2026.09.23 | AC | PCB | CCL Demand in GB300 & Rubin

摘要

• Assessments of Tier-1 vendor CCL demand, capacity, and PTFE demand trends.
• Capacity expansion cadence, monthly shipment updates, and related developments.

Expert Views:

How do you estimate material consumption for different types of PCB boards in servers? For instance, what are the typical surface areas and layer counts for orthogonal backplanes, switch boards, and NIC boards?

Material consumption can be estimated based on the area of an individual sheet of laminate. A standard sheet measures approximately 1.1 square meters. For specific boards within a server, dimensions and layer counts vary: orthogonal backplanes range from about 0.2 to 0.25 square meters, with 40 to 70 layers. Products entering mass production in 2H2026 are expected to be predominantly around 40-odd layers, whereas boards with around 70 layers will not enter mass production until 2027. Switch boards and NIC boards are much smaller, each measuring just 0.0-something square meters, typically with 20 to 30 layers. To convert PCB layer counts into material surface area, dividing total layer count by 2 gives an approximate count of material sheets used.

Regarding the area of orthogonal backplanes, some sources indicate dimensions close to 1 square meter, which differs from the 0.2–0.25 square meters you mentioned. Could you clarify its size or construction?

Sample specifications currently observed during the prototyping phase show that orthogonal backplane areas are indeed

1. 2 to 0.25 square meters. In theory, an orthogonal backplane is a single continuous board running vertically from the

bottom to the top of a rack. However, whether mass-production versions will be larger than the samples, or assembled by splicing multiple boards together, remains to be confirmed.

What is the company's current total capacity scale, and how is it distributed across its manufacturing bases?

The company currently operates five manufacturing bases, with total capacity of approximately 9.5 million sheets/month. The breakdown is as follows: Dongguan headquarters capacity is about 3.5 million sheets/month; Jiangxi base capacity is

1. 19 million sheets/month; Suzhou and Changshu bases combined have 2.2 million sheets/month; Nantong base has 800,000

sheets/month; and Shaanxi base has about 1-million-plus sheets/month.

Has the company's capacity changed in 2026 compared to 2025? And what are its capacity expansion plans for 2027?

From 2025 to 2026, the company's overall capacity has remained largely stable without broad-based expansions, apart from Thailand base, which added 300,000 sheets/month in capacity in 2025. The company plans to expand capacity in 2027, with an expected addition of 800,000 sheets/month in July 2027, followed by another 3 million to 5 million sheets/month by late 2027.

What products will the capacity added in 2027 primarily serve?

Currently, M8 is the mainstay for shipments. The capacity expansion planned for 2027 is divided mainly into two areas: one for M8 and M9 materials, and the other for ABF materials.

Which specific components in NVIDIA server racks does the company currently supply?

As of September 2026, the company has achieved mass production and supply for two main components: switch boards and NIC boards (such as boards used in CX9). It does not currently supply Compute motherboards, where M8 materials are primarily supplied by Elite Material.

What is Shengyi Technology's market share in materials for AI server NICs?

In the NIC segment, market share stands at around 30%.

For M8-grade materials used in NVIDIA AI server switches, what does the competitive landscape look like among major suppliers, and how is it expected to evolve from 2025 to 2027?

The landscape in 2025 was: Elite Material held a 50% to 60% share, Shengyi Technology accounted for about 10%, Doosan took 20%, and Panasonic held 10% to 20%. Moving into 2026, the landscape shifted to: Elite Material at 40%, Shengyi Technology rising to 20%, and Doosan and Panasonic together accounting for about 30%. Looking ahead to 2027, Elite Material is projected to maintain its 40% share, Shengyi Technology's share is expected to expand further to 30%, and Doosan and Panasonic will see their combined share decline.

At which product generation did Shengyi Technology enter the NVIDIA supply chain, and what did it initially supply? Did it participate in material supply for the Compute segment on GB200?

Shengyi Technology did not enter the GB200 supply chain. Shengyi Technology began penetrating only at the GB300 node; initially, it was not involved in NVIDIA's supply chain. Shengyi Technology began supplying from switch components, initially providing materials for a small portion of switch boards processed by Wus Printed Circuit, and used this opening to gradually expand its business.

From a consumption perspective, does single-board area differ between switch boards and compute boards?

Switch boards have relatively smaller surface areas, measuring just slightly larger than a book.

From GB200 and GB300 to the current Rubin architecture, how has CCL material consumption changed per server rack? Has board surface area expanded, or has layer count increased?

Growth in CCL consumption is primarily driven by increases in PCB layer count. In the GB200 era, PCBs had around a dozen layers; on GB300, this rose to 20 to 30 layers; and the current Rubin architecture is predominantly 30 to 40 layers. If an Orthogonal Backplane architecture is adopted on Rubin Ultra in the future, PCB layer count will reach 70 layers. Compared to GB200, board material consumption under current architectures is at least four to five times higher.

Since per-rack CCL material consumption grew four- to five-fold from GB200 to the Rubin architecture and material prices are also rising, why do industry report estimates of AI PCB market size growth (e.g., from 1.2 billion in 2025 to 2.7 billion in 2026) appear not to reflect that same magnitude?

The four- to five-fold growth discussed here refers to physical volume of CCL materials used. Final PCB value involves other factors, and pricing dynamics are not entirely clear. But purely from a material consumption standpoint, current usage has indeed increased four- to five-fold compared to GB200. At the same time, market size growth in the AI PCB sector has been significant, with suppliers broadening from only Wus Printed Circuit early on to multiple vendors including Shennan Circuits, Kinwong Electronic, and Wus Printed Circuit. Regarding market size projections, actual 2026 figures will only be confirmed once financial reports are published, and current estimates may lean conservative. This growth trend can be corroborated by supplier earnings; for instance, Doosan is listed in South Korea, and its financial reports show sharp growth in both revenue and profit.

How is Shengyi Technology's earnings growth tracking? How did it achieve high growth without significant capacity additions?

Taking 3Q2026 as an example, profit in a single quarter already exceeded the full-year total for last year (2025). Although overall capacity did not increase this year, Shengyi Technology optimized its product mix by shifting capacity previously used for mechanical applications to produce high-end materials. Last year (2025), output of high-end materials at M8 and M9 grades was around 100,000 to 200,000 sheets, whereas this year (2026) it has neared 1 million sheets. This growth represents an increase in material consumption driven by higher layer counts on AI server boards, offsetting declines in consumer PCB markets, thereby standing out prominently across the broader market.

What are Shengyi Technology's production targets for high-end materials, and what are its future capacity expansion plans?

In terms of high-end material output, volume was around 100,000-plus sheets in 2025, reached nearly 1 million sheets in 2026, and is expected to grow to 3 million sheets in 2027. To support business expansion, Shengyi Technology has invested 5 billion each in Dongguan and Suzhou to construct new production bases, totaling 10 billion in investment. This addition of 6 million sheets in monthly capacity is entirely dedicated to high-end products; once fully ramped by 2028, company-wide monthly capacity will reach 15 million sheets.

What is the company's current supply status on M9 products, particularly regarding mass production schedules and shipment expectations for orthogonal backplanes?

The company currently supplies mainly M8 materials. On M9 products, the company and Elite Material are the only two qualified mass-production suppliers for orthogonal backplanes. Orthogonal backplanes are expected to begin volume production in 4Q2026 and ramp fully in 2Q2027. Specifically for M9 material shipments, significant growth is anticipated in 4Q2026, with monthly shipments potentially reaching 100,000 to 200,000 sheets; after 2Q2027, monthly shipments are expected to surpass 500,000 to 600,000 sheets. Current M9 shipment volume is relatively small, at around 10,000 to 20,000 sheets per month, mainly used for prototyping and orthogonal backplane test samples.

In the Rubin platform, which components mainly use M9 materials, and who are the current suppliers?

In the Rubin platform, M9 materials are mainly used in 800G and higher-speed NIC boards and certain CPS boards, though overall consumption is modest. These M9 materials are currently supplied by Elite Material and Panasonic.

What are the dimensions of mid-planes in server racks, and what causes their dimensional variability?

Mid-plane areas range from roughly 0.03 to 0.05 square meters, representing relatively small dimensions. Substantial size variations arise primarily because racks are not standardized products. Different customers—such as Dell, Google, AMD, or Intel—customize designs based on differing compute requirements and system architectures (e.g., parallel versus serial configurations), leading to diverse dimensional specifications across boards inside racks.

Has the configuration for the Ultra platform been finalized?

The configuration for the Ultra platform is not yet finalized and is expected to be determined in 4Q2026.

In the Ultra platform, is orthogonal backplane a confirmed technology, and what are its application scope and necessity?

In the Ultra platform, orthogonal backplanes are 100% confirmed to be adopted and will be implemented across all racks. The necessity lies in the fact that as computing power increases and data transmission speeds accelerate, conventional mid-plane designs—which connect GPUs to rear transmission boards via cabling—face bottlenecks from excessive signal loss and distortion rates, while generating substantial heat that is difficult to dissipate. Orthogonal backplanes plug GPUs and transmission daughter cards into opposite sides of the board, using internal through-holes to transmit signals directly and reducing transmission distances to millimeter scale. This significantly enhances transmission rates and signal integrity, representing the only viable solution to meet AI server compute scaling demands.

In the upcoming Ultra platform, what is the company's anticipated share of M9 materials (whether applied to switches or orthogonal backplanes)?

For the Ultra platform, NVIDIA's total allocation guidance indicates the company's share will increase from 20% to 30%. In contrast, Elite Material's share will decline from 45% to 40%. While the specific allocation of materials across different boards has not been finalized, an increase in the company's total supply volume is certain.

In NVIDIA's supply chain, why has Elite Material's share declined while Shengyi Technology's share gained? Is it due to a price advantage?

This is primarily because Elite Material's capacity cannot fully satisfy NVIDIA's demand. As product layer counts rise to 40–70 layers, material consumption doubles compared to prior products like GB200, and Elite Material alone can no longer fulfill such enormous material demand.

Do competitors Doosan and Panasonic have expansion plans?

Panasonic is expanding with a new plant in Suzhou. Doosan likely has expansion plans as well, but has not yet entered the M9 material supply chain and is not yet a qualified M9 supplier.

Over the long term, what market share level is Shengyi Technology expected to reach in NVIDIA's supply chain?

The ceiling for market share is expected to be around 30%. This is because other vendors, such as Taiwan Union Technology Corporation (TUC) and Nan Ya PCB, are also actively seeking entry into NVIDIA's supply chain. Currently, TUC's primary partner is Google, and it has also expanded capacity in Changshu and Zhongshan, positioning it to compete for NVIDIA orders going forward. The future market structure is expected to see multiple vendors coexisting, each holding roughly 20% share, forming a landscape of four to six suppliers.

Will Shengyi Technology's market share rise alongside increased PTFE material adoption?

Shengyi Technology holds a technological lead in PTFE materials. Currently, NVIDIA has no other alternatives; thus, as long as an order specifies PTFE materials, it must be supplied by Shengyi Technology. Shengyi Technology currently faces no competitors in this domain.

Looking ahead, will PTFE material consumption continue to increase in post-Blackwell products or other types of boards?

PTFE consumption will definitely increase. Manufacturing bottlenecks currently constraining mass production are on the verge of breakthroughs. First, 400°C high-temperature laminating press equipment, currently supplied by German and Japanese vendors at roughly RMB 20 million per unit, has domestic prototypes undergoing testing at Shengquan Group; once successful, equipment will no longer be a bottleneck. Second, in drilling, diamond drill bit technology can substantially improve drilling capability on M9 materials. With breakthroughs in these two major technical bottlenecks, the likelihood of mass-producing M9 and PTFE hybrid lamination processes in 4Q this year is very high. With improved equipment precision, PTFE consumption is expected to expand further next year (2027) and beyond. From a technology roadmap perspective, future transmission rates will advance to 1.6T, 3.2T, or higher, at which point glass-cloth-containing materials—including M9 and M10—will fail to satisfy technical requirements, leaving PTFE as the only viable material. Therefore, PTFE adoption is an inevitable trend; it is merely a matter of time.

If equipment and drill bit bottlenecks are resolved, how long can Shengyi Technology sustain its technological lead in PTFE materials? Does this advantage stem primarily from formulation?

Shengyi Technology's technical advantage stems primarily from formulations. Even after equipment bottlenecks are resolved, Shengyi Technology will maintain its lead over competitors. Currently, Shengyi Technology leads Elite Material by a full product generation, representing a substantial gap.

In the ASIC space, has Shengyi Technology refrained from supplying related board materials? For example, which vendors primarily fulfill Google's orders?

Yes, Shengyi Technology does not supply any ASIC-related board materials. Google's orders are fulfilled primarily by vendors such as TUC. Due to Shengyi Technology's ownership background, Google will not use Shengyi Technology products under any circumstances, and the company is disqualified during the vendor profile setup stage.