Two distinct glass fiber shortages are running in parallel through 2026: electronic-grade T-glass cloth for AI server substrates, where a single supplier holds roughly 90% global share with no new capacity until mid-2027, and telecom optical fiber, where G.652.D and G.657.A2 grades are on allocation with China spot at approximately RMB 120 per fiber-km [S3][S5].
The bottleneck is not uniform across product forms. Construction-grade roving still trades in transparent tiers — US$0.95–1.35/kg at 50 kg MOQ from Changzhou Jlon, and US$1.40–1.85/kg at 500 kg MOQ from Jiangxi Suihua — while high-end electronic cloth and tight-tolerance single-mode fiber have moved into allocation-driven markets where lead time, not unit price, sets the negotiation [S2].
Where the Shortage Is Hardest: T-Glass for AI Substrates
T-glass cloth has become a structural constraint on AI server build-outs in 2026, with Nittobo controlling approximately 90% of global T-glass capacity and 60–70% of NER-glass capacity, and no new lines expected before mid-2027 [S5]. The material matters because AI rack designs — including cableless architectures with midplanes, orthogonal backplanes, and the inference-focused Rubin LPX rack — are pulling high-end glass fiber cloth demand sharply upward while the supply side is effectively fixed [S5]. Procurement teams that have treated T-glass as a commodity BOM line are now finding that the cloth drives the lead time of the entire server, not the GPU allocation [S6].
The substitution problem is technically narrow. Apple and other downstream players have explored Chinese glass fiber sources and lower-specification alternatives, but verification cycles are long because the specification window is extremely tight: fibers must be thinner than a human hair, perfectly round, and defect-free at sub-stratum scales [S4]. For procurement and design engineers, the operational consequence is that T-glass allocation decisions for 2026 platforms effectively had to be locked in late 2025, and 2027 build plans remain exposed to the same single-point supplier until Nittobo's next wave qualifies [S5][S9].
Optical Fiber: From Buyer's Market to Allocation
The telecom fiber market has flipped from a decade of overcapacity to what multiple industry sources describe as the tightest conditions in recent memory, with G.652.D tight on allocation and G.657.A2 — the bend-insensitive grade used in FTTH and high-density indoor builds — under similar pressure at roughly RMB 120/fiber-km ex-tax in China during May 2026 [S3][S7]. Lead times across single-mode and multimode grades have stretched to the point where some part numbers are not available to the open market in the near term, affecting nearly every category of buyer rather than a narrow segment [S10].
Demand is being pulled forward by AI data center campuses running 144-, 288-, and higher fiber-count MPO/MTP backbone trunks, against a 2024 announcement by Corning of a partnership with Nvidia to expand U.S. optical manufacturing capacity tenfold and lift fiber output by more than 50% — a signal that hyperscalers are now reaching directly into upstream supply rather than competing on the spot market [S3]. The supply response is slow because going from preform deposition through draw, coating, and proof-test typically takes 12–24 months from board approval to qualified output, so 2024 and 2025 expansion announcements are only beginning to ease specific bottlenecks now [S3][S7].
Adjacent Inputs Are Tightening in Parallel
Fiber cable cost is no longer set by glass alone: aramid yarn strength members, PBT and PE loose-tube and jacket compounds, water-blocking and filling gels, and LSZH compounds for indoor and riser constructions have all shown upward price pressure from late 2025 into early 2026, driven by feedstock costs, energy, and limited capacity in specific polymer grades [S3]. The practical effect is that even where fiber itself is allocated, finished cable lead time remains elevated and the all-in delivered cost is set by whichever component is the slowest to clear [S3][S10].
For buyers comparing offers, this means a vendor quote that ignores aramid and polymer exposure will understate the real risk window. Specifications for indoor and riser LSZH cables in particular should be reviewed for compound grade and flame-rating class, because LSZH supply is one of the more allocation-prone inputs alongside the glass itself [S3].
Criteria Comparison: Construction Roving vs T-Glass Cloth vs Single-Mode Fiber
Three glass-fiber product forms now sit in very different supply regimes, and the decision criteria for a procurement or spec engineer differ by row. Construction-grade roving from Chinese mills (US$0.95–1.85/kg at 50–500 kg MOQ) remains price-driven, multi-sourced, and short lead time; T-glass cloth for AI substrates is single-supplier-dominated at ~90% share with no new capacity before mid-2027 and is allocated rather than quoted; telecom single-mode fiber (G.652.D, G.657.A2) sits between, with China spot around RMB 120/fiber-km but with allocation common and finished-cable lead time set by aramid and polymer inputs rather than the glass [S2][S3][S5].
The selection implication: for FRP and construction use, qualify two suppliers and buy on price; for AI substrate cloth, secure allocation slots and accept that 2027 capacity is the earliest relief; for telecom and data-center fiber, treat aramid and LSZH compound exposure as part of the fiber-cable spec, not a separate purchase, because they are the binding constraints on finished lead time [S3][S5][S10]. A useful side-by-side reference for sourcing patterns is the China glass fiber manufacturers map with 2026 price bands and spec cutoffs, and the broader glass fiber supply-chain picture covering capacity, FRP demand, and sourcing risks provides context on how construction-grade and electronic-grade supply diverge.
Who This Shortage Is For — And Who Should Sit It Out
Allocation-driven sourcing applies to buyers in three lanes: AI server and accelerator platform builders needing T-glass cloth for advanced package substrates; hyperscale and FTTH operators pulling G.652.D and G.657.A2 single-mode fiber at scale; and defense or specialty cable buyers whose specs cannot be downgraded to commodity single-mode [S3][S5][S7]. These are the groups for which securing a production slot, accepting prepayment terms, and treating allocation as the priority — rather than waiting for a price correction — is the correct 2026 posture [S3].
The shortage is largely irrelevant for buyers of commodity E-glass roving for FRP pipe, tank, and pultrusion use, where Chinese supply remains multi-sourced and price-tiered in the US$0.95–1.85/kg band with 50–500 kg MOQ [S2]. It is also not the binding constraint for buyers of standard fiberglass mesh — common mesh sizes 4x4 mm, 5x5 mm, 4x5 mm, 10x10 mm, and 2.85x2.85 mm, weights 40–165 g/m², and roll lengths 50/100/200 m — which remains a construction commodity [S1].
Failure Modes and Engineering Constraints
Three failure modes recur in tight glass-fiber markets. First, single-supplier concentration: with Nittobo at ~90% of T-glass and 60–70% of NER-glass, any process event at one site cascades across the global AI substrate supply chain, and there is no qualified second source before mid-2027 [S5][S9]. Second, spec drift risk: lower-specification substitutes may clear basic mechanical tests but fail on the sub-stratum defect and roundness tolerances that AI package reliability requires, which is why Chinese alternative sources are being explored but not yet broadly qualified [S4]. Third, finished-cable bottlenecks masquerading as fiber shortages: aramid yarn, PBT/PE jacket compounds, water-blocking gels, and LSZH fire-performance compounds each carry their own capacity ceiling, so a "fiber available" signal does not guarantee a shippable cable on the requested date [S3][S10].
For a foundational reference on the base material and its properties, see the glass fiber encyclopedia entry, which covers the E-glass, T-glass, and NER-glass distinctions that define why substitution is not a like-for-like swap in this shortage cycle.
Sourcing Tactics and Trackable Signals
Two trackable signals will tell buyers whether the 2026 squeeze is easing: Nittobo's next capacity-announcement timeline relative to its stated mid-2027 first-output target, and the lead-time differential between Chinese spot G.652.D fiber and equivalent single-mode from second-source Asian and U.S. suppliers [S3][S5]. If Nittobo pulls the mid-2027 date forward or a second T-glass line qualifies, AI substrate allocation should loosen in the back half of 2027; if the date slips, expect T-glass premiums to persist into 2028 platform planning [S5][S9]. On the telecom side, watch the spread between ex-works China spot (around RMB 120/fiber-km for G.652.D in May 2026) and delivered North American or European all-in costs, because that spread is currently dominated by aramid and polymer exposure rather than the glass itself [S3][S10].
The underlying component specifications are covered under dc power supply, and switching power supply.