TL;DR

• SK Group announced AI supply chain collaboration with Nvidia, Samsung, and Broadcom, with a total framework scale exceeding $700 billion.
• The collaboration covers HBM4, AI factory, manufacturing below 2nm, and advanced packaging, with the core focus on securing AI server supply in advance.
• These arrangements are still primarily based on partnerships and MOUs, with actual purchasing volumes, prices, and delivery schedules yet to be disclosed.

The AI supply chain is expanding from hoarding GPUs to securing memory and advanced packaging in advance.

On July 24, SK Group and Nvidia announced an expanded comprehensive partnership exceeding $500 billion at the AI Summit in San Francisco, covering AI factories and next-generation memory. Samsung Electronics subsequently signed an MOU with Broadcom, expecting a collaboration scale exceeding $200 billion in memory and manufacturing over the next five years until 2030.

Bernstein's report on July 27 focused on memory rather than another round of AI chip orders. AI server expansion does not solely rely on GPUs. For GPU, ASIC, and other computational chips to truly enter data centers, HBM, DRAM, NAND, advanced packaging, and stable delivery are also required. If these segments lag behind, the shipment of AI chip companies like Nvidia and Broadcom will also be slowed down.

The two massive figures need to be understood separately. The over $500 billion is not a memory procurement contract obtained solely by SK Hynix, but a comprehensive partnership between SK Group and Nvidia, with SK Hynix taking on the HBM4 and next-generation memory collaboration, while SK Telecom is the main entity in the AI factory. Samsung and Broadcom's $200 billion MOU is clearer, covering memory such as HBM, next-generation AI accelerator support, manufacturing related to 2nm, and 2.3D/2.5D integration.

Nvidia Secures HBM4, Broadcom Discusses Memory, Manufacturing, and Packaging Together

SK Group's collaboration with Nvidia brings AI data centers and next-generation memory into the same framework.

According to official information from Nvidia and SK Hynix, the collaboration includes up to 2GW AI factory from SK Telecom, NVIDIA DSX, Vera Rubin platform, and SK Hynix HBM4. The first AI factory is planned to be launched in 2027.

The market implication of this is straightforward. AI systems following Vera Rubin continue to raise demands for HBM and system-level supply, making memory not just a standardized procurement item, but a part of whether AI chip platforms can deliver on time.

Samsung and Broadcom's MOU reads more like a bundled supply chain scheme. Samsung officially stated that the collaboration scale over the next five years until 2030 exceeds $200 billion, encompassing memory and manufacturing, supporting Broadcom's next-generation AI accelerator, and involving advanced packaging related to 2nm processes, 2.3D, and 2.5D integration.

This is not just whether Broadcom will give some AI ASIC manufacturing to Samsung. Samsung aims to package memory, advanced processes, and advanced packaging for AI chip clients, seeking to provide another set of deliverable solutions beyond TSMC.

Technologically, Samsung Cube-S and Cube-E/R belong to 2.5D/2.3D Cube packaging. Bernstein compares this to TSMC CoWoS-S/L/R, both focusing on solving the high-bandwidth connection and multi-chip integration between logic chips and HBM.

Comparing Samsung Cube-S, Cube-E, and Cube-R with TSMC CoWoS-S/L/R shows how 2.5D and 2.3D packaging connects HBM with logic chips.

Annual Revenue in Memory Expected to Reach Approximately $1.3 Trillion, AI Clients Begin to Queue Up Early

Bernstein believes that the most noteworthy aspect of this announcement is the advance locking of memory supply.

In the expansion of AI servers, while computational chips themselves are important, the supply elasticity of HBM and high-end memory is much smaller, customer certification cycles are longer, and advanced packaging will also become a limiting factor for overall delivery. Major clients do not want to wait until capacity tightens to enter the spot market, but rather want to secure future supply through multi-year frameworks.

The market consensus cited in the report indicates that the global memory industry's annual revenue is expected to be approximately $0.9 trillion in 2026, and around $1.3 trillion in both 2027 and 2028. Previous predictions by TrendForce also show that the global memory market size will be approximately $1.28 trillion in 2027. Participants include Samsung, SK Hynix, Micron, KIOXIA, and Chinese memory manufacturers.

Global memory supplier revenue consensus bar chart, approximately $0.9 trillion in 2026, around $1.3 trillion in both 2027 and 2028.

These figures cannot be directly equated with the new revenue generated by the collaboration between SK Group, Nvidia, Samsung, and Broadcom. The announcement did not disclose specific procurement volumes, product structures, pricing formulas, or delivery timelines, nor did it clarify how much of the framework amount comes from new capacity or how much is merely the long-term solidification of existing collaborations.

However, the signal is already quite clear. AI clients are starting to view HBM and advanced packaging as strategic resources, and the bargaining power of leading memory manufacturers is gaining sustained market attention. The traditional storage cycle is more influenced by PC, mobile phone, and server inventory, while under the AI cycle, HBM and advanced packaging are deeply tied, with supply-demand relationships more easily determined by a few large clients and a few suppliers.

Samsung Aims to Secure Packaging and Manufacturing, TSMC's Impact is Viewed with Caution

The MOU between Samsung and Broadcom includes processes of 2nm and below, advanced packaging, and AI accelerators, which naturally raises concerns about whether TSMC's market share will be impacted.

Bernstein's judgment is relatively cautious. Even if Broadcom shifts some AI ASICs to Samsung in the future, the recent profit impact on TSMC may be limited because demand for advanced capacity is still very full. Samsung obtaining more opportunities for AI client validation does not mean TSMC will immediately lose orders.

The real test for Samsung is its delivery capability. AI chip clients need not only single-point manufacturing ability but also stable HBM, logic chips, packaging, substrates, yield, and delivery times. If any link slows down, the technology roadmap in the MOU will be difficult to translate into actual shipments.

TSMC CoWoS three variant structural diagrams show the differences in intermediary layers, RDL, and HBM stacking for CoWoS-S, CoWoS-L, and CoWoS-R.

For Samsung, this is an opportunity to extend its storage advantages into the AI system-level supply chain. Samsung has scale in the memory field, but in terms of HBM leadership and advanced manufacturing client trust, the market has previously focused more on SK Hynix and TSMC. If the long-term framework with Broadcom is executed smoothly, Samsung can re-emphasize its "memory + manufacturing + packaging" combination capability.

Ratings Lean Toward Positive, but MOU is Not an Order Yet

Bernstein maintains Outperform ratings for Samsung Electronics, SK Hynix, Micron, Nvidia, and Broadcom in its report, while KIOXIA is rated Underperform. In terms of target prices, Samsung's common stock is at 440,000 KRW, SK Hynix at 3.3 million KRW, Nvidia at $315, and Broadcom at $550. These ratings and target prices are subject to the source report's metrics.

Main stock ratings and target prices table, with Samsung at 440,000 KRW, SK Hynix at 3.3 million KRW, Nvidia at $315, and Broadcom at $550 all rated as Outperform.

This set of ratings does not imply that all partnerships have been converted into definite performance. A more cautious understanding is that the AI memory supply is becoming more strategic, with leading memory suppliers and major AI chip clients using long-term arrangements to mitigate future supply risks.

Risks are also quite clear. MOUs and partnerships are not final procurement contracts, and prices, quantities, and delivery timelines have yet to be disclosed. Whether the collaboration between Samsung and Broadcom can yield substantial AI ASIC manufacturing or advanced packaging share will also depend on client validation, yield rates, and capacity arrangements. Even with a huge collaboration amount, it is difficult to judge how much incremental contribution it can make relative to the annual memory market of about $1.3 trillion around 2027.

Long-term pressures come from competitive dynamics in the storage industry. The progress of the Chinese storage industry, especially in NAND, will impact industry profit margins. DRAM and HBM face relatively lower short-term pressures due to higher EUV, process, and customer certification thresholds, but current supply tightness cannot easily be extrapolated into a permanent advantage.

The most certain signal from this collaboration is not "the $700 billion order has landed," but that AI giants are beginning to lock in memory and packaging resources through multi-year frameworks. Whether this can be fulfilled in the future depends on capital expenditures, HBM shipments, client prepayments, capacity expansion, and actual delivery timelines reflected in financial reports.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。