AI strategic cooperation highlights the importance of memory supply
AI summary card
AI strategic cooperation highlights the importance of memory supply
Bernstein believes the strategic cooperation among Samsung, SK hynix, NVIDIA, and Broadcom mainly reflects long-term locked-in demand in the AI era for HBM, DRAM, and advanced packaging capabilities, and it remains positive on memory leaders as well as NVDA and AVGO.
- The scale of SK hynix's cooperation with NVIDIA is described as exceeding US$500B, covering memory as well as the 2GW Vera Rubin DSX AI Factory planned to go online in 2027.
- Samsung signed an approximately US$200B MOU with Broadcom, covering memory, HBM, foundry, and advanced packaging services over the next five years through 2030.
- The report believes these announcements show that NVIDIA and Broadcom need to secure the critical memory supply required for future AI growth.
- Bernstein expects limited impact on TSMC from the announcements, because TSMC capacity still has substantial queued demand, and even if Samsung or Intel wins some share, it is unlikely to affect TSMC's visible earnings.
- The report remains positive on Samsung, SK hynix, Micron, NVIDIA, and Broadcom, but maintains Underperform on KIOXIA due to China's long-term competitive risk in NAND.
Report interpretation
Overview
This report comments on the cooperation announced at the AI Summit held by the Korean government in San Francisco: Samsung, SK hynix, NVIDIA, and Broadcom announced strategic cooperation around memory, foundry, and AI data centers. Bernstein's core judgment is that the expansion of AI computing is elevating memory supply from a cyclical component issue to a strategic resource issue.
Core views
The report's core views include: first, both NVIDIA's cooperation with SK hynix and Broadcom's MOU with Samsung indicate that HBM, next-generation AI memory, and advanced packaging have become key bottlenecks in the AI supply chain; second, the disclosed amounts are more likely to correspond mainly to memory demand, reflecting NVIDIA's and Broadcom's need to lock in stable supply; third, although it remains unclear whether Broadcom will assign AI ASIC wafers or advanced packaging to Samsung, the impact on TSMC's visible earnings should be minimal; fourth, memory has become more important to AI relative to logic semiconductors, and the recent pullback is viewed as a better entry point; fifth, NAND faces long-term competition risk from China, so KIOXIA has higher risk, while DRAM faces relatively lower competitive threat due to higher EUV barriers.
Analysis framework
The report adopts a combination of event-driven and supply-chain strategic analysis, evaluating the announced amounts, scope of cooperation, AI data center construction plans, advanced packaging technology roadmaps, and consensus industry revenue within the same framework. It does not precisely quantify the earnings contribution of each individual announcement, but instead judges the investment implications through total memory market size, downstream AI growth guidance, and valuation multiples of each company.
Methodology notes
Deriving the target price by multiplying EPS for future quarters or years by a target multiple
The report uses forward 5Q-8Q EPS multiples for Samsung, SK hynix, and Micron, a 1-year forward EPS multiple for KIOXIA, and about 25x forward or formally adjusted EPS multiples for NVDA and AVGO.
Assessing bargaining power in the industry chain through key component supply constraints
The report believes HBM, DRAM, NAND, and advanced packaging are becoming more important in AI demand, and that long-term supply security is the core motivation behind the strategic cooperation by NVIDIA and Broadcom.
Simultaneously assessing demand cycles, supply expansion, pricing environment, and the impact of new competitors
The report focuses on the risk of a favorable pricing environment ending earlier than expected, weaker demand, increased supply, changes in investor sentiment, and the long-term downside risk from China's progress in NAND.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Samsung Electronics Co LtdSupplier of memory, HBM, foundry, and advanced packaging
- Strengths
- The US$200B MOU with Broadcom improves long-term demand visibility, and the company has advanced packaging capabilities such as 2nm and below foundry and Cube-S, Cube-E, and Cube-R.
- Weaknesses
- The report does not confirm whether Broadcom AI ASICs will be produced by Samsung, and the company still faces memory cycle and execution risks.
- Comparison
- Its advanced packaging solutions are compared with TSMC's CoWoS-S, CoWoS-L, and CoWoS-R.
- Risks
- Weaker demand, increased supply, an earlier-than-expected end to the favorable pricing environment, and changes in investor sentiment.
- SK hynix IncLong-term cooperation partner for NVIDIA's next-generation AI memory supply
- Strengths
- The US$500B+ cooperation with NVIDIA strengthens its position in HBM and next-generation AI memory supply, while also benefiting from AI data center construction.
- Weaknesses
- Affected by the memory cycle, capital expenditure, and customer concentration.
- Comparison
- Compared with NAND, DRAM faces lower competitive threat from China due to higher EUV technology barriers.
- Risks
- Demand volatility, supply expansion, changes in the pricing environment, and execution risk.
- Micron Technology IncBeneficiary of the global DRAM and memory upcycle
- Strengths
- The report maintains Outperform, believing AI supports memory demand positively.
- Weaknesses
- Valuation and earnings remain affected by memory cycle volatility.
- Comparison
- Along with Samsung and SK hynix, it belongs to the report's preferred memory leader group.
- Risks
- An early end to the favorable pricing environment, weaker demand, increased supply, and Chinese memory competition.
- NVIDIA CORPAI accelerated computing demand-side player and strategic memory cooperation partner
- Strengths
- The data center opportunity is huge and still at an early stage, and supply cooperation helps ease supply concerns amid strong AI growth.
- Weaknesses
- Future growth is sensitive to the supply chain, export regulation, and the sustainability of customer demand.
- Comparison
- Like Broadcom, it needs to lock in critical AI memory supply.
- Risks
- Short-term business volatility, slower-than-expected revenue growth in key end markets, competitors pressuring share or price, customers shifting to in-house chips, and technology export regulation risk.
- Broadcom IncDemand-side player for AI ASICs, networking chips, and advanced packaging
- Strengths
- Its MOU with Samsung covers memory, HBM, 2nm and below foundry, and advanced packaging; the report believes the AI trajectory in 2026 is strong and may accelerate after 2027.
- Weaknesses
- It remains unclear whether AI ASICs will be produced on Samsung wafers or through Samsung advanced packaging.
- Comparison
- Like NVIDIA, it reduces AI growth supply risk through strategic cooperation.
- Risks
- Unexpected weakness or gaps in AI demand, loss of share or sockets at key customers, failed M&A synergy execution, insufficient cash returns, and unexpected management changes.
- KIOXIA Holdings CorpNAND memory-related asset
- Strengths
- If NAND demand is stronger than expected, pricing improves, Japanese government capex subsidies are obtained, or cost declines are better than expected, upside risk may emerge.
- Weaknesses
- The report assigns Underperform due to the long-term competitive threat from China in NAND.
- Comparison
- Compared with DRAM, NAND faces higher competition risk from China.
- Risks
- Chinese NAND competition, deterioration in the pricing environment, insufficient demand, and cost improvements falling short of expectations.
- TSMCReference asset potentially affected by Samsung's foundry and advanced packaging cooperation
- Strengths
- The report believes TSMC still has queued demand for capacity, so even if some share shifts to Samsung or Intel, visible earnings are unlikely to be materially affected.
- Weaknesses
- If Broadcom shifts more AI ASIC wafers or advanced packaging to Samsung, it could create share disruption.
- Comparison
- Samsung's Cube-S, Cube-E, and Cube-R are respectively compared with TSMC's CoWoS-S, CoWoS-L, and CoWoS-R.
- Risks
- Intensifying competition in high-end advanced packaging and customer supply-chain diversification.
Key data
- Scale of SK hynix-NVIDIA cooperationUS$500B+The cooperation covers memory and the 2GW Vera Rubin DSX AI Factory, planned to go online for SK Telecom in 2027.
- Additional memory supply cooperation of SK GroupUS$250B over next 5 yearsThe report says SK Group will also cooperate with other global technology companies to provide memory supply.
- Samsung and Broadcom MOUUS$200B through 2030Covers memory, HBM, 2nm and below foundry, as well as 2.3D/2.5D advanced packaging.
- Consensus memory industry revenue约 US$1.3T annual revenue in 2027 and 2028Used to help investors understand where the scale of the cooperation stands within the overall industry.
- NVDA AI opportunity guidance约 US$1T across CY25-27The report believes this implies an opportunity size approaching US$500B in the following year.
- AVGO AI opportunity guidanceUS$100B+ next yearThe report believes the supply cooperation helps ease concerns about AVGO's AI growth supply.
- NVDA rating and target priceOutperform, US$315 PTThe report says the data center opportunity is huge and still in the early stage, with substantial upside still possible.
- AVGO rating and target priceOutperform, US$550 PTThe report believes the AI trajectory in 2026 is strong and may accelerate in 2027 and beyond.
Impact & implications
The investment implication is that value distribution across the AI supply chain may continue to concentrate in companies with advanced memory supply capabilities. Samsung, SK hynix, and Micron benefit from AI memory demand and DRAM supply barriers; NVDA and AVGO reduce future supply uncertainty through strategic cooperation; TSMC's short-term earnings risk is considered limited; while KIOXIA participates in the memory cycle, its risk-reward is weaker because NAND is more vulnerable to competitive pressure from China.
Risks
- The favorable pricing environment for memory may end earlier than expected due to weaker demand or increased supply.
- Changes in investor sentiment may compress memory stock valuations.
- China's progress in memory, especially NAND, poses a long-term downside risk.
- NVDA faces risks from slower-than-expected revenue growth in key end markets, competitors cutting prices or taking share, customers migrating to in-house chips, and technology export regulation.
- AVGO faces risks from unexpected weakness in AI demand, loss of share or sockets at key customers, failed M&A synergy execution, insufficient cash returns, and management changes.
- If KIOXIA's NAND demand, pricing, policy support, and cost declines fall short of expectations, investment returns may weaken further.
What to watch
- The technology roadmap, supply volume, and pricing arrangements for HBM4 and subsequent AI memory in the SK hynix-NVIDIA cooperation.
- Whether Samsung's MOU with Broadcom converts into actual orders, and whether it covers AI ASIC wafer or advanced packaging production.
- Changes in capacity, yield, and customer adoption between TSMC CoWoS and Samsung's Cube-series advanced packaging.
- Whether global memory revenue in 2027 and 2028 approaches the consensus expectation of about US$1.3T.
- Whether NVIDIA's and Broadcom's guidance for future AI growth continues to be revised upward or delivered on.
- The technological progress of Chinese manufacturers in NAND and DRAM, especially whether they break through key process constraints that affect the competitive landscape.