Bernstein Analysis: Behind the $700 Billion Cooperation, What AI Might Lack Most is Not GPUs
- Core Viewpoint: SK Group and NVIDIA, as well as Samsung and Broadcom, recently announced AI supply chain cooperation frameworks totaling over $700 billion. The core objective is to secure HBM memory and advanced packaging resources in advance, rather than representing direct purchase orders. This signals a shift in the AI supply chain from vying for GPUs to抢占 high-end memory and packaging segments.
- Key Elements:
- SK Group and NVIDIA announced comprehensive cooperation exceeding $500 billion, covering AI factories (SK Telecom’s 2GW scale, with the first project planned for launch in 2027) and next-generation memory like HBM4.
- Samsung and Broadcom signed an MOU worth over $200 billion, encompassing HBM memory, 2nm foundry services, and 2.3D/2.5D advanced packaging, aiming to provide an integrated "memory + foundry + packaging" solution.
- These two frameworks are primarily based on MOUs and partnerships. Specific procurement volumes, prices, and delivery schedules have not been disclosed, and they do not constitute firm orders.
- A Bernstein report indicates that AI clients are using long-term frameworks to secure memory supply like HBM in advance. This is because HBM has limited capacity flexibility and long certification cycles, making it a bottleneck for AI server deliveries.
- The annual revenue of the global memory market is projected to be around $1.3 trillion from 2027 to 2028, but the contribution of these framework amounts to incremental growth is difficult to quantify.
- The cooperation between Samsung and Broadcom is expected to have a limited impact on TSMC’s market share, as demand for advanced capacity remains strong. Samsung faces a test of its delivery capability and needs to ensure the coordinated stability of HBM, logic chips, and packaging.
- Bernstein maintains an “Outperform” rating for Samsung, SK Hynix, Micron, NVIDIA, and Broadcom but clarifies that an MOU does not equal an order. Risks include competition from Chinese memory manufacturers and the realization of production capacity.
TL;DR
- SK Group, NVIDIA, Samsung, and Broadcom have successively announced AI supply chain partnerships, with the total framework scale exceeding $700 billion.
- These collaborations cover HBM4, AI factories, foundry production below 2nm, and advanced packaging, primarily aimed at securing AI server supply in advance.
- These arrangements are still mainly partnerships and MOUs; actual procurement volumes, pricing, and delivery schedules have not yet been disclosed.
The AI supply chain is extending from securing GPUs to locking in memory and advanced packaging in advance.
On July 24, SK Group and NVIDIA announced at the AI Summit in San Francisco an expanded comprehensive partnership exceeding $500 billion, covering AI factories and next-generation memory. Subsequently, Samsung Electronics signed an MOU with Broadcom, anticipating a collaboration scale exceeding $200 billion in memory and foundry over the next five years until 2030.
In a report on July 27, Bernstein focused on memory rather than another wave of AI chip orders. AI servers rely on more than just GPU expansion. For computing chips like GPUs and ASICs to truly enter data centers, they require HBM, DRAM, NAND, advanced packaging, and stable delivery. If these links fall behind, shipments from AI chip companies like NVIDIA and Broadcom could also be delayed.
The two massive figures also need to be examined separately. The over $500 billion is not a procurement contract won solely by SK hynix for memory; rather, it represents SK Group's comprehensive partnership with NVIDIA. SK hynix handles the HBM4 and next-generation memory cooperation, while SK Telecom is the main entity for the AI factory. The $200 billion MOU between Samsung and Broadcom is more specific, covering HBM and other memory, support for next-generation AI accelerators, 2nm-related manufacturing, and 2.3D/2.5D integration.
NVIDIA Locks in HBM4; Broadcom Negotiates Memory, Foundry, and Packaging Together
The SK Group-NVIDIA partnership places AI data centers and next-generation memory within the same framework.
According to official information from NVIDIA and SK hynix, the collaboration includes SK Telecom's up to 2GW AI factory, NVIDIA DSX, the Vera Rubin platform, and SK hynix's HBM4. The first AI factory is planned to go live in 2027.
The market implication is straightforward. AI systems post-Vera Rubin will continue to increase demands on HBM and system-level supply. Memory is no longer just a standardized procurement item but a part of whether AI chip platforms can be delivered on schedule.
The Samsung-Broadcom MOU resembles a bundled supply chain solution. Samsung officially stated the cooperation scale over the next five years until 2030 exceeds $200 billion, encompassing memory and foundry, supporting Broadcom's next-generation AI accelerator, and involving advanced packaging related to the 2nm process, along with 2.3D and 2.5D integration.
This involves more than just whether Broadcom will entrust some AI ASICs to Samsung for manufacturing. Samsung aims to package memory, advanced nodes, and advanced packaging for AI chip clients, striving to offer a deliverable solution outside of TSMC.
In technical terms, Samsung's Cube-S, Cube-E/R fall under 2.5D/2.3D Cube packaging. Bernstein compares these to TSMC's CoWoS-S/L/R, with the core goal being high-bandwidth connectivity and multi-chip integration between logic chips and HBM.
Comparison of Samsung Cube-S, Cube-E, Cube-R with TSMC CoWoS-S/L/R, illustrating how 2.5D and 2.3D packaging connect HBM with logic chips.
Memory Industry Annual Revenue Could Reach ~$1.3 Trillion; AI Clients Start Queueing Early
Bernstein believes the most noteworthy aspect of these announcements is that memory supply is being locked in ahead of time.
In the expansion of AI servers, computing chips are certainly important, but the supply elasticity for HBM and high-end memory is smaller, client qualification cycles are longer, and advanced packaging can become a bottleneck for system delivery. Major clients prefer not to wait until capacity is tight to enter the spot market; instead, they secure future supply through multi-year frameworks.
The market consensus cited in the report shows the global memory industry's annual revenue at approximately $0.9 trillion in 2026, and around $1.3 trillion in both 2027 and 2028. TrendForce's previous public forecasts also indicated the global memory market size in 2027 would be about $1.28 trillion. Participants include Samsung, SK hynix, Micron, KIOXIA, and Chinese memory manufacturers.
Consensus bar chart of global memory supplier revenue, approximately $0.9 trillion in 2026, and around $1.3 trillion in both 2027 and 2028.
These figures cannot be directly equated to new revenue generated from the collaborations between SK Group, NVIDIA, Samsung, and Broadcom. The announcements did not disclose specific procurement volumes, product mixes, pricing formulas, delivery schedules, nor how much of the framework amount comes from new capacity versus simply extending existing cooperation.
However, the signal is clear enough. AI clients are treating HBM and advanced packaging as strategic resources, and the bargaining power of leading memory manufacturers is consequently gaining sustained market attention. Traditional memory cycles were more influenced by PC, mobile phone, and server inventory. Under the AI cycle, HBM and advanced packaging are deeply intertwined, making supply-demand dynamics more likely to be determined by a few large clients and a few suppliers.
Samsung Aims for Packaging and Foundry; Impact on TSMC Viewed with Restraint
Given the appearance of 2nm and below processes, advanced packaging, and AI accelerators in the Samsung-Broadcom MOU, the market naturally wonders whether TSMC's market share will be affected.
Bernstein's assessment is relatively restrained. Even if Broadcom moves some AI ASICs to Samsung in the future, the near-term impact on TSMC's profitability might be limited, as demand for advanced capacity remains very tight. Samsung gaining more validation opportunities from AI clients does not mean TSMC immediately loses orders.
The real test for Samsung lies in its delivery capability. AI chip clients require more than just manufacturing capability; they need stable performance across HBM, logic chips, packaging, substrates, yield rates, and delivery schedules. A delay in any link makes it difficult to translate the technological roadmap in the MOU into actual shipments.
Diagram of three TSMC CoWoS variants, showing differences in interposer, RDL, and HBM stacking for CoWoS-S, CoWoS-L, and CoWoS-R.
For Samsung, this represents an opportunity to extend its memory advantages into the AI system-level supply chain. Samsung holds scale in the memory sector, but in terms of HBM leadership and client trust in advanced foundry, the market has historically focused more on SK hynix and TSMC. If the long-term framework with Broadcom is executed smoothly, Samsung can re-emphasize its combined capability of "memory + foundry + packaging."
Ratings Lean Bullish, but MOUs Are Not Orders
In its report, Bernstein maintained Outperform ratings for Samsung Electronics, SK hynix, Micron, NVIDIA, and Broadcom, while rating KIOXIA as Underperform. Regarding target prices, Samsung common stock was set at 440,000 KRW, SK hynix at 3,300,000 KRW, NVIDIA at $315, and Broadcom at $550. Related ratings and target prices are subject to the source report.
Table of key stock ratings and target prices: Samsung 440,000 KRW, SK hynix 3,300,000 KRW, NVIDIA $315, Broadcom $550, all rated Outperform.
This set of ratings does not mean all collaborations have been translated into confirmed performance. A more prudent interpretation is that AI memory supply is becoming more strategic, with both leading memory makers and major AI chip clients using long-term arrangements to mitigate future supply risks.
The risks are also clear. MOUs and partnerships are not final procurement contracts; prices, quantities, and delivery timelines remain undisclosed. Whether the Samsung-Broadcom collaboration translates into substantial AI ASIC foundry or advanced packaging market share depends on client validation, yield rates, and capacity planning. Even with the immense scale of the cooperation, it is difficult to assess how much incremental contribution it can make relative to the roughly ~$1.3 trillion annual memory market around 2027.
Longer-term pressure comes from the competitive landscape of the storage industry. Progress in China's storage industry, especially competition in the NAND segment, will impact industry profit margins. Due to higher barriers in EUV, process technology, and client qualification for DRAM and HBM, short-term pressure is relatively smaller. However, the current supply tightness cannot simply be extrapolated into a permanent advantage.
The most definite signal from this set of collaborations is not that "$700 billion in orders have been secured," but that AI giants are beginning to lock in memory and packaging resources through multi-year frameworks. Whether these can be realized will depend on capital expenditures, HBM shipments, client prepayments, capacity expansion, and actual delivery schedules reported in financial results.
