AI Materials on the Attack: Upgrading to M9/M10, Copper-Clad Laminate Moving Toward Higher Performance

```

AsAI computing power nodes are comprehensively upgraded from H100 (M7) to GB200 (M8), and continue to advance towards the Rubin platform (M9/M10), copper clad laminates, as the core substrate for “signal transmission” in AI servers, are undergoing a paradigm shift from “traditional cyclical products” to “strategic materials for computing power.” The amount of CCL per AI server is 3-5 times that of traditional devices, and the iteration frequency of high-end materials (M8→M9→M10) has changed from the traditional 5-8 years to 2-3 years. This material upgrade is rapidly restructuring the industry’s value distribution at an unprecedented pace.

In2026, a critical window for global computing power infrastructure iteration, copper clad laminate (CCL), as the core substrate of printed circuit boards (PCBs), is experiencing its most profound technological leap since the beginning of the industry. As AI servers make a full leap from H100 (M7 grade) to GB200 (M8 grade) and stride toward the large-scale demand for M9/M10 after 2026, CCL is no longer a “cyclical bulk product” in the traditional sense, but has become a “high-performance strategic material” that determines the interconnection efficiency of AI computing power clusters.

I. What is happening? Material upgrade continues

The entire industry isadvancing toward the large-scale M9/M10 demand after 2026. CCL is no longer a “cyclical bulk product” in the traditional sense, but has become a “high-performance strategic material” that determines the interconnection efficiency of AI computing power clusters.

Core points:

① Technical side: 224G high-speed interconnect technology requires CCL dielectric constant (Dk) and dielectric loss (Df) to enter an “extremely low loss” blind spot. Iteration of M9/M10 materials is essentially a challenge to the physical limits of high-frequency electromagnetic waves.

② Fundamentals: The “rise in both volume and price” caused by supply-demand imbalance is ongoing. The CCL price hike wave starting at the end of 2025 heralds a shift from industry losses and elimination to a high-boom expansion period.

③ Industry competition: Although Taiwanese leaders (EMC, TUC, etc.) hold high-end market dominance, mainland Chinese manufacturers (Shengyi Technology, Nan Ya New Material, Wazam New Material, etc.) are speeding up replacement in the M7/M8 field, and are embracing a golden window to challenge M9.

④ Industry logic: Watch for the CCL value in the AI server BOM cost growing by 3-5 times, as well as the domestic price premium for upstream specialty resins, ultra-thin copper foil, and low-loss glass cloth.

Although globalCCL capacity is huge, there are very few manufacturers that can stably supply M7/M8 in large volumes and develop M9 samples. In the global high-frequency high-speed CCL market, EMC (Elite Material), Doosan, and TUC together account for over 80% share.

Copper clad laminate (CCL) is the core substrate for PCBs, made from copper foil, resin, glass fiber cloth, and other materials, and is responsible for core signal transmission. In scenarios with high requirements for signal integrity and heat dissipation, such as AI servers, 5G/6G base stations, and data centers, CCL’s dielectric properties directly determine the quality of system-wide signal transmission.

High-endCCL can be categorized into different grades such as M7/M8/M9 according to signal transmission performance. The higher the grade, the lower the loss of the board. Panasonic’s material system is currently the most commonly used industry classification: M7 corresponds to Df about 0.002, M8 corresponds to Df ≤ 0.001, M9 corresponds to Df ≤ 0.0007. For AI computing boards, AI will drive the use of the full series of high-speed copper clad foils from M2-M8, and large-scale application of M9 is imminent.

A singleAI server uses 3-5 times more CCL than traditional servers. For major boards, high-speed CCL must be made into high-multilayer boards and 2-5 stage HDI boards, with a trend towards even higher levels. The evolution from 112G to 224G standards requires the Dk/Df of high-frequency high-speed CCL to be optimized from 3.2/0.0012 to 3.0/0.0009, aiming to sharply reduce energy attenuation and delay in ultra-high frequency signal transmission by pushing the dielectric properties to their limits.

To understand the significance of this numerical change, one must understandthe physical meaning of Dk and Df. Dk (dielectric constant) affects signal propagation speed, Df (dielectric loss factor) directly determines signal attenuation. At a 112Gbps SerDes rate, the insertion loss budget for PCB traces is already extremely tight; when the rate doubles to 224Gbps, the loss budget shrinks further, requiring Df to drop from 0.0012 to 0.0009, a reduction of up to 25%. This demands a systemic redesign of CCL in terms of material formula, resin system, and glass fiber cloth matching.

M9 CCL has a significant improvement in high-frequency performance over M8 and is about to lead a new high-speed bandwidth revolution, with single-channel interface speeds rising to 224Gbps. M9 series CCL’s Df is even lower than the M8 series, meaning that the widely-used PPO/PPE high-frequency high-speed resins can barely meet M9 requirements; breakthroughs in dielectric performance have further driven upgrades in upstream resin materials. M9 CCL may use more hydrocarbon resins and special hydrocarbon resins.

The Df gap between M8 and M9 is significant: M8 Df is 0.001, while M9 Df drops to ≤0.0007, a 30% reduction; M9 Dk is further reduced to the 2.6-2.8 range.

NVIDIA hasstarted testing the next-generation CCL material (M10) with PCB manufacturers, targeting the orthogonal backplane and switch blade mainboards of the Rubin Ultra and Feynman platforms. Sampling will start in Q1 2026 with initial test results expected in Q2 2026. The main target applications of M10 include: (1) orthogonal backplanes (midplane) designed to replace the current slot architecture; (2) switch blade mainboards for Rubin Ultra and Feynman platforms. These applications are at the core of NVIDIA’s next-generation AI server architecture and set much higher requirements for PCB material performance than current solutions.

II. Why is it important? Industry supply and demand restructuring

EMC expectsthe high-end CCL market to accelerate from 2024-2027 with a 40% CAGR, higher than the 21% CAGR in 2018-2021. This acceleration is mainly driven by relentless demand growth for high-speed CCL in AI servers, 5G, data centers, and other cutting-edge fields.

On a broader global CCL market level, according toQYR, global CCL sales in 2024 reached $17.58 billion, expected to reach $23.96 billion by 2031, with 2025-2031 CAGR of 4.6%. High-end market growth is much higher than overall market growth, reflecting how AI computing demand is reshaping the growth structure of the CCL industry.

On the supply side, production bottlenecks face three constraints.

① Constraint one: Structural shortage of raw materials

Core raw materials for CCL—copper foil, electronic-grade glass fiber cloth, resin—are all in short supply.

Copper foil: In 2025, the average copper foil price jumped from 82,000 RMB/ton in September to over 92,000 RMB/ton by December. With global copper mine suspensions and processing fee cuts at smelters, a rigid “tight ore – tight smelting – tight finished product” logic has solidified along the supply chain.

Electronic-grade glass fiber cloth: AI computing power demand is forcing manufacturers to shift toward high-end categories such as LDK glass fiber cloth, quartz cloth, and other low-CTE products, causing structural shortages of common 1080, 2116, and regular 7628 standard cloths. As of December 2025, spot lead times have stretched beyond one month. Nitto Boseki raised glass fiber prices 20% across the board in August 2025, with industry-wide follow-up price hike expectations. Rapid AI server demand growth in 2026 is expected to bring phased price hikes of 10% to high-end cloth, and pull up middle and low-end product price centers.

Resin: Driven by Middle East tensions, prices for epoxy resin, natural gas,and chemical raw materials like TBBA have risen sharply and supply is tight. On April 3, top CCL maker Kingboard sent a notice, raising plate and PP prepreg prices by 10% due to “sharp price surges and tight supply” in upstream resin and glass fiber cloth. Linzhou Guangyuan, Chongqing International, etc., raised 7628 cloth prices on April 1, and resin manufacturers plan monthly price increases of at least 5% starting April.

② Constraint two: Slow release of high-end CCL capacity

High-endCCL uses complex resin systems, glass fiber cloth, and copper foil matching, making effective production capacity slow to release. Compounded by limited supply of ultra-thin copper foil and high-end fiber cloth upstream, a stage of acute shortages has formed. Routine order lead times have grown from 7 days to 20-45 days, and most manufacturers have adopted order suspension or quantity-limited modes.

③ Constraint three: High industry concentration in CCL itself

According toPrismark, for major global CCL companies in 2023, the top five are Kingboard (14.7%), Shengyi Technology (14.0%), EMC (10.3%), Nan Ya Plastics (9.2%), Panasonic (6.6%), with CR5 at 54.8%. In high-performance CCL, in 2024 the top three are EMC, Doosan, and TUC with CR3 up to 81.4%. Mainland China’s top Shengyi Technology holds 5.7%, and Nan Ya New Material 1.0%.

On the demand side, three drivers spark explosive growth.

① Engine one: AI servers—the core incremental market

A singleAI server uses 3-5 times more CCL than traditional servers. NVIDIA’s computing node has fully upgraded from H100 (M7) to GB200 (M8), and M9 grade CCL may deliver large volumes in 2026. With each platform generation, there is a marked increase in CCL consumption and a leap in material grade.

EMC holdsan absolute edge in the AI server CCL market with a 70% AWS share, Google TPU share rising from 20% to 50%, and exclusive supply for Meta. In 2026, AI server revenue is projected to reach 48%. TUC holds a 30% AWS share, has entered the Google/Meta chain, with AI server revenue expected at 31% in 2026.

② Engine two: Switches and optical modules

Switches and optical modules requirehigh-speed materials M6, M7, M8, and even M9 to make high-multilayer PCBs and 2-5 stage HDI boards. At 200G-400G port speed, the industry shifts to seventh-gen materials (M7); to support 800G Ethernet ports, suppliers have released eighth-gen M8 materials (Df<0.002). From the second half of 2026, platforms such as Vera Rubin CPX server and 1.6T switches will use M9-grade quartz materials.

③ Engine three: Automotive electronics and IC substrates

High-endHDI benefits from AI terminal integration demand. According to Sullivan Research, global market size is projected to reach $16.9 billion in 2029; global automotive PCB market may reach $12.2 billion in 2030. For IC substrates, handled by high prosperity in downstream memory, BT substrate lead times have grown to 16-20 weeks, offering a window for domestic manufacturers to enter.

III. What’s next to watch? Where is the core value

The current market cycle highly resemblesthe “one price per day, limited supply” extreme boom of 2020. Since December 2025, Kingboard, Nan Ya, and others have issued dense price adjustment notices, with weekly CCL price increases reaching 10%-20%.

There are three major causes for restricted capacity on the supply side: (1) High-end production depends on imported equipment and precision processes, with long ramp-up periods; (2) Upstream electronic cloth production tilts towards low dielectric, ultra-thin grades, squeezing out supply of standard 7628 cloth for ordinaryCCL; (3) Copper foil production is diverted to lithium battery foil, causing copper foil for electronic circuits to become tight.

On the demand side, high-end application markets show very strong growth momentum.AI servers, 800G switches, automotive electronics, and IC substrates are all growing in tandem, with CCL demand showing “rising volume and price.” BT substrate lead times have grown to 16-20 weeks, offering a rare window for domestic suppliers to be introduced.

We believe this round ofCCL supply-demand imbalance will last longer than before, mainly because demand for AI servers and 800G switches will still be accelerating through 2026-2027, while high-end CCL capacity release is limited by upstream raw materials and equipment bottlenecks and cannot quickly fill the gap.

The CCL industry is experiencing a paradigm shift from“mature cyclical products” to “strategic computing power materials.” We divide this leap into four phases:

Phase 1 (2023-2024): AI demand germination period. Represented by the popularization of NVIDIA H100, demand for high-end CCL from AI servers becomes explicit. Overseas leaders such as EMC benefit first; Chinese manufacturers mainly follow. M7/M8 materials are mainstream at this stage.

Phase 2 (2025-2026): Supply-demand mismatch accelerates. AI server demand explodes, M8 becomes GB200 standard, upstream shortages of raw materials, CCL prices rise rapidly. EMC’s M9 is exclusively supplied to NVIDIA; CR3 market share reaches 81.4%. Nan Ya New Material’s M6-M8 are supplied in bulk to domestic top computing clients, entering the mainstream. We are now in the core acceleration phase.

Phase 3 (2026-2028): Domestic substitution window. M10 test suppliers diversify, domestic manufacturers begin entering the certification system of NVIDIA and other top overseas clients. Nan Ya New Material launches M10 worldwide, Wazam New Material’s Extreme Low Loss participates in overseas certification, Yongqiang Technology’s Gallop9Q joins next-generation certification. The core conflict in this phase is “certification progress” and “capacity ramping speed.”

Phase 4 (post-2028): Pattern reshaping period. The high-end CCL market shifts from oligopoly by a handful of firms such as EMC to diverse competition among “overseas leaders + emerging domestic players.” Enterprises with technical strength and customer certification will share in a trillion-yuan high-end CCL market.

Key conclusions are as follows—

Conclusion 1:AI computing demand is restructuring the value distribution of the CCL industry. Per server CCL use is 3-5 times traditional equipment, and rapid M7→M8→M9→M10 iteration keeps high-end CCL prices soaring. Enterprises with high-end CCL capacity and customer certification will embrace boom cycles of volume and price increases.

Conclusion 2: The supply-demand gap is forming a structural resonance; the price rally is likely to span2026-2027. Copper foil, electronic cloth, and resin prices are rising in tandem with tight supply, CCL lead times stretch to 20-45 days, and most manufacturers are limiting orders; on the demand side, AI servers, 800G switches, automotive electronics, and IC substrates are all breaking out at once, which will drive CCL company profitability to continue improving.

Conclusion 3: The window for domestic substitution has fully opened. The global high-performance CCL CR3 is as high as 81.4% in 2024, but diversified M10 test suppliers indicate that top clients like NVIDIA are seeking supply chain diversity. Domestic firms, with faster technical response and local material supporting abilities, are likely to accelerate access to core computing power supply chains at home and abroad.

Conclusion 4: High-endCCL material upgrades drive coordinated development of the upstream industry chain. M9/M10 materials require even higher standards of resin, glass cloth, and copper foil; there is vast potential for domestic substitution of specialty hydrocarbon resins, low-dk glass fiber cloth, HVLP copper foil, etc., upstream.

Risk Disclosure and DisclaimerThe market has risks. Investment needs caution. This article does not constitute personal investment advice, nor does it consider the special investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their specific circumstances. Invest accordingly at your own risk. ```