Samsung Accelerates P5 Fab 2 Construction; Memory Chip Shortage Expected to Persist Through 2028
nashnova research
Samsung has moved the P5 Fab 2 groundbreaking forward by six months to July 2026 and locked in multi-year supply contracts; the memory chip shortage is expected to persist through 2028 and beyond, with equipment delivery bottlenecks emerging as the biggest variable for capacity ramp.
Why is P5 Fab 2 breaking ground six months early?
Samsung pulled the P5 Fab 2 start date from early 2027 to July 2026; structural column installation begins January 2027.
This means → Samsung sees AI-driven memory demand growing far faster than expected and is racing to build capacity before the supply gap widens further.
P5 Fab 1 is already under construction, targeting 2028 production; Fab 2 is slated for 2029 operations. Samsung is also pushing to move Fab 1 equipment installation from Q3 to Q2 2027.
Where is the 2026 spending going?
Samsung's 2026 full-year capex is projected at roughly ₩84.4 trillion, of which the Device Solutions division accounts for ₩76.8 trillion.
In plain terms = the vast majority of Samsung's capital is going straight into chip manufacturing — new fabs and infrastructure.
This reflects a "build capacity first, fill orders after" strategy for this AI-driven demand cycle.
How much capacity is locked up in long-term contracts?
Samsung has signed multi-year memory supply agreements with 10 major customers and is negotiating with 10 more, targeting long-term contract coverage of 60%–70% of capacity by 2030.
For HBM — high-bandwidth memory, a high-speed stacked memory designed specifically for AI chips — Samsung's available capacity at target pricing for 2027 is fully sold out, and key customers are still requesting additional supply.
SK Hynix has similarly locked in 14 major customers (including global cloud providers), covering over 50% of total capacity for 2027–2028, and expects demand to exceed supply through 2030.
How severe is the equipment delivery bottleneck?
ASML, the world's sole EUV lithography supplier, says its planned output for 2027–2028 cannot fully meet customer demand.
The company plans to boost low-NA EUV — the most advanced production-grade lithography system — and DUV immersion lithography capacity by 30% in 2027 and is studying a larger expansion for 2028; customers have begun placing equipment order inquiries for 2029 and even 2030.
This means → even if Samsung and SK Hynix finish building their fabs on time, production lines cannot start without the lithography tools — equipment lead time is now the critical bottleneck for capacity ramp.
What will P5 Phase 1 and Phase 2 each produce?
Phase 1 focuses on DRAM and HBM production; supplier selection is nearly complete, with potential equipment orders estimated at over ₩10 trillion.
Phase 2 is being discussed for advanced NAND processes; delivery lead time is identified as the key factor for production start.
In plain terms = Phase 1 tackles the memory type AI needs most urgently, Phase 2 fills in the other major memory category — a two-step rollout that shows Samsung is prioritizing within its own product line.
How long can this memory upcycle last?
On the supply side, new capacity from Samsung and SK Hynix cannot come online before 2028–2029 at the earliest, and equipment bottlenecks could push the timeline further.
On the demand side, Samsung expects AI-driven memory demand in 2027 to exceed industry supply by an even wider margin than 2026.
This means → supply ramps slowly while demand grows fast — the two lines likely will not cross until at least 2028. Until then, memory chips will almost certainly remain in shortage. The real test: whether AI demand can keep delivering incremental growth throughout this window.
市场有风险，内容仅供研究参考，不构成投资建议。