Nvidia is pursuing glass substrates to replace organic Ajinomoto build-up film (ABF) in AI chip packages, as modelers place HBM cost at $3,917 to $7,373 for a 288 GB GPU between 2026 and 2027. Samsung Electro-Mechanics (Semco) has pushed glass substrate mass production to after 2028, following a prototype failure in a customer reliability review. Through-glass via (TGV) packaging solutions are targeting 1,000 Watts of power delivery and 1,000 Amperes of current delivery for next-generation AI accelerators. The launch window matters because larger packages create room for more HBM as each gigabyte grows costlier, according to Yahoo Finance’s Nvidia glass packaging analysis.
Why glass replaces organic ABF in high-end AI packages
Glass offers larger, flatter packages with denser connections and more HBM room than traditional organic ABF. Those larger packages accommodate increasingly expensive high-bandwidth memory: the modeled HBM cost for a 288 GB GPU runs $3,917 to $7,373 in the 2026 to 2027 window. Nvidia is pursuing this substrate shift while ABF is pushed toward larger form factors, higher bandwidth, and integration. For a closer view of packaging and substrate developments, readers can follow the AI chip manufacturing hub.
TGV formation, panel warpage, and reliability testing
Scaling glass substrates depends on resolving manufacturing and standardization issues. Industrywide adoption requires resolving manufacturing and standardization issues first. Lam Research is helping scale glass-panel based packaging solutions, and glass has long been used as a temporary round carrier in advanced packaging manufacturing. AUO is extending large-format panel manufacturing into advanced semiconductor packaging. On September 10, 2026, AUO glass substrate sizing fueled TSMC collaboration speculation. LG Innotek applies in-house AI to glass substrate manufacturing targeting microcracks and is aiming for the 2028 glass substrate production race.
Semco’s timeline illustrates the reliability hurdle. The company delayed glass substrate commercialization after a prototype failed a customer reliability review. Its Q2 earnings call on August 3, 2026 signaled production in 2027, but an August 14, 2026 update pushed mass production to after 2028. Semco is also expanding its SoulBrain partnership on glass substrate materials for AI packaging and will make its first glass substrate appearance at Semicon Taiwan on September 1, 2026. At the same time, ceramic cores emerge as a challenger to glass as advanced packaging substrates hit limits. FOPLP leads production while CoPoS advances glass packaging, and by September 16, 2026, glass substrates and TGV were set to reshape the 2027 packaging race. The delay does not affect the long-standing use of glass as a temporary round carrier in advanced packaging manufacturing, but panel-level and substrate-level reliability is a different problem.
More detail on these packaging trajectories is available in Lam Research’s glass substrate packaging update and DIGITIMES glass substrate coverage.
Key figures and milestones
| Item | Figure / Milestone | Source / Company |
|---|---|---|
| Modeled HBM cost for 288 GB GPU | $3,917 to $7,373 (2026–2027) | Nvidia glass packaging analysis |
| TGV power delivery target | 1,000 Watts | TGV packaging solutions |
| TGV current delivery target | 1,000 Amperes | TGV packaging solutions |
| Intel–Lens Technology MOU | July 25, 2026 | Intel, Lens Technology |
| Lam Research glass substrate content | October 29, 2026 | Lam Research |
| Semco glass substrate production signal | 2027 (Q2 call, August 3, 2026) | Samsung Electro-Mechanics |
| Semco mass production slip | after 2028 (August 14, 2026) | Samsung Electro-Mechanics |
| LG Innotek production target | 2028 | LG Innotek |
Intel signed a memorandum of understanding with Lens Technology on July 25, 2026, and the two companies are collaborating to explore glass substrate semiconductor packaging using TGV advanced packaging. The China glass substrate supply chain is taking shape around that Intel–Lens MOU as of July 29, 2026.
What infrastructure teams should verify before procurement
Glass substrate adoption is not yet a qualified, volume-ready option for production AI accelerators. The unresolved items that infrastructure buyers should track include through-glass-via reliability, customer qualification of glass substrates, industrywide manufacturing and standardization issues, Semco glass substrate mass production timing after 2028, and whether glass substrates can scale reliably for production AI accelerators. Buyers should request package-level power and current validation against the 1,000 W and 1,000 A TGV targets, plus warpage and microcrack data from LG Innotek’s in-house AI inspection. Teams planning around HBM-dense packages should also compare memory choices using the choosing memory for AI workloads guide and watch packaging readiness via the AI Chip Buyer’s Guide. For semiconductor industry standards and roadmaps, see SEMI Global Semiconductor Alliance.