How to Compare Glass Substrate Capacity Claims
A comparison has started circulating that puts a Chinese glass substrate project at 100,000 square meters a year against 12,000 for Absolics, SKC's plant in Georgia. That is an 8.3x gap on paper. It is also a comparison of two plans that describe different products at different process stages, and neither number is output.
The useful question is not who is building the bigger factory. It is what you need to know before two glass substrate capacity numbers can be compared at all. I count four things, and public filings answer surprisingly few of them.
KEY TAKEAWAYS
1. Absolics' 12,000 m² Phase 1 (announced November 2022) and WG Tech's 100,000 m² Phase 1 (company notice, June 2024) are both nameplate targets. WG Tech's line is shared across chip packaging, RF antennas and Micro LED displays.
2. Target dates keep moving. Absolics went from "production from Q2 2024" in 2022 to "reliability testing, no date" in September 2026. Samsung Electro-Mechanics went from "2026-2027" in April 2024 to "2028" in July 2026.
3. On public disclosures, none of five players (Absolics, Samsung Electro-Mechanics, LG Innotek, WG Tech, Intel) has reached qualified volume supply, and none discloses yield.
Yardstick 1: square meters of what
A "glass substrate" is a stack of process steps. Someone makes the glass panel or core. Someone drills and fills through-glass vias (TGV). Someone metallizes it and builds up routing layers until a chip can sit on it. A capacity number only means something once you know which of those steps it covers.
WG Tech's own June 2024 notice describes a project at its Hubei subsidiary, TGV Circuits, sized at 1 million m² a year in total, with a CNY 500 million first phase of 100,000 m². The stated products include advanced packaging substrates, 6G RF antennas and Micro LED display backplanes. The Absolics figure, from SK's November 2022 groundbreaking release, is for finished glass substrates for semiconductor packages.
Samsung Electro-Mechanics shows why the stage matters. The joint venture it announced on July 2, 2026 with Dongwoo Fine-Chem, a Sumitomo Chemical subsidiary, makes glass core material rather than finished substrates. Total investment is about KRW 480 billion, Samsung holds roughly 66%, and operations are targeted for the second half of 2027. It is a glass substrate investment, but at the front of the line.
Yardstick 2: plan or running line
Nameplate capacity describes a finished, fully utilized plant. The more telling data is how the start date has moved each time a company updated it.
Absolics said in 2022 that Phase 1 would start production in the second quarter of 2024. When the U.S. Commerce Department finalized up to $75 million of CHIPS funding in December 2024, the stated schedule was first customer deliveries in 2025 and capacity coming online in 2027. As of September 2026, SKC describes the embedded product as having finished preliminary evaluation and entered reliability testing, with a non-embedded product entering supplier selection in the second half.
Samsung Electro-Mechanics CEO Chang Duck-hyun talked about 2026-2027 mass production in April 2024. On the July 30, 2026 earnings call the company said full-scale production would start in 2028. LG Innotek announced a 2028 target at KPCA Show on September 10.
The Chinese side reads the same way once you go to the source. At its May 2026 results briefing, WG Tech management gave no mass production date, said TGV products were in sample validation across several applications, and described revenue at the glass substrate subsidiary as very small and loss-making, according to Cailian Press.
Yardstick 3: where the customer is
A substrate earns revenue only after a customer puts its own silicon on it, runs reliability qualification and signs off on volume. That step is the one most often skipped in capacity headlines.
Intel, which put glass substrates on the map with its September 18, 2023 announcement, said it had spent more than a decade on reliability research and planned complete solutions in the second half of the decade. In January 2026 it showed a 78 x 77 mm thick-core glass engineering sample with EMIB bridges, as reported by TrendForce. In 2025, Korean press reported Intel was shifting toward outside sourcing and licensing; Intel has not confirmed that publicly. When the first large buyer is still undecided, a capacity ranking is not a supply ranking.
Yardstick 4: yield, the number nobody publishes
No company in this group discloses glass substrate yield. Korean reporting ties Absolics' delays to the difficulty of fine via formation, and WG Tech did not give yield figures at its briefing. So the only honest way to show the effect is an illustrative model.
Start with announced capacity at 100. Assume half of it goes to chip packaging, the line runs at 60% utilization and yield is 70%. Effective supply is 21 (illustrative calculation, not actual company figures). Halve any one ratio and the result halves. An 8x area advantage says nothing about supply until those three boxes are filled.
What I actually watch
| Checkpoint | Why it matters | Where it shows up |
|---|---|---|
| Absolics supplier selection (2H 2026) | First sign a customer has moved past reliability testing | SKC disclosures, earnings call |
| SKC capital deployment | KRW 589.6bn of the May 2026 raise is earmarked for Absolics, KRW 472.6bn of it capex; KRW 281bn goes in first | Korean DART filings |
| Samsung glass core JV | Incorporation by year-end, operations in 2H 2027 | Samsung Electro-Mechanics IR |
| WG Tech subsidiary revenue | Whether shared-line capacity turns into packaging sales | Shanghai-listed periodic reports |
| Intel sourcing decision | Who the first volume customer buys from | Intel Foundry announcements |
The Korean filings are the part most English coverage misses. SKC's disclosures on the Absolics rights offering spell out how much goes to equipment versus operating costs, which is a better read on progress than any square-meter figure.
Value chain read-through
| Stage | What changes | What to verify |
|---|---|---|
| Glass panel / core | New material supply chain outside organic laminates | Core JV start-up, material qualification |
| TGV drilling and fill | Laser and plating steps widely cited as the hardest | Via yield commentary, tool orders |
| Build-up and finished substrate | Substrate makers compete on reliability, not area | Customer reliability milestones |
| Package customers | Adoption decides which capacity is real | Product announcements naming glass cores |
Risks to this view
• All production dates are company targets, and this segment has a record of targets slipping.
• WG Tech's second-phase investment, customer validation and the "Korea and China are at parity" claim come from unnamed industry sources in press coverage. I did not use them as evidence.
• Absolics' 12,000 and 72,000 m² figures are from the 2022 plan. I could not find an official update to those numbers.
• Yield and utilization are undisclosed everywhere. The ratios in the fourth chart are assumptions.
• Nothing here estimates any company's customers, orders or volumes.
Area is the first number a company announces and the last one anyone can verify. For glass substrates, the finish line is a qualified customer, and on public records nobody has crossed it yet. Next, I will break down TGV processing, the step most often cited as the bottleneck, by equipment and materials.
Sources: Intel Newsroom (Sep 18, 2023); SK press release on the Absolics groundbreaking (Nov 1, 2022); U.S. Commerce Department and NIST CHIPS award (Dec 2024); The Elec (Sep 4 and Sep 27, 2026); WG Tech company notice (Jun 13, 2024); Cailian Press (May 25, 2026); Korea Economic Daily (Apr 12, 2024); Newspim (Jul 2 and Sep 10, 2026); Digital Daily (Jul 30, 2026); TrendForce (Aug 22, 2025 and Jan 26, 2026).
Comments
Post a Comment