The Substrate Line — 72 Hours Three Korean Glass Players Cleared TGV, and Samsung Rewrote HBM-Foundry

For 72 hours ended September 7, three Korean glass substrate developers pushed Through-Glass-Via (TGV) technology from lab curiosity to mass-production readiness — and Samsung answered by folding HBM and foundry into a single ASIC pitch. This is what the packaging pivot looks like when it happens simultaneously up and down the stack.

The most concrete data point came from PhiOptics at KPCA SHOW 2026, where the Korean materials supplier displayed 2mm-thick TGV glass substrates at 0ppm defect rate. Thickness matters here — 2mm is the range required for multi-die AI packaging with tall HBM stacks, and the industry has spent two years arguing whether glass could survive the drilling-and-metallization sequence at that thickness without cracking. PhiOptics just answered the question in public.

The same week, KULS — a startup founded by ex-Samsung engineer Cho Jin-hyun — announced solutions to the TGV metallization bottleneck. Metallization has been the second wall for glass substrates: once you drill the vias, filling them with conductive material without voids, cracks, or copper-glass adhesion failure is where most industrial pilots have stalled. That two independent Korean players cleared complementary parts of the same problem in the same week is the pattern.

The third piece is capital. SKC committed 281 billion won to Absolics as part of a 589.6 billion won plan running through 2028 — three years of substrate capacity build-out anchored on the Korean-US glass substrate joint venture that has already been sampling to major GPU customers. Add Hanyang E&G's 148 billion won contract at Samsung's Pyeongtaek P5 fab for cleanroom piping (13.6% of P5 Phase 1), and you have four supplier-side announcements in one week that all describe the same movement: Korean domestic packaging capex is being funded on the assumption that HBM alone will not carry the next AI cycle.

At the chip level, Samsung's move rhymes with the substrate story. Inews24 reported that Samsung is now marketing HBM and foundry as a bundled ASIC platform — vertically integrating the two capabilities specifically to target the AI ASIC market where hyperscalers want a single Korean vendor to deliver both the accelerator die and the memory stack. This is a strategic answer to the market-share problem the news cycle keeps returning to: SK Hynix holds 50% HBM share versus Samsung's 33%, and the gap is narrowing rather than widening. Samsung cannot close that gap with a better HBM4 alone. It can close it by offering something SK Hynix structurally cannot — HBM plus advanced-node logic on the same purchase order.

The macro backdrop justifies the capex. Korean semiconductor exports hit $32.7 billion in July (YoY +166%), following $33.6 billion in June (+174%) and $29.4 billion in May (+154%). The DDR5 16Gb spot price sat at $54.13 on September 7, and analysts at Mirae Asset raised Samsung's target to 400,000 won on the view that memory supply will remain tight through 2028. KB Securities telegraphed that Samsung will resume buybacks this year as earnings recover, and the company is guiding toward 200 trillion won in H2 operating profit. When the export line runs at +170% YoY and the largest customer is signaling ten-year memory demand, the choice to fund substrate lines and vertical HBM-foundry stacks is not speculative — it is defensive.

The competitive framing sharpens the point. Micron entered HBM and CXMT is reportedly beginning HBM3E production, which means the pure-memory margin advantage Korea has enjoyed since 2024 is now on a countdown. The Korean response, as the last 72 hours shows, is to move up the stack: TGV substrates that can host tall HBM plus logic; HBM-foundry integration that no CXMT or Micron can match; and a domestic packaging supply chain (PhiOptics, KULS, Absolics/SKC, Hanyang E&G) that keeps the value-added integration inside Korean borders. If CXMT wants to compete in three years, it needs not just DRAM cells but a comparable packaging ecosystem — and that is much harder to build than a fab.

There is a domestic wrinkle. Foreign retail investors sold Korean semiconductor large-caps in September, while leveraged ETFs accumulated, and SK Hynix will list an ADR-linked leveraged ETF on September 13-14. This is a re-rating in progress: institutional flows into instruments, retail flows out of names, valuation being decided by ETF creation rather than fundamental accumulation. The substrate and vertical-integration story is what those instruments are pricing.

The signal for the next quarter is that Korean equity re-rating is no longer a memory-cycle bet. It is a bet on whether Korea can extract the packaging premium in the ASIC decade — and PhiOptics' 0ppm, KULS' metallization, and SKC's 281 billion won are three receipts that the bet is being funded in real time.

Key Sources: - PhiOptics achieves 0ppm defect rate for 2mm thick glass substrates for AI packaging (thelec, 2026-09-07) - KULS solves critical TGV metallization bottleneck for glass substrate mass production (thelec, 2026-09-04) - SKC Invests 281B Won in Absolics Glass Substrate Expansion (thelec, 2026-09-04) - Samsung targets AI ASIC market with HBM-foundry collaboration (inews24, 2026-09-07) - Hanyang E&G Wins 148B Won Samsung P5 Fab Cleanroom Piping Contract (thelec, 2026-09-07) - plus 4 more