Glass Substrate Series #15

Glass Substrate vs. Silicon Interposer
Economics
Cost, Yield, and Scalability in Real Numbers

Glass substrates cost 2–3x more than silicon interposers today. So why are AMD, Intel, and TSMC moving to glass? The numbers tell a more interesting story than the current price tag.

Paradigm Shift Lab  ·  May 2026

2–3×

Current glass substrate cost
vs. organic equivalent

75–85%

Glass substrate yield today
(silicon: 90–95%)

50%

Warpage reduction
vs. silicon interposer

Why Pay More for Glass? — Starting With the Right Question

Basic economics says: if performance is equal, buy the cheaper option. Yet the semiconductor industry is moving toward a more expensive material. Glass substrates cost more than silicon interposers today — and 2–3x more than organic substrates — yet AMD, Intel, and TSMC are all moving toward glass.

Understanding whether this is rational requires looking at total cost of ownership (TCO) rather than unit price — and understanding how yield and scalability reshape the economics over a 3–5 year horizon.

💡 What Is an Interposer — and Why Does It Matter?

An interposer is an intermediate connection layer between chips and the package substrate. It’s essential for connecting different chips at high density — like a GPU and HBM memory. The current standard for high-performance AI accelerators is the silicon interposer, as used in TSMC’s CoWoS packaging. Glass substrates are being developed as the replacement for this silicon layer.

The Core Comparison — Real Numbers Today

Metric	Silicon Interposer	Glass Interposer	Organic (ABF)
Current Cost	High (baseline)	2–3× baseline	Lowest
Manufacturing Yield	90–95%	75–85%	90–95%
Minimum Line Width	1–2 μm	2–10 μm	10–15 μm
Surface Flatness	Moderate	Best-in-class	Low
CTE Match to Chip	Low (good match)	Low (tunable)	High (mismatch)
Warpage	Moderate	50% less than silicon	Severe
Dielectric Loss	Moderate	Lowest (best signal integrity)	High
Panel Size Scaling	Wafer-limited (300mm)	510×515mm+	Large panel possible
Via Process	TSV (complex, costly)	TGV laser (simpler)	Not applicable
2030 Cost Outlook	Stable at current level	40–60% reduction expected	Marginal reduction

The Cost Structure — Why Glass Is Expensive Now

Glass substrates aren’t expensive because the material is expensive. The glass itself is cheap. The cost comes from process immaturity and lack of scale.

⚡ Silicon Interposer Cost Structure

• Wafer materialExpensive
• TSV drillingComplex, costly
• PhotolithographyRequired (costly)
• Max size300mm wafer limit
• Process maturityMature (high yield)
• Scaling ceilingHard wafer size limit

✦ Glass Interposer Cost Structure

• Glass materialCheap (advantage)
• TGV laser drillingSimplifying rapidly
• PhotolithographyNot required (advantage)
• Max size510×515mm+
• Process maturityEarly (yield improving)
• Scaling potentialPanel-level, unlimited

The key insight: glass material is cheaper than silicon, but process costs are still high. TGV laser drilling, metallization, and build-up layer deposition are all early-stage processes with high per-unit cost. This is exactly the type of cost that drops sharply as scale increases.

Yield — The Biggest Problem Right Now

Glass substrate yield today is 75–85%, versus 90–95% for silicon and organic substrates. This is the single most direct driver of today’s cost premium.

The primary cause is the SeWaRe problem — Seam, Warp, and Residue defects in glass processing. In January 2026, Intel debuted the first “No SeWaRe” glass core substrate sample, marking a decisive step toward solving this challenge.

When Yield Hits 90%, the Economics Flip

Once glass substrate yield reaches silicon-level 90%, three advantages converge simultaneously: lower material cost + larger panel size + simpler process. That combination makes glass cheaper than silicon interposers. The industry targets this crossover at 2028–2030. This is the exact timing that explains why AMD chose 2028 for its first glass substrate product adoption.

Scalability — Where Glass Beats Silicon Decisively

The silicon interposer has a structural ceiling: 300mm wafer size. As AI accelerator scale grows, larger interposers are needed — and silicon cannot deliver them.

Glass is different. SK Absolics’ current substrate measures 510×515mm. Rapidus unveiled a 600×600mm sample at SEMICON Japan 2025. Panel-level processing removes size limits entirely in principle.

This is glass’s structural advantage in the AI era. The larger AI clusters grow, the harder silicon’s limit becomes — and the more valuable glass becomes.

Where Glass Wins, Where Silicon Holds

✦ Glass Wins

Large AI accelerator interposers, high-frequency signal paths, CPO optical waveguide integration, 300mm+ area packaging, post-2028 new chipset designs

≈ Contested Zone

Mid-scale AI inference chips, 2026–2027 transitional products, HBM connection interposers (outcome depends on yield improvement pace)

⚠ Silicon Holds

Currently in-production CoWoS-based products, low-volume high-spec custom chips, yield-sensitive leading-edge nodes (N2 and below)

The Cost Reduction Roadmap — When Does the Crossover Happen?

Timeline	Glass Cost Level	Yield	Key Driver
2026 (Now)	2–3× organic equivalent	75–85%	Pilot line stage. AMD pilot production starts
2027–2028	1.5–2× level	85–90%	Samsung EM & SK Absolics volume production. TGV process improvement
2029–2030	At or below silicon interposer cost	90%+	Panel-level mass production, economies of scale. SeWaRe resolved
2030+	Below silicon interposer	95%+	Material cost advantage + panel size advantage + mature process: all three at once

🔑 Why Pay Today’s Price for Tomorrow’s Economics

When AMD chose glass substrates for 2028, it was buying the 2028 price, not the 2026 price. If you don’t invest in pilot qualification now and lock in your supply chain, you won’t have access to glass substrates when they become cost-competitive in 2028. In semiconductor materials transitions, early qualification is supply chain pre-emption. That’s why AMD is moving despite today’s premium. Note: this is not investment advice — actual decisions should involve professional guidance and your own judgment.

The Bottom Line

Glass substrates are expensive today. Yield is lower than silicon. Both are true.

But the raw material is cheaper than silicon, panel scalability is unlimited, and signal characteristics are superior. When process maturity arrives and yield hits 90% around 2028–2030, the economics flip. AMD targeting 2028 is the strategic move to be positioned at that crossover before competitors are.

This concludes the glass substrate series. Supply-side materials and processes, demand-side Big Tech roadmaps, and economic convergence — all three vectors are crossing at the same point in 2028–2030. That intersection is when glass substrates become the new standard for semiconductor packaging.

The glass substrate series covered materials, process, demand, and economics in full. Paradigm shifts always follow the same pattern: an expensive, immature technology gradually replaces a cheap, mature one. Glass substrates are in the middle of that process right now.

Paradigm Shift Lab  ·  Documenting the moments when paradigms shift

Glass Substrate Series

Previous: #14 AMD, Apple, Tesla & Broadcom’s Glass Substrate Roadmaps — The Demand Side Is Finally Moving
Continuing: Solid-State Battery & Space series ongoing