Nvidia reportedly buys glass substrate samples from Korea's JNTC

The 2mm glass core is the tell. This is not a generic packaging upgrade—it is a direct response to warpage and CTE-mismatch limits on 75mm-plus accelerator dies. TGV at that depth demanding sub-10μm via pitch forces simultaneous breakthroughs in femtosecond laser drilling and electroless copper metallization, pulling Corning and Schott into the critical path upstream while restructuring the CoWoS/SoIC stack downstream, potentially eliminating certain RDL layers entirely. JNTC's entry into Nvidia's AVL is a supply-chain hedge, not a routine procurement. With US export controls on advanced packaging equipment tightening, a Korean substrate node dilutes concentration risk on Japan (Ibiden, Shinko) and China Taiwan (Unimicron). The playbook mirrors SK Hynix's 2023 insertion into the HBM ecosystem. Intel publicly pegged glass-substrate mass production at 2030. Nvidia sampling in 2025 compresses that timeline by three to five years. Ibiden and Unimicron face immediate ABF margin pressure; Samsung's in-house packaging will accelerate its TGV roadmap to defend its foundry moat. Twelve to twenty-four months out: TGV yield is the binding constraint—industry estimates sit at 60-70% for 2mm cores versus 95%+ for organic. Expect first CoWoS-L pilot lots by Q2 2026. The structural inflection lands in 2027-2028, when HBM4's 16-high stacks demand thermal and flatness specs that only glass can deliver.