaily.com/business/global-chip-stocks-tumble-as-chinas-cxmt-unveils-nextgeneration-dram-plans/12917865) — Korea JoongAng Daily, October 13, 2026 (third-party)
14. [China's CXMT begins HBM3E production as chip battle with Korean powerhouses widens](https://www.koreajoongangdaily.com/business/chinas-cxmt-begins-hbm3e-production-as-chip-battle-with-korean-powerhouses-widens/12854260) — Korea JoongAng Daily, September 1, 2026 (third-party)
15. [Samsung Electronics, SK hynix rebound on foreign, institutional buying; Kospi closes up 1.8%](https://www.mt.co.kr/en/stock/2026/10/19/2026101915563310038) — MoneyToday, October 19, 2026 (third-party)
16. [HBM4 Validation Expected in 2Q26; Three Major Suppliers Poised to Shape NVIDIA Supply Landscape](https://www.trendforce.com/presscenter/news/20260213-12929.html) — TrendForce, February 13, 2026 (third-party)

---

Scores are computed deterministically from each asset's data file: composite = Moat^0.40 × Growth^0.30 × Valuation^0.30, scaled to 0–100.

Full analyses for every name above: https://www.investmoat.com/stocks. Site index for agents: https://www.investmoat.com/llms.txt

InvestMoat is an open-source research and education framework. Nothing here is financial advice.