2026.09.23 | AC | PCB | Rubin Ultra PTFE Materials Boom

摘要

· The RubinUltra platform's NVL576 switch boards have been upgraded to 18 boards, with electrical performance required to reach 337Gbps, driving PTFE materials to become the core intra-rack transmission solution for Scale-up architectures. · PTFE CCL exhibits a substantial cost advantage: PTFE resin (100–200 RMB/kg) is far cheaper than M10 (5,000–10,000 RMB/kg) and requires no expensive quartz cloth, with future market prices expected to fall below M9. · The primary bottlenecks reside in PCB manufacturing yields and dedicated high-temperature laminator equipment capacity, with equipment orders already booked through 2027–2028. · Shengyi Technology plans to expand its monthly PTFE CCL capacity; Shennan Circuits and Wus Printed Circuit serve as the main PCB suppliers for orthogonal backplanes. · PTFE adoption is broadening from switch boards into CPU motherboards, midplanes, and LPU boards, with an addressable CCL market estimated at tens of billions of RMB and a PCB market spanning tens of billions to hundreds of billions of RMB.

The following insights are from a senior industry expert:

Q&A

What does the current competitive landscape look like for PTFE materials, and which vendors are entering the market aside from Isola and Rogers?

In the copper-clad laminate (CCL) sector, beyond Isola, Rogers, and Taconic, players such as Elite Material (EMC), Taiwan Union Technology (TUC), ITEQ, and Doosan are progressively submitting PTFE-related solutions. Upstream film suppliers remain predominantly W. L. Gore, with no meaningful shifts. On the resin side, supplies consist primarily of PTFE resin from Dongyue Group, while the filler supplier is Novoray Corporation. Broadly speaking, the upstream vendor landscape has not seen major alterations compared to before.

How specifically are PTFE materials being implemented across the two Rubin Ultra platform variants (Kyber NVL144 and NVL576)?

The Rubin Ultra platform does indeed come in two versions. The Kyber NVL144 variant incorporates an orthogonal backplane, which represents the earliest trial application for PTFE materials. That orthogonal backplane first tested a PTFE lamination structure in 2024. While electrical performance met specifications, fabrication was exceptionally difficult and very few PCB makers could process it. Subsequent evaluations tested M9Q material, but its electrical performance lagged by 20%; M10 was tested as well. The currently evolved solution pairs a PTFE Core with M9Q prepreg (PP) layers in a 26x3 structure to reduce processing hurdles. Fabrication yields currently scrape 40% to 50%, but because few PCB fabricators can produce it, available capacity remains constrained.

The other variant is the Oberon-based NVL576 open rack, which houses eight NVL72 units linked across racks via CPO optical interconnects. To deliver Scale-up high-speed transmission speeds within the rack, its switch boards were increased from 9 to 18 boards, requiring electrical performance to rise accordingly to 337Gbps, which led to the adoption of PTFE materials. The current architecture laminates two 16-layer PTFE Cores using M9Q PP. Yields for an individual 16-layer board reach around 60%, and laminating them into a 32-layer board sustains a comparable yield level. End customers view this approach as viable for volume production.

This progress has altered PCB fabricators' postures. Previously, PCB makers offered minimal backing because PTFE was hard to process, yielded poorly, and consumed disproportionate line capacity. Now, however, once a competitor succeeds in fabrication, others are compelled to follow suit or risk trailing behind in future PTFE technology competition. Consequently, whether it is EMC and TUC who originally specialized in M9Q, or other PCB fabricators, firms are aggressively recruiting PTFE specialists to rapidly stand up relevant technical capabilities and capacity over the near term.

In the Rubin Ultra 144 (Kyber NVL144) system, what are the specific function, structure, and per-system PTFE material requirements for the orthogonal backplane?

The orthogonal backplane connects compute trays to switch trays, superseding the massive assemblies of copper cabling previously required. Designed much like a computer motherboard, it accommodates 18 compute trays plugged into the front and 9 switch trays plugged into the rear to achieve mutual interconnection, eliminating the need for more than 5,000 connecting cables.

The orthogonal backplane currently utilizes a 26x3 structure. Based on bill-of-materials calculations, a single orthogonal backplane consumes roughly 10 square meters of PTFE Core CCL and 10 square meters of M9Q PP. Because an individual Kyber NVL144 system employs two orthogonal backplanes, the total material consumption per system is double that figure.

What is the current unit price of PTFE CCL? How does it compare to M9 materials?

PTFE CCL does not yet have an established volume production price. In terms of performance, it outperforms M9, so it should logically command a higher price. In the initial phases of volume production around 2027, the market will likely face supply shortages, keeping pricing relatively elevated. Looking further out to 2028–2029, however, as additional CCL vendors enter the fray, PTFE CCL could become cheaper than M9 under competitive pressure, given that its underlying manufacturing costs are potentially lower than M9's. For reference, M9 currently trades in the market at approximately 2,000 to 3,000 RMB per square meter.

Why is the use of PTFE materials mandatory in the orthogonal backplane architecture?

Adopting PTFE materials on the orthogonal backplane is driven primarily by stringent electrical performance demands. This application requires Scale-up intra-rack transmission speeds, which necessitates materials with very low dielectric loss (Low Df). Currently, neither M9 nor M10 grade materials feature loss characteristics that satisfy this performance bar, prompting the industry to turn toward PTFE materials.

Taking the 144 server as an example, what are the orthogonal backplane PCB value, the PCB supplier base, and the material profile?

In the 144 server, the PCB value of the orthogonal backplane is high, with an individual board priced at roughly 500,000 RMB and each server utilizing two orthogonal backplanes. Currently, the leading PCB suppliers in this space are Shennan Circuits and Wus Printed Circuit. Notably, the 144 server represents the debut platform for M10 materials; the earlier standard Rubin release did not incorporate M10.

What is the unit price of M10 material, and who are its key upstream suppliers?

M10 material unit prices are estimated at 2,000 to 3,000 RMB per square meter. Incremental value in its upstream supply chain appears predominantly in resins and glass cloth. Primary resin suppliers include Dongcai Technology and Disai New Material. For glass cloth, Q-Fiber (quartz cloth) is typically utilized, with Feilihua serving as the primary vendor in this segment, although certain suppliers deploy designs that omit glass cloth entirely. Upstream silica powder suppliers remain unchanged, consisting chiefly of Novoray Corporation.

For the NVLink 576 cluster, what effect will the switch board upgrade have on PTFE CCL consumption? What are the projected PTFE consumption and dollar content per rack?

The NVLink 576 cluster links eight 72-unit systems via optical interconnects. The switch board count will jump from 9 to 18 boards, coupled with upgrades to the switch boards themselves, which will prompt a massive increase in PTFE usage. Under current projections, an individual NVLink 576 rack is estimated to require 400 square meters of PTFE CCL. Priced at 3,000 RMB per square meter, the total value of PTFE CCL in a single rack stands at roughly 1.2 million RMB.

Across the NVLink 576 cluster, which specific components utilize this 400 square meters of PTFE CCL? Factoring in scrap and manufacturing yields, what is the actual material consumption?

Within the NVLink 576 cluster, PTFE CCL goes primarily into orthogonal backplanes and switch boards, with the bulk concentrated across the 18 switch boards. Fully transitioning switch boards to PTFE designs is an active industry initiative that has demonstrated encouraging test results. The 400-square-meter figure is a theoretical calculation assuming all switch boards transition to PTFE. Factoring in current manufacturing yields of around 60%, along with losses from panel cutting utilization and edge trim, actual PTFE CCL consumption roughly doubles, approaching 800 to 1,000 square meters.

Beyond switch boards, could PTFE materials expand into other server assemblies going forward? If adopted, how much incremental demand would that unlock?

Yes. Proving out volume production on switch boards will bolster industry confidence in adopting PTFE materials, creating scope to expand into additional components over time. The industry is currently evaluating PTFE on CPU motherboards, midplanes, and LPU boards. If CPU switch motherboards shift to PTFE, consumption would rise by approximately 20% over current levels. Adoption on midplanes would add another 30% to 40%. LPU boards present the greatest volume upside: if converted to PTFE, their consumption would nearly double current levels, though overall LPU rack shipments may not match NVLink 576 rack volumes.

What is the current status of Avary Holding's PTFE validation on CPO switches?

Avary Holding is actively building out its PTFE portfolio and has brought in numerous fabrication specialists for this initiative. The company is currently running extensive qualification runs at its Huai'an facility while continuously sourcing and evaluating Low Df materials from multiple suppliers.

Given that the overall value of PTFE boards does not appear exceptionally massive, why are tier-one companies like Avary Holding deploying resources here?

Even though the value of PTFE laminates on an isolated basis may not appear staggering, firms like Avary Holding are likely positioning themselves out of longer-term strategic considerations.

For high-tier PCB makers like Avary Holding, PCB value per rack appears modest—for example, on an NVL576 footprint consuming 1,000 square meters, total rack value may only be two to three million RMB, far lower than the value of a switch board. What motivates fabricators to engage?

For top-tier fabricators like Avary Holding, the primary impetus stems from strategic partnership agreements signed with NVIDIA, which mandate joint development efforts. Looking at operational scale, overall GB-related (Blackwell GB200) business represents tens of billions of RMB. From the standpoint of corporate strategic growth, companies naturally maintain strong interest in these programs.

Regarding Avary Holding's product development roadmap, is work currently concentrated mainly on the CPO switch segment?

Avary Holding is indeed running trials on CPO switches, while concurrently testing 32-layer boards (in a 16x2 configuration) intended for switches.

Industry rumors suggest that because CPO requires higher board transmission speeds, previous architectures ran into technical issues, prompting trials with PTFE materials. Is this true?

This is largely accurate. The initial architecture ran into two primary hurdles: first, yield deficiencies in the CPO chips themselves; second, the subsequent pivot toward NPO (Near Packaged Optics) architectures, which necessitated upgrades to underlying PCB materials. These factors represent the core reason PTFE materials are required on CPO and NPO boards.

Regarding PTFE consumption within CPO switches, what is the consumption per switch unit? What is its incremental market headroom?

CPO switch consumption volumes are not particularly massive at present. A system integrates 8 CPO switches in total, with one per NVL72 rack, which is far lower in unit count than standard switches. Specifically, CPO switches account for 8 PCBs, whereas switches total 18 multiplied by 8, exceeding 100 PCBs. Thus, the incremental demand from CPO switches is comparatively modest, amounting to roughly 10% to 20% of the switch segment's volume. While there are only 8 boards, their layer counts are high, requiring nearly 5 square meters of PTFE CCL per PCB. High layer counts boost material usage per board, but because total board counts remain small, overall PTFE material consumption is limited.

In an NVL576 single-rack system footprint, does the 400-square-meter PCB consumption figure include the 8 CPO switches?

No, it does not. The 400-square-meter consumption figure refers exclusively to the switch board segment.

From an industry perspective, how large is the addressable market for PTFE materials? Over the next three to four years, will adoption be confined to switch trays in Rubin Ultra and next-gen platforms, capping overall market headroom?

PTFE addressable market size in the CCL segment is expected to reach tens of billions of RMB, representing a highly attractive market for laminate manufacturers. In the PCB segment, the market could reach tens of billions or even hundreds of billions of RMB, depending on the eventual breadth of downstream adoption. For CCL suppliers with typical annual revenues in the single-digit billions of RMB, this represents vast market capacity. Furthermore, technical barriers for PTFE materials are higher than standard materials, allowing gross margins to reach 40% to 50% or higher, well above conventional FR-4 CCL and PCB products. Driven by sizable market capacity and robust gross margins, industry participation is enthusiastic across the board.

Is PCB fabrication yield currently the primary bottleneck for industrial PTFE adoption? What are the cost and yield dynamics in CCL manufacturing?

Yes, the primary bottleneck today lies in PCB fabrication yields. Yields in CCL manufacturing are no longer the central issue, regularly reaching 80% to 90% depending on individual vendor manufacturing capabilities. One can consider CCL manufacturing relatively mature.

If CCL manufacturing is relatively mature, what is the defining constraint on large-scale PTFE adoption?

The linchpin is whether sufficient production capacity can be supplied once the technology is validated. PTFE laminates represent pioneering, first-of-their-kind products globally, requiring vendors to demonstrate patience during product development and deploy dedicated capital expenditure. Capacity expansion bottlenecks—especially in critical processing machinery—determine whether high-volume adoption can be achieved. Currently, numerous manufacturers are attempting to procure expansion equipment, but these tools face severe supply shortages. The single most critical piece of equipment is the laminator. High-end laminators are indispensable whether expanding capacity for ABF substrates, BT substrates, M8/M9 grade materials, or PTFE boards. Laminators designed for PTFE are specialized tools that cannot be shared with lines producing M8, M9, or FR-4; fabricators must construct new lines and procure dedicated laminators. Currently, only a handful of vendors worldwide—such as Kitagawa in Japan, Bürkle in Germany, and LAUFFER in Germany—can produce these high-end laminators. These equipment vendors are running at full capacity, with backlogs booked out to 2027 and 2028. For instance, Kitagawa's incremental press capacity will not come online until 2028; Bürkle is also expanding; LAUFFER is partnering domestically with Hefei Metalforming Intelligent Manufacturing to scale production, aiming to add 200 to 300 presses over the next one to two years. Upstream equipment constraints, compounded by aggressive baseline industry demand, have created an acute equipment shortage. As NVIDIA has noted, its systems incorporate numerous novel materials never previously deployed at global scale, such as PTFE and quartz cloth. Incumbent global capacity cannot instantly absorb such massive demand surges, and capacity buildouts lag far behind explosive demand growth, necessitating a buffer period.

Regarding the laminator expansion project between LAUFFER and Hefei Metalforming Intelligent Manufacturing, what are the planned capacity, equipment unit pricing, and customer mix?

The laminator expansion project between LAUFFER and Hefei Metalforming Intelligent Manufacturing is planned for an annual capacity of 200 to 300 units. Unit pricing varies based on equipment specifications and press dimensions. Dedicated PTFE laminators carry unit prices of roughly 5 million to 10 million RMB. For laminators handling M8 and M9 grade materials, smaller units are priced at around 2 million to 3 million RMB, while larger units run 3 million to 4 million RMB. The partnership covers high-end, large, and small form factors, with tailored configurations adjusted to customer dimension requirements. In terms of customer mix, high-temperature PTFE laminators are procured primarily by CCL manufacturers, whereas laminators for high-end M8 and M9 materials are purchased mostly by PCB fabricators.

From a cost breakdown standpoint, how do PTFE CCL and M9/M10 CCL differ across raw material consumption and procurement costs?

PTFE CCL and M9/M10 CCL exhibit stark differences in bill-of-materials costs, primarily across resin systems, glass cloth, and silica fillers. First, on resins, both material types consume similar amounts per square meter of CCL at roughly 0.1 to 0.2 kg, depending on thickness. Price disparities, however, are extreme: PTFE resin costs 100 to 200 RMB/kg, M9 resin runs 2,000 to 5,000 RMB/kg, and M10 resin reaches 5,000 to 10,000 RMB/kg. Second, on reinforcement materials, PTFE formulations do not use glass cloth, reducing this cost component to zero. Conversely, M9 and M10 materials generally require Q-cloth (quartz cloth), which costs about 300 RMB per ply, or 600 RMB per board if two plies are used. Finally, regarding silica fillers, PTFE materials require slightly higher silica loadings, but silica is relatively economical at approximately 300 to 500 RMB/kg. Consequently, within a single panel's cost structure, resins and glass cloth represent the dominant variance drivers, shifting unit panel costs by hundreds of RMB, whereas silica filler differences are modest, amounting to only tens of RMB.

In high-speed CCL, what are the specific consumption rates, prices, supplier landscape, and market size for silica powder?

In high-speed CCL, silica powder consumption per square meter of PTFE laminate stands at roughly 0.1 to 0.2 kg. Because PTFE eliminates glass cloth, its silica loading trends closer to 0.2 kg; M9 and M10 formulations trend closer to 0.1 kg because they incorporate glass cloth, with exact consumption dictated by dielectric thickness. High-end spherical silica powder commands roughly 300 to 500 RMB/kg. On the supply side, vendors capable of delivering such high-grade chemical-method spherical silica powder include Novoray Corporation, Lingwe Technology, and Kingboard Advanced Materials. Extrapolating from per-board loadings, total silica powder demand in high-speed CCL reaches thousands of metric tons.

Could you outline the basics of PCB layer counts, such as what 16x2 designates in switches, as well as layer configurations across different product tiers like Rubin and Ultra?

PCB layer count denotes the number of conductive circuitry planes embedded within the board. For instance, in switch applications, a 16x2 architecture entails fabricating two independent 16-layer sub-boards and then laminating and bonding them together to produce a finished 32-layer PCB. Regarding platform architectures, standard Rubin platforms typically utilize 26-layer or 28-layer PCBs, whereas Ultra platforms require 32-layer PCBs.

Taking a 32-layer Ultra platform PCB as an example, what does its internal stackup look like? If PTFE is implemented, how is it applied?

A 32-layer PCB does not contain 32 discrete CCL sheets. A representative stackup comprises 14 CCL sheets and 16 PP (prepreg) sheets, yielding a CCL-to-PP ratio close to 1:1. The prepreg serves as the bonding dielectric that laminates adjacent CCL cores together. If the Ultra platform adopts a PTFE solution, all 14 internal CCL cores are swapped out for PTFE CCL.

What are the specific layer count, material composition, and architecture of the orthogonal backplane?

The orthogonal backplane is a high-density 78-layer backplane. Its stackup integrates roughly 30-odd core laminates, all composed of PTFE materials.

What is the layer count on CPO switch boards?

A CPO switch incorporates 8 board cards in total, with each card running roughly 40 to 50 layers.

What are the layer count, material configuration, and PTFE demand profile for the midplane?

Midplanes currently run around 48 layers. Should they transition to PTFE in the future, it would unlock substantial incremental PTFE demand. Each PCB accounts for approximately 2 square meters of CCL consumption.

What drives the heavy CCL consumption seen in LPU racks?

Substantial CCL consumption in LPU racks is driven primarily by high PCB layer counts per rack and intense chip integration density, scaling to 256 chips per individual rack. However, LPU racks are deployed primarily in inference workloads, so overall rack procurement volumes may remain modest depending on customer adoption. PCBs in this rack form part of the compute tray cluster.

What is the single most critical checkpoint to monitor when tracking PTFE material adoption?

The primary metric for tracking PTFE progress is the fabrication yield rate achieved by PCB makers on PTFE boards. Currently, even top-tier market leaders like Wus Printed Circuit and Shennan Circuits achieve yields of only around 60%.

Why is market attention so heavily focused on laminator equipment?

Elevated market interest in laminators stems from two key dynamics: first, laminators used for PTFE CCL command high equipment value, running roughly double the cost of conventional laminators; second, equipment suppliers across the sector are operating at full capacity. As a result, the market is closely tracking the capacity expansions at Hefei Metalforming Intelligent Manufacturing and LAUFFER, with Hefei Metalforming Intelligent Manufacturing well positioned to capture a share of these incremental equipment orders.

What are Hefei Metalforming Intelligent Manufacturing's positioning and capacity plans in the PTFE laminator segment?

Hefei Metalforming Intelligent Manufacturing does not produce PTFE materials itself; rather, it specializes in building the dedicated laminators required to produce PTFE CCL. Its planned capacity of 200 to 300 units will be allocated toward manufacturing dedicated PTFE laminators based on order intake, supplying CCL manufacturers with active procurement demand, such as Shengyi Technology and Taiflex.

How much CCL capacity can a single PTFE laminator support?

An individual PTFE laminator can support approximately 20,000 to 40,000 square meters of monthly CCL capacity.

What is Shengyi Technology's current PTFE CCL capacity, and what are its future expansion plans?

Shengyi Technology currently operates PTFE CCL capacity of approximately 100,000 square meters per month. In response to future market demand, the company plans to expand its monthly capacity.

Where do the primary manufacturing capability differences lie between tier-one and trailing vendors in PTFE PCBs?

Currently, process capability gaps among PCB fabricators regarding PTFE board manufacturing are relatively narrow, and that technical divergence is expected to compress further over the next one to two years. The defining difference lies in capacity footprint: leading players like Shengyi Technology and Taiflex already possess operational PTFE CCL lines and capacity, whereas vendors with no prior exposure to PTFE lack dedicated lines and capacity. As for the PCB fabrication process itself, no specialized machinery is required.

What is causing the currently low manufacturing yields for PTFE PCBs?

Subdued yields for PTFE PCBs today stem primarily from the high intrinsic complexity of the fabrication process.