Nanya Technology's LTAs cover up to 70% of capacity as Pingtung 3D IC plant targets 2H28 production

Nanya's 50-70% LTA coverage is not a margin story—it is the DRAM industry's structural pivot from cyclical commodity to contractual allocation. Customers pushing three-year extensions reveal a deeper signal: supply-trust erosion in the AI inference memory segment, not demand euphoria. The Pingtung 3D IC plant (2H28 target) slots into the bandwidth gap between HBM and conventional DDR5. If TSV stacking and hybrid bonding yield targets are met, upstream equipment makers (BESI, ASMPT) see order pull-forward, while downstream inference chip designers gain a 30%+ power-efficiency alternative to HBM. Competitively, this is a deliberate mis-positioning play. SK Hynix and Samsung are locked into HBM4 roadmaps for 2027; Nanya competes on cost-per-Gbps and supply elasticity rather than peak bandwidth. Micron's hesitation on 3D DRAM opens an 18-24 month window. 12-24 month outlook: LTA lock-in will compress spot-market liquidity, shifting DRAM pricing from quarterly auctions to annual anchoring. The real risk is not technical—if AI capex corrects in 2027, three-year fixed capacity becomes a balance-sheet liability, not an asset.