NVIDIA's next-generation AI accelerator Vera Rubin's high-bandwidth memory supply landscape is taking shape, with Samsung Electronics and SK Hynix both selected as HBM4 suppliers, and the shipping time window may have opened.
On Monday, according to industry sources cited by TrendForce, a market research firm, the two South Korean memory giants have been included in NVIDIA's component supplier list for Vera Rubin. As the production cycle for HBM4 from DRAM wafers to final packaging exceeds six months, both companies are expected to start mass production as early as this month. This timing coincides closely with NVIDIA's GTC conference scheduled for next week, and the market is highly focused on the HBM4 supply dynamics for Vera Rubin.
In terms of supply share, sources expect SK Hynix to take on more than half of NVIDIA's total HBM supply, including HBM3E, by 2026, while Samsung is expected to dominate the HBM4 business specifically for Vera Rubin.
According to TrendForce's forecast, SK Hynix will continue to lead the industry with a 50% share of global HBM bit output, although this is a decline from 59% in 2025; Samsung's share is expected to rise from 20% to 28%, showing a clear upward trend.
<h2>Technical Threshold: NVIDIA Sets Exceeding Performance Requirements</h2>
The core dividing line in the HBM4 supply competition lies in data transmission rates. According to the Korea Economic Daily, NVIDIA has set a data rate requirement of over 10Gb/s for the HBM4 used in Vera Rubin, far exceeding the JEDEC industry standard limit of 8Gb/s.
In terms of qualification progress, Samsung has successfully passed NVIDIA's two-tier qualification tests for HBM4—corresponding to data rates of 10Gb/s and 11Gb/s. In contrast, SK Hynix is still optimizing its products to pass the higher-level test at 11Gb/s. Analysts point out that the data processing speed of HBM4 may become a key differentiating factor between the two suppliers.
In terms of product specifications, HBM4 uses an architecture of 8 to 16 layers of DRAM chips stacked on a control base. NVIDIA's Vera Rubin is expected to feature 16 HBM4 stacks, achieving a total capacity of 576GB, surpassing the 432GB limit supported by AMD's upcoming MI450, thus gaining an advantage in the memory specification competition for flagship AI accelerators.
<h2>Supplier Landscape: Micron Positioned in Mid-range, SK Group Executives Attend GTC</h2>
Among the three major HBM suppliers, Micron's positioning is somewhat different. According to the Korea Economic Daily, Micron has not completely exited the HBM4 market and is expected to provide HBM4 for mid-range AI accelerators aimed at inference scenarios (such as Rubin CPX), rather than the flagship version Vera Rubin.
Notably, SK Group Chairman Chey Tae-won will personally attend NVIDIA's GTC conference for the first time. Industry sources say that his presence underscores SK Group's high level of importance, and Chey Tae-won will directly participate in overseeing SK Hynix's supply progress arrangements and negotiations
In terms of shipment progress, Samsung Electronics began shipping HBM4 in February this year, while SK Hynix has yet to announce its delivery plans, raising market concerns about the time gap between the two.
<h2>Samsung's Leverage: Ordinary DRAM Price Increases Reshape Supplier Negotiation Landscape</h2>
The significant rise in ordinary DRAM prices is changing the strategic considerations of HBM suppliers and providing Samsung with additional negotiation leverage. TrendForce pointed out that since the fourth quarter of 2025, ordinary DRAM prices have surged sharply, narrowing the historical profitability advantage of HBM. As a result, memory manufacturers are beginning to readjust the capacity allocation between HBM and ordinary DRAM to balance overall revenue growth and customer commitments.
Against this backdrop, NVIDIA's excessive reliance on a single supplier is becoming increasingly evident—if the supplier shifts capacity towards ordinary DRAM, the ramp-up progress of the Rubin platform will face constraints.
Currently, server-grade ordinary DRAM modules (such as SOCAMM2) are priced at approximately $1.30 per Gb, nearing the price level of NVIDIA's flagship HBM3E. This means that for Samsung, producing ordinary DRAM without the expensive stacking process may not be inferior to HBM4 in terms of profit margins. Therefore, by simultaneously mastering the capacities of HBM4 and ordinary DRAM, Samsung is able to present NVIDIA with diversified pricing options, taking a more proactive position in supply negotiations

NVDA

SSNGY

SOXL

The memory competition for NVIDIA's next-generation AI ace, Vera Rubin, has already begun—Samsung and SK Hynix have both made it onto the HBM4 supplier list, with mass production expected to start as early as this month. In terms of technical thresholds, NVIDIA has set a requirement for a transmission rate of 10Gb/s that exceeds industry standards by 25%. Samsung has taken the lead by passing dual-level certification, while SK Hynix is still catching up

- NVIDIA's next-generation AI accelerator, Vera Rubin, is shaping its HBM4 memory supply landscape with Samsung and SK Hynix set to supply high-bandwidth memory (HBM) starting this month, coinciding with the GTC conference.  
- The competition centers on data transfer rates, with NVIDIA requiring over 10Gb/s for HBM4 while Samsung has passed qualification tests at 10Gb/s and 11Gb/s, giving it an edge over SK Hynix.  
- Rising standard DRAM prices have altered the strategic considerations for HBM suppliers, allowing Samsung to negotiate better terms with NVIDIA amid concerns over dependency on single suppliers.

Wallstreetcn

NVIDIA

Samsung Electronics 

Direxion Semicon Bull 3X

Spdr Select Tech

GraniteShares 2x Long NVDA Daily ETF

Invesco Semiconductors ETF

Direxion Daily NVDA Bull 2X Shares

First Trust Nasdaq Food & Semicon

VanEck Semiconductor ETF

T-Rex 2X Long NVIDIA Daily Target ETF

SPDR S&P Semicon

T-Rex 2X Inverse NVIDIA Daily Target ETF

iShares Semiconductor ETF

XL2CSOPNVDA

XI2CSOPNVDA

YieldMax NVDA Option Income Strategy ETF

Direxion Daily NVDA Bear 1X ETF

Samsung and SK Hynix have been selected as NVIDIA's Rubin HBM4 suppliers, with shipments expected to begin in March