Question: Does Tower Semiconductor plug into the glass substrate build-out, and how meaningfully?
Question: Does Tower Semiconductor plug into the glass substrate build-out, and how meaningfully?
TL;DR: No direct linkage. Close to zero indirect linkage. TSEM and the glass substrate thesis are two separate bets on the same macro (AI compute → bandwidth and packaging bottlenecks). They do not overlap in the supply chain. If someone pitches TSEM as a glass substrate beneficiary, that’s a category error — likely caused by conflating two different “glass” stories on fintwit. TSEM is a silicon photonics / SiGe foundry; the glass thesis belongs to Corning, AGC (5201), SKC/Absolics, Samsung Electro-Mechanics, Ibiden, LPKF, and Intel as the anchor consumer.
Before anything else, split the term. Most of the confusion comes from people (and model write-ups) smashing these together.
Glass panel replaces the organic (ABF) core of a flip-chip package substrate. Laser-drilled through-glass vias (TGVs) are copper-filled to carry signal and power between the die and the board. Target end-market is very large AI/HPC packages (Intel Clearwater Forest, future GPUs) where ABF hits limits on warpage, flatness, and fine-pitch routing.
This is a packaging substrate story, not a wafer story. The beneficiaries sit downstream of the wafer fab — between the front-end foundry (TSMC, Intel, Samsung, TSEM, GlobalFoundries) and the finished package that lands on a server motherboard.
Who wins Story A: - Glass material suppliers: Corning (GLW), AGC (5201), NEG (5214), SCHOTT (private) - Substrate assemblers / finishers: Absolics (SKC affiliate, 011790.KS), Ibiden (4062.T), Shinko (6967.T, going private), Samsung Electro-Mechanics (009150.KS), LG Innotek, Toppan (7911.T) - TGV laser drilling equipment: LPKF (LPK.DE) — cleanest pure-play on LIDE laser drilling; Han’s Laser - Metrology for TGV: Onto Innovation (ONTO), Camtek (CAMT) - Anchor consumer: Intel (INTC) — Clearwater Forest Xeon 6+ launched at CES January 2026 with an 800-micron glass core, 10 RDL layers each side
2026 state of play: Intel Clearwater Forest is the first mass-produced commercial CPU with a glass core — produced at Chandler, AZ. Absolics’ Covington, GA line is ramping (~20k sheets/month target, $75M CHIPS Act award). Samsung’s Sejong pilot line slipped from 2026 to 2027 mass production. Ibiden announced ¥500B (~$3.3B) FY26-28 capex with glass-core R&D called out. TSMC is conspicuously absent as a glass adopter — matters because TSMC controls the AI package flow via CoWoS.
Glass used as the wafer itself or as an interposer for a device — leveraging glass’s low dielectric loss, high resistivity, and optical transparency. Use cases: RF interposers for 5G/6G front-ends, through-glass-via photonic interposers for co-packaged optics, optical MEMS lids, high-resistivity glass wafers for antenna-in-package.
Different customer base (foundries, MEMS/RF device makers, CPO vendors), different timeline, smaller near-term TAM than Story A.
Who plays Story B: - Specialty glass wafer suppliers: Corning (GLW), SCHOTT, AGC (5201), Plan Optik (PLO.DE) - Foundries that could process glass-based photonic/RF wafers: theoretically TSEM, GlobalFoundries (GFS), Intel Foundry, TSMC — but in practice SiPh today is overwhelmingly on silicon-on-insulator, not glass - CPO / photonics device makers pulling the demand: Coherent (COHR), Lumentum (LITE), Marvell (MRVL), Broadcom (AVGO), Ayar Labs (private), OpenLight (private) - TGV equipment: LPKF, Han’s Laser (same equipment as Story A, different end market)
2026 state of play: CPO volumes inflect on NVIDIA Quantum-X (early 2026) and Spectrum-X (H2 2026). Glass interposers and glass-based optical sub-assemblies are part of the toolkit but not the dominant substrate. SCHOTT launched optical MEMS lids for OCS at OFC 2026. The TAM for glass-specific photonics substrates in 2026 is small compared to Story A.
The critical point for this memo: Story B is where TSEM would theoretically appear if TSEM were a glass play. It isn’t. See below.
TSEM is a specialty analog wafer foundry — 45-180nm “more-than-Moore” processes — built around six platforms:
| Platform | What it does | End markets |
|---|---|---|
| RF-SOI | Smartphone front-ends, 5G | Qorvo-adjacent, handset TAM |
| SiGe BiCMOS | TIAs, drivers, modulators for optical modules; auto radar | 50%+ market share in optical TIAs |
| Silicon photonics (PH18) | SiPh PICs on 200mm now, 300mm ramping at Albuquerque / Migdal | AI data-center optical interconnect, CPO |
| BCD power | Up to 700V power management | Auto, industrial, consumer |
| CMOS image sensors | BSI sensors | Mobile, industrial vision |
| MEMS | Various | Industrial, medical |
Fabs: Migdal HaEmek (150/200mm, IL), Newport Beach (200mm, CA — SiPh workhorse), San Antonio (200mm, TX), TPSCo Japan (300mm, 51%-owned), Agrate (Italy, shared with STM), and the disputed Intel Fab 11X Rio Rancho (300mm) deal.
Recent numbers (Q4 2025 / April 2026): FY2025 revenue $1.566B (+9%), Q4 $440M record. Silicon photonics doubled to $228M (+115% YoY). Q1 2026 guide $412M ±5%. Stock rallied from ~$15B market cap in February to ~$26-27B on the NVIDIA 1.6T deal (Feb 2026) and SiPh ramp. Trailing P/E 70-80x; ~20x on management’s 2028 net profit target of $750M.
Key recent partner/design-win signals (all silicon-based): NVIDIA 1.6T optical module (Feb 2026), InnoLight, Coherent, Scintil, Xscape, Salience Labs, LightIC, Xanadu. None reference glass substrates.
TSEM exposure: zero. TSEM is a front-end wafer fab, not an OSAT and not a substrate assembler. It does not produce glass cores, does not buy glass cores, and its output (finished wafers) feeds into OSATs (ASE, Amkor, JCET) and IDMs who may or may not use glass-core substrates downstream — but that’s four layers removed.
TSEM does not appear in any 2025-2026 market report on the glass substrate segment. It does not partner with Corning, AGC, SKC/Absolics, Samsung Electro-Mechanics, Ibiden, Shinko, or LPKF. Management has not referenced glass substrates on Q3 2025, Q4 2025, or Q1 2026 earnings calls.
TSEM exposure: effectively zero. Silicon, not glass. TSEM’s PH18 silicon photonics platform is built on silicon-on-insulator. The November 12, 2025 “CPO foundry technology” announcement explicitly describes silicon-to-silicon 300mm wafer bonding — stacking a SiPh PIC on a SiGe EIC, extending the BSI image-sensor bonding process. No glass interposer, no TGV, no glass carrier. Heterogeneous integration at TSEM means InP-on-Si (lasers via OpenLight) and SiPh-on-SiGe (3D-IC for CPO). All silicon and III-V.
One small qualifier: SiPh fabs use glass wafer carriers as process consumables for temporary bonding during wafer thinning. This is a de minimis, undisclosed consumable relationship — not a thesis. It’s the equivalent of saying “TSEM is a quartz play because they use quartz furnace tubes.”
On the Q4 2025 call (Feb 11, 2026) management disclosed that Intel has expressed intent not to perform on the Rio Rancho 300mm BCD arrangement and the parties are in mediation. Customer flows have been redirected to Fab 7 (Japan). TSEM’s 2028 model now excludes Fab 11X entirely.
This matters because Intel’s glass substrate HVM line is in Chandler, AZ — not Rio Rancho. The Fab 11X dispute is about silicon wafer capacity for BCD power, not glass. So even the one Intel-TSEM thread that remains in live dispute is orthogonal to glass.
Three specific developments — none of which are on the table in 2026:
TSEM adds TGV capability to its SiPh line — announces laser drilling, equips Newport Beach or Albuquerque with LPKF-style LIDE tools, qualifies a glass-based photonic interposer with a named customer (Broadcom, Marvell, NVIDIA). No sign of this. TSEM’s investment dollars are going to 300mm silicon SiPh capacity expansion ($920M capex, 5x capacity by end-2026) — not substrate diversification.
Glass wafers become the standard SiPh substrate — a technology shift from SOI to glass for photonic ICs. No one is calling for this. SOI is entrenched for SiPh PICs; glass is a complement (carrier, interposer) not a replacement.
TSEM announces a partnership with Corning, AGC, SCHOTT, or SKC/Absolics for co-development of a glass-based platform. None disclosed. The SiPh partnerships TSEM has announced (OpenLight for InP lasers, Xscape for on-chip lasers) are about III-V integration, not glass.
If any of these happen, revisit. They are plausible 2027+ storylines (particularly #1 if CPO scales and glass interposers become the lowest-cost photonic substrate), but nothing to underwrite today.
For Story A (glass-core packaging substrates): - Corning (GLW) — highest-quality pure-play on glass material for both stories; also display glass, fiber, specialty glass diversification - AGC (5201.T) — already in vault ([[5201]]); Japan’s biggest glass maker, EUV photomask blanks, display glass; glass substrate optionality - LPKF (LPK.DE) — cleanest equipment pure-play via LIDE laser drilling - SKC (011790.KS) via Absolics — the Covington GA ramp - Samsung Electro-Mechanics (009150.KS) — 2027 ramp, but expensive optionality - Ibiden (4062.T) — incumbent substrate leader pivoting to glass - Intel (INTC) — the anchor consumer; already in vault ([[INTC]])
For TGV inspection / metrology sub-theme: - Onto Innovation (ONTO) — already in vault ([[ONTO]]) - Camtek (CAMT) — already in vault ([[CAMT]])
For Story B (device/interposer photonic substrates) — this is genuinely small TAM in 2026: - Plan Optik (PLO.DE) — specialty glass wafers - Corning (GLW) — same name wins here too - SCHOTT — private
For the CPO / SiPh / optical interconnect thesis (which is what TSEM actually is): - TSEM itself — cleanest public SiPh foundry; already in vault ([[TSEM]]) - GFS — SiPh via AMF acquisition ([[GFS]]) - COHR, LITE, AVGO, MRVL — CPO device makers - 5801 Furukawa, 5802 Sumitomo, 6777 Santec, 6754 Anritsu — Japanese photonics supply chain; all in vault
TSEM is a silicon photonics / SiGe / CPO thesis, not a glass substrate thesis. The two stories are adjacent — both are downstream beneficiaries of the AI compute build-out and the bandwidth bottleneck — but they occupy different positions in the supply chain and have different catalysts, different customers, different equipment bases, and different timelines.
Any pitch deck, note, or tweet that frames TSEM as a glass substrate beneficiary is wrong or muddled. If the underwriting thesis is specifically glass, buy Corning, LPKF, AGC, or Samsung Electro-Mechanics. If the thesis is AI bandwidth / CPO, buy TSEM. Don’t mix them.
Related vault pages: [[TSEM]], [[5201]] AGC, [[5214]] NEG, [[7741]] Hoya, [[silicon-photonics-ai-datacenters]], [[optical-components-primer]], [[inp-sige-photonic-materials]], [[GFS]], [[INTC]], [[ONTO]], [[CAMT]]