Samsung Electronics
2026-07-30
Samsung locks 60% capacity in long-term deals, HBM4E sampling, memory shortage through 2028
Summary
Key Takeaways
Samsung's Q2 2026 revenue reached 171.5 trillion won, operating profit 89.5 trillion won (margin 52.2%). DS division contributed 99.7% of profit, with memory revenue (DRAM+NAND+HBM) at 120.8 trillion won, up 62% QoQ and 471% YoY.
In HBM, Samsung is expanding HBM4 sales, expecting HBM4 revenue to triple QoQ in Q3 and account for over 60% of total HBM revenue in H2. First HBM4E samples delivered to key customers. Next-gen HBM5 will use 2nm GAA process, with speed 50%+ over HBM4E.
Supply strategy: Samsung locks 60-70% of total capacity in 5-year rolling LTAs, signed with 5 global data center customers, 5 more in final negotiations. Contracts include prepayment clauses (multi-year deposits), ~25% already received.
Outlook: Memory shortage will worsen in 2027 vs 2026, lasting through 2028; 2029 unclear. Samsung accelerating expansion: Taylor Phase 2 fab to start construction end-2026, targeting 2030 production; surging inquiries for 1.4nm process.
DX division posted loss of 0.8 trillion won on 48 trillion won revenue, due to memory cost pass-through.
Why It Matters
Samsung's record profit masks a strategic lock-in: LTAs with prepayments anchor AI customers, defending against SK Hynix and Micron by securing capacity commitments. The hidden trap is the 2nm GAA process for HBM5 – Samsung's GAA yield issues (e.g., 3nm) cast doubt on HBM5 timeline, yet customers are locked via multi-year contracts. By locking 60-70% capacity, Samsung squeezes spot market supply, raising costs for non-contracted buyers. This ecosystem restructuring reduces enterprise architectural flexibility: customers tied to Samsung's roadmap cannot pivot to competitors' advanced HBM (e.g., SK Hynix's Hybrid Bonding) for specific AI workloads. The prepayment deposits create sunk cost barriers, limiting vendor switching.
PRO Decision
[Vendors] SK Hynix and Micron should exploit Samsung's LTA rigidity: promote their own advanced HBM (e.g., Hybrid Bonding) with more flexible, shorter-term contracts and no prepayment demands. Highlight Samsung's 2nm GAA yield risks and offer co-development partnerships to attract AI customers seeking architectural flexibility.
[Enterprises] CIOs must audit Samsung's LTA for lock-in: demand guarantees on HBM5 2nm GAA timeline and yield, include technology switch clauses, and maintain dual sourcing (SK Hynix, Micron) to preserve leverage. Negotiate shorter lock periods and lower prepayment percentages to retain flexibility for future AI workload shifts.
[Investors] Look beyond the profit surge: Samsung's LTA concentration on 5 hyperscalers raises customer dependency risk. Heavy capex (Taylor Phase 2) pressures FCF. If Samsung's 2nm GAA HBM5 faces delays, its competitive edge may erode. Evaluate whether prepayment income is sustainable and whether LTAs become a liability when memory cycle turns.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)