2026.09.22 | AC | PCB | PCB Trends: mSAP & Pricing

摘要

1. NVIDIA Rubin Platform PCB Value Breakdown: Compute boards (26-layer HDI) account for $50,400 per rack, with Victory

Giant Technology holding a dominant 50%-60% share; mid-planes (44-layer) and switch boards (32-layer) are led by Wus Printed Circuit.

1. Severe PCB Raw Material Price Volatility: Laminate prices rose by over 100% cumulatively in 2026, primarily driven by

thin glass fabric capacity constraints from Japan's Toyota looms (lead times extended to two years); price increases are expected to continue through June 2027.

1. mSAP Process Emerges as Core Incremental Driver for Optical Modules: 800G/1.6T optical module PCB unit prices reach

RMB 100-220, prompting leading vendors such as Shennan Circuits and Wus Printed Circuit to expand capacity on a large scale.

Expert Insights:

Could you discuss the current business footprints, technical focus areas, and capacity comparisons among mainstream vendors in compute PCBs?

In compute PCBs, leading vendors have distinct strategic focuses. Wus Printed Circuit primarily concentrates on high-layer-count boards, including server motherboards, NVIDIA mid-planes, and switch boards; while it also manufactures compute boards, they are not its primary business focus. Victory Giant Technology has the opposite focus, centering primarily on HDI boards, particularly compute boards. Avary Holding only began entering the AI field in 2025; its business previously centered on handset HDI, IC substrates, and related products, serving customers mainly like Apple.

Shennan Circuits focuses primarily on the domestic Chinese market. Its Guangzhou plant mainly produces IC substrates for NVIDIA, while its main AI board customers are Google and Huawei. Shengyi Electronics' primary AI customer is Amazon, with about half of its AI-related capacity dedicated to Amazon. Fibocom Wireless is also involved, though its specific customer base remains unclear; its traditional business mainly supplies high-layer-count boards to traditional server customers both domestically and abroad, such as Dell and Inspur.

Founder Technology is reportedly qualifying NVIDIA's HDI products, but its output in 2026 is expected to remain modest. Shenzhen Kinwong Electronic's business in 2026 continues to center on high-layer-count boards, primarily supplying mid-planes and switch boards for NVIDIA as a tier-two supplier, with Wus Printed Circuit serving as the tier-one supplier. The year 2026 marks the first year Shenzhen Kinwong Electronic ramps up volume in this space.

Currently, capacity expansion across the NVIDIA supply chain is also predominantly concentrated among these vendors.

Taking NVIDIA's Rubin platform as an example, what board types make up its PCB architecture, and what are their respective content values, technical specifications, and supplier share distributions?

Taking NVIDIA's latest Rubin platform as an example, the rack PCB architecture and dollar content breakdown are as follows:

• Compute Board: This represents the largest portion of value. It is a 26-layer HDI board utilizing M8-grade materials. Priced at roughly $1,400 per unit with 36 boards per rack, it accounts for approximately $50,400 in content per rack. In terms of supplier allocation, Victory Giant Technology holds the dominant share.
• Switch Board: This is a 32-layer board using M8.5 (M8U)-grade materials. Priced at about $1,450 per unit with 9 boards per rack, it accounts for roughly $13,050 in content per rack. The primary supplier is Wus Printed Circuit with about a 40% share; Shenzhen Kinwong Electronic accounts for roughly 30%; the remaining 30% is mostly secured by Victory Giant Technology, likely around 20%, with the possibility of 10% allocated to other vendors.
• Mid-plane: This is a 44-layer board using M9-grade materials. Priced at approximately $1,500 per unit with 18 boards per rack, it represents about $27,000 in content per rack. The primary supplier is likewise Wus Printed Circuit with about a 40% share; Shenzhen Kinwong Electronic accounts for roughly 30%; the remaining 30% is expected to be split between Victory Giant Technology (around 20%), Avary Holding, and others.

Beyond NVIDIA, what are Shengyi Electronics' positioning and market shares in compute boards and server boards across other cloud computing giants like Google, Meta, and Amazon?

Shengyi Electronics has completed qualifications with major cloud customers including Google, Meta, and Amazon. Its business is not limited to NVIDIA compute boards; it also supplies products directly to these hyperscalers. For instance, in Amazon's server board business, including 3D boards, Shengyi Electronics holds over 50% share. For Google, Meta, and Amazon compute board requirements, Shengyi Electronics is likewise a primary supplier, capturing the majority of share.

What was Avary Holding's entry timing into the AI server PCB market, and what is its current competitive strategy?

Avary Holding entered the AI server PCB market starting in 2025, representing a relatively late entry. Its primary motivation was seeing profit growth and capacity expansion trends among domestic Chinese vendors in this space. The current competitive landscape largely depends on how much capacity Avary Holding can bring online and how much market share it can win from incumbent suppliers.

What are Shenzhen Kinwong Electronic's business progress and expectations with these hyperscale customers?

Currently, although companies such as Google, Meta, and Amazon have rack product shipments, large-scale volume ramp-up has not yet materialized. Shenzhen Kinwong Electronic is actively pursuing business with these accounts; for example, it has provided samples for Amazon server motherboards, though mass production has not started, and has developed high-layer-count server motherboards for Meta.

Since early 2026, what have been the material price adjustment frequencies and pricing mechanisms in the PCB industry? Has price pass-through progressed smoothly?

Since early 2026, PCB raw material prices have risen practically every month, with high-speed materials adjusting on a slightly slower cadence of roughly once every two to three months. The price pass-through dynamic differed noticeably across phases:

Prior to May, because the market underestimated sustained material cost hikes, price pass-through was not smooth. For example, a planned 10% price hike in March resulted in only 3% to 5% being accepted after negotiations with major accounts, forcing the company to absorb 5% to 7% of the cost increase internally and causing losses on certain orders. The May round saw a similar dynamic; a planned 10% to 15% increase was not fully accepted by customers either, typically resulting in a 50/50 split, or shared at ratios of 60/40, 70/30, or even 80/20, with only small customers accepting 100% price pass-through.

The inflection point came in June. The market broadly recognized that material prices were continuously climbing and supplies were constrained; for instance, material suppliers began capping purchase volumes, limiting single orders to no more than 2,000 sheets or cutting monthly allocations proportionally (a 10%-20% reduction for large accounts and a 50% cut for small accounts). Against this backdrop, an industry consensus emerged: starting in June, all new and repeat orders could smoothly pass cost increases entirely along to downstream customers. If raw material prices continue to rise in subsequent months, prices are adjusted accordingly even for unfulfilled orders or orders where materials have not yet been prepared.

Has the industry reached a tacit understanding on spontaneous monthly price increases? How have quotation validity windows changed compared to historical norms?

Yes, the industry has essentially reached a tacit understanding. As long as upstream material suppliers hike prices, PCB makers spontaneously adjust their product quotes. Correspondingly, quotation validity windows have contracted dramatically. Currently, quote validity for prototype facilities is typically just one to three days, while conventional facilities have generally shortened it to half a month. In contrast, during periods of stable material prices, quote validity was usually one month, and even long-term for products such as automotive boards. Later, validity was first trimmed to one month, which still failed to hedge cost volatility risks, leading to a further reduction to half a month. Currently, many vendors in the market maintain validity periods of only one week or three to five days to cope with rapid raw material price shifts. If customers do not place orders within the validity window, quotes must be recalculated subsequently.

What is the final quoting unit for PCB products? How do cost structures differ between double-sided and multilayer boards?

Regardless of the PCB type, quotes provided to customers are ultimately denominated in "pieces/panels"—that is, the price per piece. This unit price is derived from the square-meter price combined with the customer's specified panel dimensions and batch quantities.

Cost structures vary by layer count. Taking double-sided boards as an example, core material costs (CCL laminate costs) account for roughly 50% to 60% of total cost. For multilayer boards, such as eight-layer boards, core material costs (including core boards, PP prepreg, and copper foil) represent a relatively lower share of around 40%, while the remaining 60% covers labor, utilities, administrative overhead, scrap, and other expenses. Once all costs are calculated, the target profit margin is added to establish the final market quotation.

Among various PCB raw material costs, which component is currently experiencing the steepest price hikes?

The steepest price increases are currently in base laminates—specifically copper-clad laminates and prepreg used for multilayer lamination—both of which have seen price hikes virtually every month. In addition, other key materials remain elevated. After rising to RMB 100,000 per ton, copper foil prices have not experienced a notable retracement and currently linger above RMB 90,000 per ton. Meanwhile, gold prices have not shown any meaningful decline either.

What has been the cumulative price increase for base laminates year-to-date?

Since the beginning of 2026, base laminate prices have more than doubled. Specifically, through May, monthly price increases averaged around 10%; after June, monthly increases expanded to roughly 15%, with PP (prepreg) posting a 20% jump. On a blended basis, the average monthly price increase has been about 15%, and cumulative gains through September have exceeded 100%.

Who are the major suppliers in the CCL (copper-clad laminate) industry, and what is the underlying root cause of current material shortages?

There are numerous suppliers in the CCL industry. Domestically in China, they primarily include Shengyi Technology, Kingboard Holdings, Yinha, Shanghai Nanya, and several smaller players. Kingboard Laminates mainly produces low-end materials. Taiwanese vendors include Elite Material (EMC), Taiwan Union Technology (TUC), ITEQ Corporation, and Nan Ya Plastics. Internationally, major players include Panasonic of Japan, Doosan of South Korea, and Isola of the United States. The root cause of current material shortages is the constrained supply of glass fabric, especially thin fabrics used in AI server boards, high-layer-count boards, and HDI boards, such as models 1080, 1067, or 2116. Surging demand for compute boards in AI has triggered massive consumption of thin fabrics used in these high-layer-count boards, exceeding current supply. Material vendors tend to allocate limited thin fabric capacity toward producing higher-margin high-speed materials, which restricts standard material output and exacerbates overall supply tightness.

What are the specific bottlenecks limiting thin glass fabric capacity, and how long is this cycle of price increases expected to persist?

The capacity bottleneck in thin glass fabric stems primarily from loom equipment supply. The looms required to weave such thin fabrics are exclusively supplied by Japan's Toyota, whereas domestic Chinese manufacturers can only produce machines for thick fabrics (such as model 7628, with thickness above 0.2 mm). In late 2025 and early 2026, delivery lead times for Japanese looms were approximately 3 to 6 months; however, as of August and September 2026, lead times have stretched to two years due to capacity constraints. Consequently, even for new thin fabric lines initiated in late 2025 or early 2026, commercial operation will not occur until June 2027 at the earliest. Therefore, this round of price increases is expected to potentially persist through late June 2027. If compute board demand growth moderates by then, tight material conditions may ease in the second half of 2027.

What have been Kingboard Laminates' recent market dynamics and operating strategies?

Kingboard Laminates has recently begun cutting prices proactively. Taking its A0-grade or A3-grade materials used in consumer electronics as an example, selling prices stood at roughly RMB 340-plus in July, falling to RMB 289 by early August. Market rumors suggest the price cut occurred because two major customers halted procurements due to product quality issues; combined demand from these two accounts reached hundreds of thousands to over a million sheets monthly, creating idle capacity. Pricing strategy was another driver. Prior to July, Kingboard Laminates' prices were RMB 20 to 30 higher than comparable products from Kingboard Holdings, which carries a higher quality reputation in the industry. Its strategy appeared to leverage raw material (e.g., glass fabric, resin) cost inflation and tight market conditions to attach excess profit margins above costs, realizing substantial markups. Although prices have dropped, its profitability in 2026 remains solid due to the earlier substantial price increases.

What are current net margin levels for compute board businesses (such as Rubin platform-related products)?

Net profit margins for compute board operations generally achieve 25% to 30%. Even when yield rates and material utilization rates are sub-optimal and scrap rates are relatively high, net margins exceeding 25% remain achievable.

Beyond NVIDIA's demand, what are the application status and expansion trajectories of the mSAP process in areas such as optical modules?

The mSAP process is currently a primary focus of industry capital expenditure, applied predominantly in optical modules. Leading vendors including Shenzhen Kinwong Electronic, Avary Holding, Shennan Circuits, Wus Printed Circuit, and Redboard are all actively expanding capacity. Specifically, Avary plans to build 36 lines; Shenzhen Kinwong Electronic plans to add 15 lines in 2026 and another 10 in 2027. Shennan Circuits and Wus Printed Circuit each plan to deploy roughly 15 lines. Optical modules generate immense demand for mSAP; while 400G optical modules rarely use it, a portion of 800G modules have adopted it, and 1.6T and 3.2T optical modules almost universally adopt the mSAP process.

Compared to conventional processes, what technical advantages and manufacturing complexities does the mSAP process bring?

The mSAP process primarily resolves fine-pitch trace fabrication challenges that traditional subtractive processes cannot meet. PCB manufacturing capabilities broadly fall into three tiers: line widths exceeding 3 mil (roughly 75 microns) can be produced by the vast majority of facilities; line widths above 2 mil (roughly 50 microns) can be handled by most HDI plants or newer facilities. However, when line width and spacing drop below 2 mil—namely sub-40 microns—traditional lines cannot cope, requiring fine-line processes like mSAP (modified Semi-Additive Process). As chip integration increases and routing tightens, 800G optical modules require line width/spacing of 25 to 40 microns, while

1. 6T modules demand 15 to 20 microns. Achieving this precision requires comprehensive equipment upgrades across LDI

(laser direct imaging), etching, and developing tools to dedicated high-end systems, driving costs up sharply. For example, standard LDI equipment exposing line widths above 2 mil costs around RMB 1 million-plus, whereas tools delivering sub-40-micron exposure precision jump to RMB 3 million or over RMB 5 million. Equipment investments across other stages, such as electroplating, rise even more steeply.

How high are equipment capital expenditures for the M-SAP process, and how does this affect the competitive landscape?

Equipment expenditures for the M-SAP process are substantial. Taking electroplating lines as an example, a standard line costs roughly RMB 5 million to 9 million, whereas an M-SAP electroplating line costs at least RMB 20 million even when sourced from TKK (Tech-Shine); when sourced from brands like Amatek, costs reach tens of millions of RMB. This multi-fold increase in equipment expenditure substantially raises entry barriers. Consequently, smaller facilities struggle to enter AI-related PCB manufacturing due to capital constraints, concentrating competition among a few leading publicly listed enterprises. For these large players, investments in new greenfield facilities routinely reach several billion or even tens of billions of RMB, requiring expansion support through large-scale equity financings or stock listings. In contrast, smaller listed peers struggle to sustain such massive capital commitments even if they can raise several hundred million RMB, putting them at a structural competitive disadvantage.

What are the unit prices for M-SAP PCBs used in 800G and 1.6T optical modules?

The unit price of M-SAP PCBs used in 800G optical modules ranges roughly from RMB 60 to 120, with an average pre-tax price of around RMB 100. For 1.6T optical modules, PCB unit prices stand at roughly RMB 200 to 220 following price increases. The broader range spans RMB 160 to 240, varying with design architectures and HDI layer counts across vendors; for example, 1.6T optical module PCB prices procured by Innolight fall within the RMB 200 to 220 range.

Given that vendors including Avary, Shenzhen Kinwong Electronic, and Shennan Circuits all have large-scale expansion plans, is there a risk of future oversupply in the M-SAP market?

Based on current expansion plans, roughly 150 production lines will come online if all planned additions materialize. Projections indicate that 150 lines could deliver annual capacity of 300 million optical module PCBs, whereas 2027 market demand is projected at 200 million units, implying capacity would exceed demand by roughly 50%. Therefore, while vendors have drafted aggressive expansion plans, actual deployment will likely adjust based on incoming order momentum rather than installing all planned capacity at once. A phased approach is expected: for instance, if a vendor plans ten lines, it will likely install five first, monitor utilization and demand, and only then decide whether to proceed with the remaining five. Other vendors are expected to adopt similar prudent strategies.

When is technology enabling direct chip packaging on PCBs expected to reach commercial viability?

Packaging chips directly onto PCBs implies that the surface-layer trace precision of the board must match IC substrate levels, requiring MSAP or SAP processes for outer or surface layers. The key hurdle lies in material breakthroughs rather than PCB fabrication capabilities. Specifically, it requires a material with a low Z-axis coefficient of thermal expansion approaching ABF, BT resin, or glass substrates to replace conventional FR-4. Such a material would resolve thermal expansion mismatch between chips and PCBs. Once commercialized, packaging and substrate intermediate steps can be eliminated, dramatically lowering costs while shifting technical complexity entirely into PCB manufacturing—essentially integrating substrate functionality onto the surface layer of the PCB. Sample materials exist today, but Z-axis thermal expansion coefficients still fall short of requirements.

What are the application prospects for glass substrates or ceramic materials in direct chip packaging architectures?

Glass substrates represent a critical material for realizing direct chip packaging. In the substrate domain, glass cores have already begun seeing adoption, such as using glass cores in intermediate layers followed by ABF lamination. Future adoption hinges on whether glass base material costs can decline; if costs become manageable, glass bases could be laminated directly onto PCB surfaces with chips packaged directly on top, addressing heat dissipation and Z-axis expansion issues simultaneously. Ceramic materials can also be used in chip packaging, particularly for chips with extreme heat dissipation requirements. In fact, embedding chips directly into PCBs already has commercial precedents, such as IGBT modules. Today, IGBT chips no longer use standalone boards; instead, they are packaged inside copper blocks, which are embedded directly into the surface layers of circuit boards such as OBCs. Companies including Shennan Circuits have mass-produced these products for customers like Infineon and BYD, with voltage ratings reaching 800V. Embedding copper blocks or aluminum nitride (ceramic) carriers represents a form of directly integrating chips into PCBs and has already entered volume shipments.

What are Bomin Electronics' customer profile, order backlog, capacity footprint, and expansion plans?

Bomin Electronics' customer base is primarily overseas-oriented; it maintains a US sales team of over a dozen staff generating tens of millions of RMB in annual revenue. The company's order book is fully committed through the end of 2026, with current orders already booking into 2027. In terms of capacity, its Shenzhen facility currently runs at over 30,000 square meters per month, the Nanjing facility at 150,000 to 180,000 square meters, and the Zhuhai facility (acquired from Shuohong) at roughly 50,000 square meters, bringing combined monthly capacity to about 230,000 square meters. The company continues to expand: the Zhuhai plant has headroom for another 50,000 square meters, the Thailand plant is expected to add 50,000 square meters, and the Nanjing plant may add another line. Monthly capacity is projected to increase by at least 100,000 square meters in 2027. Regarding product mix, Bomin Electronics primarily manufactures mid-to-low-end through-hole boards with modest technical barriers; HDI boards are currently undergoing only 1-step and 2-step sample trial runs at the Shenzhen facility.