---
title: "SK Hynix fell 10% by midday on an unverified Nvidia HBM4 rumor the shipment record contradicts"
description: "Korea's chip complex fell 7% on a Chinese memory announcement that has shipped nothing, while the HBM4 despec rumor behind SK Hynix's slide points the wrong way against the shipment record."
publisher: "The Inference"
section: "Silicon"
published: 2026-10-13T15:30:40.775Z
modified: 2026-10-13T15:30:40.775Z
canonical: https://theinference.org/article/sk-hynix-fell-10-by-midday-on-an-unverified-nvidia-hbm4-rumor-the-shipment-record-contradicts
language: en
keywords: "Semiconductors, Memory, Nvidia, South Korea, Markets"
---

# SK Hynix fell 10% by midday on an unverified Nvidia HBM4 rumor the shipment record contradicts

> Korea's chip complex fell 7% on a Chinese memory announcement that has shipped nothing, while the HBM4 despec rumor behind SK Hynix's slide points the wrong way against the shipment record.

## Two fears went into one basket on Tuesday

Seoul's Kospi [fell more than 7% intraday on October 13, its lowest level since July 31](https://www.businessinsider.com/kospi-stock-index-korea-chip-stocks-china-cxmt-ai-trade-2026-10), with [Samsung Electronics](https://theinference.org/markets/companies/samsung) down 9% and [SK Hynix](https://theinference.org/markets/companies/sk-hynix) down 10% by midday [2]. The day before, [CXMT's memory announcement](https://theinference.org/markets/companies/cxmt) had already taken Samsung down 3.6% and SK Hynix 3.4% in won terms [6]. SanDisk [slid 2.8% premarket](https://finance.yahoo.com/markets/stocks/articles/sandisk-shares-fall-2-8-105400456.html) even though the Chinese advance concerns DRAM, not the NAND market it sells into [3]. One basket, two very different fears, and only one of them has shipped anything.

## What China actually announced

On October 10 at a Shanghai conference, CXMT president Cao Kanyu unveiled progress on a 4F² DRAM cell, a vertical-gate transistor layout that packs roughly 30% to 50% more density into the same silicon area, and promised [a DDR5 server RDIMM by year-end](https://www.trendforce.com/news/2026/10/12/news-cxmt-accelerates-4f%c2%b2-dram-push-eyes-year-end-server-ddr5-rdimm-launch/) [5][7]. That is the whole announcement. No speeds, no capacities, no named customers, no volume commitments; Korean reporting suggests the first products may only partly use 4F², and [volume production is a separate hurdle](https://theinference.org/article/china-s-cxmt-hired-27-samsung-and-sk-hynix-engineers-at-up-to-double-pay-26-are-already-gone) [5]. Cantor Fitzgerald called the reaction a case of "the headline is worse than reality" [3]. The scoreboard says the same: CXMT holds 9.5% of global DRAM revenue, fourth place, against Samsung's 39.4% [7].

*Chart: **CXMT is the number four in DRAM at 9.5% of revenue, not yet the incumbent killer** Column chart of global DRAM revenue share in Q2 2026: Samsung 39.4%, SK hynix 24.9%, Micron 23.3%, CXMT 9.5%, with CXMT highlighted.*

*CXMT is the number four in DRAM at 9.5% of revenue, not yet the incumbent killer [7]*

## What the HBM4 record shows while the rumor mill spins

The rumor behind SK Hynix's slide says [Nvidia](https://theinference.org/markets/companies/nvidia) cut HBM4 stacks from 12-high to 8-high. Here is the documented sequence instead. The JEDEC HBM4 standard sat at 8 gigabits per pin; SK Hynix designed for 10 on its proven MR-MUF bonding process, which it says lifts productivity about 3 times and heat dissipation 2.5 times; Samsung, beaten in HBM3, gambled in 2024 on newer 1c dies over its own 4nm base die [1]. Nvidia then raised Rubin's requirement to 10 gigabits and past 11 in the third quarter of 2025, forcing all three makers to resample, and [Samsung announced mass-production shipments at 11.7 gigabits in February](https://www.damnang.com/p/what-is-happening-in-hbm4) [1]. The gamble bought headroom. But headroom is not share: supply-chain tracker Silicon Analysts reports [SK Hynix mass-producing a 48GB, 16-layer HBM4 stack for Rubin in the third quarter](https://www.msn.com/en-us/technology/hardware-and-devices/sk-hynix-is-shipping-16-layer-hbm4-for-nvidia-rubin-where-do-micron-and-samsung-stand/ar-AA2chUXF), while Samsung's 12-layer HBM4 sits in final qualification [4]. In June, [Jensen Huang](https://theinference.org/markets/people/jensen-huang) named all three suppliers for Rubin, with analysts estimating SK Hynix at 60 to 70% of the volume [8].

*Chart: **Nvidia raised Rubin's HBM4 speed bar twice, above what either maker first designed for** Column chart of HBM4 per-pin speed targets in Gbps: JEDEC standard 8, SK Hynix design 10, Nvidia Rubin requirement over 11 (estimate), Samsung shipped 11.7.*

*Nvidia raised Rubin's HBM4 speed bar twice, above what either maker first designed for [1]*

## The counterpoint, stated fairly

The 12-high reversal claim comes from a reconstruction by Damnang Research, an ex-[AMD](https://theinference.org/markets/companies/amd) engineer's own modeling, and the decisive part sits behind a paywall [1]. [Hock Tan](https://theinference.org/markets/people/hock-tan) warned on Monday that capital, not chips or electricity, could become the AI industry's binding constraint, and that fear is legitimate [2]. CXMT's DRAM share did jump from 7.6% to 9.5% in a single quarter; the direction is real even if the level is not yet [7].

## The dated things that settle it

Three readings: CXMT's year-end RDIMM, 11 weeks out, will show whether volume follows the announcement [5]; Samsung's 12-layer HBM4 qualification result [4]; and SanDisk's October 29 earnings call on whether memory margins hold [3]. The rumor said SK Hynix lost HBM4. The shipment tracker says it is the only one shipping the hardest part of it.

## Takeaway

Korea's chip selloff folded two unrelated fears into one: CXMT announced a DRAM technology it has not yet shipped in volume, and an unverified paywalled model claimed Nvidia cut SK Hynix's HBM4 stack count, even though trackers show SK Hynix is the only maker mass-producing the hardest part of Rubin's HBM4.

## Sources

1. [Damnang Research, What Is Happening in HBM4, 13 October 2026](https://www.damnang.com/p/what-is-happening-in-hbm4)
2. [Business Insider, Kospi is plunging again as AI investors find more reasons to worry, 13 October 2026](https://www.businessinsider.com/kospi-stock-index-korea-chip-stocks-china-cxmt-ai-trade-2026-10)
3. [Yahoo Finance, SanDisk Shares Fall 2.8% as Chinese Memory Chip Advances Raise Competition Concerns, 13 October 2026](https://finance.yahoo.com/markets/stocks/articles/sandisk-shares-fall-2-8-105400456.html)
4. [Benzinga via MSN, SK Hynix is shipping 16-layer HBM4 for Nvidia Rubin: Where do Micron and Samsung stand?, 15 September 2026](https://www.msn.com/en-us/technology/hardware-and-devices/sk-hynix-is-shipping-16-layer-hbm4-for-nvidia-rubin-where-do-micron-and-samsung-stand/ar-AA2chUXF)
5. [Tom's Hardware via MSN, CXMT plans 4F² DDR5 server memory by year-end, 13 October 2026](https://www.msn.com/en-us/technology/hardware-and-devices/cxmt-plans-4f-ddr5-server-memory-by-year-end/ar-AA2e9ybL)
6. [EconoTimes, Samsung, SK Hynix Shares Fall as China's CXMT Advances DRAM Technology, 12 October 2026](https://www.econotimes.com/Samsung-SK-Hynix-Shares-Fall-as-Chinas-CXMT-Advances-DRAM-Technology-1754622)
7. [TrendForce, CXMT Accelerates 4F² DRAM Push, Eyes Year-End Server DDR5 RDIMM Launch, 12 October 2026](https://www.trendforce.com/news/2026/10/12/news-cxmt-accelerates-4f%c2%b2-dram-push-eyes-year-end-server-ddr5-rdimm-launch/)
8. [Tech Times, Nvidia Vera Rubin Enters Full Production: Samsung, SK Hynix, Micron Named HBM4 Suppliers, 2 June 2026](https://www.techtimes.com/articles/317539/20260602/nvidia-vera-rubin-enters-full-production-samsung-sk-hynix-micron-named-hbm4-suppliers.htm)


---

SK Hynix fell 10% by midday on an unverified Nvidia HBM4 rumor the shipment record contradicts — The Inference. Canonical: https://theinference.org/article/sk-hynix-fell-10-by-midday-on-an-unverified-nvidia-hbm4-rumor-the-shipment-record-contradicts