Electronic Fiberglass Fabric in 2026: The Year 7628 Stopped Being Boring
My family has been pulling glass since 1978. I grew up around greige cloth and warp beams, and for most of my life 7628 was the boring product in the catalog. Six yuan a meter. Everybody had it. Nobody got excited about it.
That is over.
As I write this in October 2026, electronic fiberglass fabric is one of the tightest materials in the whole supply chain. Standard 7628 cloth that sold for about $0.60 a meter at the bottom of the cycle in the third quarter of 2025 is now trading near $1.47 a meter. That is a 140% move in twelve months, and the forecasts I see put it at $1.75 to $2.19 before the year is out.
We only weave 7628. That is the product we can make. Right now, that is not a weakness, and I want to explain why and what I think happens next.
Everybody blames the AI servers. They are only half right.
The story you read is that AI servers ate the cloth. There is truth in that. A traditional business server sits on an 8- to 14-layer board and eats 3 to 5 meters of glass cloth. An AI training box runs 16 to 30 layers, and some backplanes go past 70. One machine can swallow 30 to 50 meters. Hyperscalers are taking tens of thousands of those machines a quarter. Do that math, and the demand step is enormous.
But demand alone would not have emptied the standard market. What emptied it is what the big weavers did next. They pulled air-jet looms off 7628, 2116, and 1080 and pointed them at Low-Dk fabric. Low-Dk sells for 160 yuan a meter against 6 to 9 yuan for ordinary cloth. If you ran a weaving mill, you would do the same thing. So would I.
That is the paradox of this year. The shortage is worst in the cheap product. The electronic fiberglass fabric market is not short because 7628 became fashionable. It is short because everybody walked away from it to chase the high-margin grades. When Fuqiao pushed through its July increases, ordinary E-glass went up 30% and low-Dk2 only 15%. That number tells you exactly where the hole is.
The air-jet loom is the bottleneck nobody can unblock.
Here is the part that keeps me up.
More than 90% of the electronic-grade air-jet looms in the world come from a handful of machine builders in Japan. Lead time on a new loom today runs 18 to 24 months. So a mill that ordered equipment in early 2026, when the shortage was already obvious, does not see new output until late 2027. Maybe later.
I know what that wait feels like. We waited fourteen months for our last loom delivery. Fourteen months of watching a hot market run past us while a machine sat in a shipyard. You cannot rush a precision loom. Our run is several hundred picks a minute and holds yarn tension in grams. Push the speed, and you get warp streaks, weft bar, and broken filaments. Then the treater rejects the roll, and you have made expensive scrap.
Inventory across the industry has fallen to somewhere between three and ten days of cover. Normal is two to three months. Jushi was reported at about one week of ordinary fabric. When a company that size is running hand to mouth, you know the supply is genuinely gone.
Why we stayed on 7628
You might ask why a company that has been in fiberglass since 1978 did not chase the thin fabric and the low-DK money.
Honest answer. We did not have the bank or the customer base for 1080 and 2116. Low-Dk is a different chemistry problem, not just a different weave. And I watched a customer get burned. They tried to swap a thin cloth in for 7628 on a power board to shave weight. Two months later, they were fighting conductive anodic filament failures around the vias. Wrong tool, wrong job. A cheaper stackup is not cheap if you are re-spinning the board.
I will add one more thing about 7628, because it matters more every year. That 44 by 31 plain weave is not electrically invisible. The yarn bundles sit in a regular grid, and the resin sitting between them does not have the same Dk as the glass. On a tight differential pair, the weave period can show up as skew and impedance ripple. People call it the glass weave effect. It is the reason some fabricators pay extra for spread yarn or flat yarn. We do not make spread yarn. Standard 7628 is what comes off our looms. If your design is running past 25 Gbps per lane, pick 7628 on purpose, not by habit.
7628 is a heavy, thick, plain weave. One ply of it builds a dielectric section that three plies of 1080 cannot reach. For conventional FR-4, industrial multilayer, automotive control boards, and power supplies, that thickness is the entire point. We are good at it, and we have been good at it for a long time. Our warp and weft are ECG75 1/0 electronic-grade yarn, and the greige cloth runs through desizing and silane finishing on our own line.
I am not going to pretend we are a low-DK house. We are not. If you need DK down near 4.2 for an 112G channel, call Nittobo. But if you need electronic fiberglass fabric in the 7628 style on a truck at a price that has not been re-cut three times since spring, that is exactly what we do.
What 7628 actually is, in numbers
Numbers keep people honest. So here are ours, dry cloth, before resin.
Nominal thickness lands between 0.17 and 0.18 mm. Weight runs 203 to 210 grams per square meter. The weave is plain. Warp and weft are both ECG75 1/0, at 44 ends and 31 picks per inch, and we hold the count within plus or minus two. Tensile is at least 400 newtons per 25 mm in warp and 300 in weft. Dielectric strength is 40 kV per mm or better, and continuous service temperature is 550 degrees Celsius. We finish with an epoxy-compatible silane, and we keep the finish pickup in a tight window, because that one number decides whether your treater wets out clean or makes fish eyes.
Roll width is 1000 to 1270 mm as standard. Roll length runs 100 to 300 meters depending on your line. Moisture stays under 0.10%. If you laminate it into prepreg, plan on 46 to 51% resin content in a common build and roughly 0.20 to 0.22 mm of finished ply thickness.
That is the whole product. A buyer of electronic fiberglass fabric does not need a brochure. He needs those numbers, and he needs them to be the same next month.
In 2023, a yarn house shipped us a batch with moisture over 0.15% and alkali drifting past our line. On the spool it looked fine. On the loom it shed filaments, and the cloth came off with the weft bar. We scrapped nine hundred meters and ate it, then spent a month getting anyone to admit the number. That is why our moisture spec sits at 0.10% max and why we test every incoming cake. Your treater will not catch it for you either. It shows up in the prepreg yield a week later, and by then the argument is about whose fault it is. Nobody wins that argument. The roll is already dead.
The mistake I watch buyers make right now
Most electronic fiberglass fabric buyers I talk to are shopping on the wrong number.
A laminate maker in Jiangsu asked me in June for six months of volume at March pricing. I said no. We have a mill to feed and yarn to buy, and yarn is what moves every month now. He got angry. He bought at the top from a trader in Shenzhen instead. Two months later, the trader could not deliver because the trader never had an allocation. He had a phone number.
The second mistake costs more. Engineers are redesigning perfectly good boards to dodge the cloth price. Moving from 7628 to a thinner stack sounds clever until you re-run thermal cycling and the CAF test. Then you are three months behind, and your original 7628 quote has expired anyway. I get it. I still think it is the wrong trade when the market is this unstable.
The third mistake is small, and it wrecks treater lines. Buyers shop on price per meter and ignore finish pickup and yarn count tolerance. That cloth runs fine flat. Then it goes through a hot treat, and the warp streaks show up. I have watched a full roll get scrapped over two broken filament yarns nobody caught on incoming inspection. We check ours. We have to. We are not the biggest weaver in China, so our name rides on every drum we ship.
If you are quoting a laminate this quarter
Do three things.
Ask your weaver for the yarn source, the alkali number, and the moisture number in writing, per lot. Not a certificate from 2023. The actual lot.
Then hold enough physical stock to cover six weeks. Yarn lead times are not going back to normal this year, and every electronic fiberglass fabric supplier in China is quoting on a monthly basis right now, including us.
And stop asking for a full year at a locked price. Nobody can honor that. Anyone who signs it is either lying to you or planning to renegotiate in ninety days.
Since you are reading this from a weaver, I will say the obvious part. We price 7628 on a monthly basis against a fixed volume commitment. You get the allocation, we take the yarn risk on the front end, and neither side pretends the November price will look like the May price.
Where this goes
Jushi fired up its Huaian zero-carbon base in March. That line adds 100,000 tons of electronic yarn and 390 million meters of fabric a year, the largest single line anyone has built. Most of it feeds thin and high-end products, but it is real capacity. More mainland expansion lands through 2027. Outside the mainland, the pace is slower, and that matters because the bottleneck is looms, not furnaces.
My read. Tight through the rest of 2026. The standard grades will be the last to loosen. Somewhere in the second half of 2027, ordinary cloth should come off the peak. I do not expect it back at $0.60. The cost base moved, the yarn moved, and every buyer in the chain now knows that electronic fiberglass fabric can go scarce.
Until then, if you need electronic fiberglass fabric and you need it in 7628, come talk to us early. Not because we are the only ones holding cloth. Because in this market, the buyer who booked his slot in September is the one shipping in December.
Post time: Oct-10-2026
