Invesaro
Investment thesis · Live

The AI revolution's glass ceiling is literal: low-loss laminate fabric sold out through 2027

AI hardware doesn't get built without low-loss glass fabric from a handful of firms: supply is fixed through 2027, prices climb every quarter, and the margin flows to the producers.

Published July 31, 2026 · 120-day horizon · supply and demand

Basket return
+37.1%
equal weight, since publication
The market over the same window
+3.0%
benchmark for this basket
Edge over the market
+34.2%
in percentage points
Causal chain: AI hardware forces boards with ultra-low signal loss → Low-Dk fabric output concentrated in a few furnaces → Fabric demand steps up with every hardware generation → A new furnace takes about 2 years to build, supply is fixed → Capacity sold out through 2027, quarterly hikes cascade into laminates → Nitto Boseki: fabric leader sets the price → Taiwanese CCL makers: Elite Material, ITEQ, TUC

What this thesis rests on

Each one is a statement that has to be true. We go back to every live thesis on a schedule and check these against public sources, so each leg carries either what we found or the date we look next.

Checked every 14 days. Last review August 28, 2026, next one September 11, 2026.

  1. Nitto Boseki's low-Dk and T-glass cloth capacity stays fully booked into calendar 2027, with no downward revision of that backlog disclosed through Q1 2027.

    operational · Supported

    A source confirms this · August 28, 2026

    FY2026 Q1 briefing (1 Aug 2026): electronic materials sales +29.5% to ¥14.6bn, operating profit +70% to ¥7.0bn; management said special glass sales are expanding at a pace exceeding production capacity, kept T-glass/NER-glass capacity expansion at FY2027-FY2028, and raised full-year guidance to ¥141.6bn sales / ¥30bn operating profit. No backlog cut disclosed.

    See the source ↗

  2. Low-Dk glass cloth prices rise in each of the two quarters after 31 July 2026, evidenced by announced price increases to CCL customers or reported ASP gains.

    financial · Open

    Checked August 28, 2026, nothing published either way

  3. Elite Material's gross margin in Q3 2026 is at least equal to its Q2 2026 gross margin, showing the glass cloth increases are passed through in full.

    financial · Open

    Checked August 28, 2026, nothing published either way

  4. No Chinese electronic glass yarn producer is qualified as a low-Dk cloth supplier at a top-five CCL maker before 31 March 2027.

    competitive · Open

    Checked August 28, 2026, nothing published either way

  5. AI accelerator boards and 800G/1.6T switch boards keep woven glass cloth reinforcement through Q1 2027, with no glass-free or reduced-ply design cutting cloth use per board.

    operational · Open

    Checked August 28, 2026, nothing published either way

Inside every printed circuit board in AI hardware, from 800G switches to accelerator boards, sits a material the market barely talks about: glass fabric, the reinforcement in copper clad laminate (CCL). At signalling speeds of 224 Gb/s per lane, ordinary E-glass generates too much loss, so designers have to reach for low dielectric constant (low-Dk) grades, woven from filament thinner than a human hair. Every hardware generation adds board layers and pushes fabric consumption per system higher, while quality requirements get tighter.

Supply of this material is extremely concentrated. Japan's Nitto Boseki leads the top grades (NE-glass, T-glass), backed up by a narrow field of Taiwanese producers. The barrier to entry is decades of know-how in glass chemistry and in weaving ultra-thin yarns, and a new furnace takes roughly two years to build and then has to run without interruption. The investment decisions were made before the AI boom, so the market has entered a phase of structural shortage: order books run into 2027, price lists move up quarter after quarter, and customers are accepting long-term contracts and prepayments just to lock in allocations.

This thesis, over time

What has happened since publication. Every entry is computed as of its own date, from price history.

August 28, 2026

Review: 1 of 5 assumptions confirmed

We went back to this thesis on day 28 of 120 and checked each of its 5 assumptions against public sources. Confirmed by the sources: 1. No news either way: 4. The equal-weight basket stands at +37.1% since publication day. The benchmark (.INX) returned +3.6% over the same window, so the basket is ahead of its market by 33.5 percentage points. 92 days remain until the verdict.

August 11, 2026

Basket crossed +20% since publication

This basket is at +22.9% since publication, 11 days after the thesis was posted. The benchmark (S&P 500) returned +3.2% over the same window, so the basket is ahead of its market by 19.7 percentage points. The strongest contributor is 6213 (+71.9%), the weakest is 3110 (+0.5%). This is not a change of status: the horizon keeps running and the verdict only lands once it expires.

August 4, 2026

Basket crossed +10% since publication

This basket is at +10.4% since publication, 4 days after the thesis was posted. The benchmark (S&P 500) returned +3.3% over the same window, so the basket is ahead of its market by 7.1 percentage points. The strongest contributor is 6213 (+17.9%), the weakest is 6274 (+2.0%). This is not a change of status: the horizon keeps running and the verdict only lands once it expires.

Open the interactive thesis

Interactive causal map, the basket charted against its market, and the full timeline.

Open in Invesaro →

New theses straight to your inbox

Subscribe and we will send you every new thesis and every verdict. No spam, one-click unsubscribe.

This is analysis, not investment advice and not a recommendation to buy or sell anything. We publish it and track it in public, mistakes included. Any decision is yours and yours alone.