Is control over memory design shifting to NVIDIA and TSMC? Samsung and SK Hynix risk being relegated to downstream foundries

$NVDA  $META  $GOOGL #AI  #谷歌  #Meta #NvidiaInvestigatesAfterHuang’sAffairApology 
AI chip leader NVIDIA and TSMC have recently moved to reshape the HBM (high-bandwidth memory) supply chain, raising alarm in South Korea’s semiconductor industry. According to a November 9 report by South Korea’s JoongAng Ilbo, as NVIDIA’s influence over HBM design and process selection continues to grow, if South Korean companies fail to retain control of key areas such as base dies, hybrid bonding, and heat dissipation, SK Hynix and Samsung may be left with only the role of “downstream contractors” supplying DRAM cells. The value layer of the memory industry is quietly shifting toward NVIDIA and TSMC.
Report: TSMC Evaluating In-House 3D DRAM Production at Nvidia’s Request? Company Quickly Denies It
According to foreign media reports, TSMC is said to be evaluating production of next-generation 3D DRAM at Nvidia’s request, and may build a new fab or acquire an existing facility and convert it. After the news broke, TSMC said the following day that it would continue to focus on its core foundry business.
Since TSMC is also a key partner in SK Hynix’s HBM production, the statement was interpreted as reaffirming its business principle of “not competing with customers.” However, South Korean industry analysts believe that, in light of Nvidia’s recent series of moves, this incident can hardly be dismissed as an isolated episode.
(Is Nvidia Pushing NVHBM to Drive Down HBM Prices? Analysts Worry Memory Giants Could Lose Bargaining Power)
Nvidia Moves into Memory Design: NVHBM and 3nm Base Dies
In August this year, Nvidia announced NVHBM, a customized memory product that integrates its own memory controller into the Base Die. The Base Die is the chip at the bottom of the HBM stack responsible for memory control. In addition, last month it was reported that Nvidia had asked Samsung Electronics, SK Hynix, and Micron to use TSMC’s 3nm process to manufacture Base Dies.
(Samsung Gives In! Reportedly Accepts Nvidia’s Request to Make NVHBM Chips Using TSMC’s 3nm Process)
For Nvidia, designing its own memory architecture and interface can boost performance while reducing dependence on specific suppliers. For TSMC, since it needs HBM to produce GPUs, increasing its influence over memory manufacturing can reduce the risk of GPU shipments being disrupted and allow it to capture a larger share of the AI chip supply chain.
Scholars Outline Four Paths for TSMC: Capturing High Value-Added Business Without Building DRAM Fabs
Kwon Seok-joon, a professor in the Department of Chemical Engineering at Sungkyunkwan University, explains that TSMC can enter high-value-added areas of the memory industry through four avenues without building DRAM fabs of its own:
Expand Base Die manufacturing
Use hybrid bonding to lead stacking and integration
Partner with Taiwanese DRAM manufacturers
Manufacture peripheral circuits for 3D DRAM
Hybrid bonding is key. As the number of HBM stack layers approaches 20, hybrid bonding—which directly bonds chips without using microbumps—will become increasingly important. If TSMC also takes on this process, then even if SK Hynix continues supplying DRAM core dies, Nvidia and TSMC will still control the design and manufacturing of the finished product, reshaping the balance of power in the supply chain.
A former senior SK Hynix executive warned that as HBM moves beyond 20 layers, if South Korea’s two major memory manufacturers cannot take the lead in Base Dies, bonding, and thermal processing technologies, they may be reduced to downstream contractors that can only supply DRAM cells.
South Korea’s Last Line of Defense: Secure Co-Design Rights and Mass-Produce Hybrid Bonding
Facing pressure from both Nvidia and TSMC, Kwon Seok-joon recommends that South Korean companies make two things their last line of defense: first, become co-designers of Base Dies; second, achieve mass-production capability for hybrid bonding ahead of TSMC. They should also broaden their customer base for customized HBM to include Google, AMD, and Meta, protecting their design autonomy beyond Nvidia.
Kim Yong-seok, dean of the Semiconductor Education Institute at Gachon University, bluntly noted that in business, there are no permanent friends or allies. He urged academia to strengthen early-stage research in 2.5D/3D packaging, HBM interfaces, and high-performance memory system design. The government should also support talent development and infrastructure for integrated logic and memory design.
This article, “Is Memory Design Leadership Shifting to Nvidia and TSMC? Samsung and SK Hynix Risk Being Reduced to Downstream Contractors,” first appeared on .
【Further Analysis】
【My Further Thoughts】
This trend reveals a paradigm shift in the AI hardware supply chain, from assembling “standard components” to “deep customization and advanced packaging.” The core value of HBM has shifted from DRAM storage cells alone to logic control (Base Die) and advanced packaging (hybrid bonding). Nvidia’s leadership in defining structural specifications and TSMC’s role in providing high-value manufacturing processes are reshaping the distribution of chip value. If Korean manufacturers lose their leadership in Base Die design and packaging, memory giants will face the risk of commoditization, and their gross margins could be eroded.
【Market Impact】
* **U.S. stocks:**
  * **Nvidia (NVDA, bullish):** Developing its own NVHBM and controlling the Base Die strengthen its control over the supply chain and lower procurement costs, helping it maintain high gross margins over the long term.
  * **Micron (MU, neutral to bearish):** If bargaining power shifts to IC design and foundry giants, the pricing power of major memory manufacturers may be constrained.
* **Taiwan stocks:**
  * **TSMC (2330/TSM, bullish):** It can secure high-margin orders for 3nm Base Dies and hybrid bonding without taking on the massive capital expenditure of building DRAM fabs, further cementing its position as the leading AI foundry.
  * **Advanced packaging and equipment supply chain (bullish):** As hybrid bonding becomes more important, suppliers of related advanced packaging equipment and materials may see a new wave of orders.
* **Cryptocurrency:**
  * **AI-related tokens (such as RENDER, FET, and NEAR, neutral to bullish):** Improvements in computing efficiency and lower costs for hardware infrastructure will benefit the long-term expansion of decentralized AI computing and application ecosystems.
【What to Watch Next】
1. Before mass production of HBM with more than 20 layers begins, can Samsung and SK Hynix successfully improve hybrid bonding yields and maintain their leadership?
2. How NVHBM progresses, and whether major tech companies such as Google and Meta follow suit by adopting customized HBM solutions.
【Risk Reminder】
If antitrust regulators subsequently scrutinize Nvidia and TSMC for their excessive dominance in the AI supply chain, or if yields for the new hybrid bonding process fail to improve as expected, this could pose short-term risks to technology adoption and cause volatility in related share prices.