Foreword:

With the escalation of computing power density, the factors of current, heat, signals, and material stability all experience a simultaneous amplification. Components such as MLCC, PCB, copper foil, and electronic fabric, which were once overlooked, are now emerging as another cornerstone of the computing power industry chain.

The true driving force behind computing power lies in every abrupt change in current and every segment of high-speed signals.

Author | Fang Wensan

Image Source | Internet

Computing Power: Restructuring Industries and Shifting Value to Foundational Materials

As the density of GPU computing power doubles, power consumption rises proportionally, and signal transmission rates increase exponentially, prompting a comprehensive reconfiguration of power supply and cooling systems. This pressure cascades through the industry chain, ultimately impacting the most fundamental electronic components and materials.

Profits in the industry chain have traditionally gravitated towards the most rigid segments, with the core supply bottleneck in this computing power cycle centered on upstream foundational materials.

Downstream capacity adjustments offer greater flexibility, whereas upstream expansion in high-end copper foil, electronic fabric, and MLCC faces numerous constraints.

Electronic fabric, copper foil, PCB, and MLCC—four products originally from disparate niche markets—are now inextricably linked to the same growth trajectory driven by AI computing power.

Each time AI chip power consumption and signal rates escalate, higher demands are placed on the entire industry chain: electronic fabric must become thinner and more uniform, copper foil must have a lower profile, PCBs must incorporate more layers, and MLCCs must be used in larger quantities with enhanced performance.

The physical limits of materials and processes are continually being pushed, yet the pace of capacity expansion consistently lags behind the growth in demand.

Supply-demand conflicts are transmitted step-by-step along the chain from "fibrous materials—metal materials—manufacturing integration—passive components," with the closer to upstream foundational materials, the longer the expansion cycle, the fewer the participating manufacturers, and the weaker the supply elasticity.

Ultra-thin electronic fabric, ultra-low-profile copper foil, and high-end ceramic powders—these three types of materials at the forefront of the industry chain are precisely the most rigid segments in terms of supply-demand gaps across the entire computing power supply chain.

The transmission chain can be succinctly summarized as follows: increased GPU power consumption drives upgrades in power architecture, which in turn necessitate increases in the quantity and specifications of high-end MLCCs. Enhanced high-speed interconnection density drives increases in PCB layers, area, and material grades, while PCB upgrades push high-end copper foil, low-dielectric electronic fabric, and low-thermal-expansion electronic fabric into a constrained supply zone.

Any shortfall in any segment will be exacerbated by the high power consumption, high speed, and high density of AI servers.

The underestimated aspect lies here: the market perceives GPU orders but often overlooks the consumption of material certification, process yield, and capacity elasticity associated with these orders.

The Revaluation of MLCCs: Driven by Power Integrity

AI chips operate in low-voltage environments, where instantaneous current changes can reach extreme levels. GPU loads fluctuate rapidly between training, inference, and communication switching, necessitating the power system to maintain voltage stability within extremely short timeframes.

MLCCs play a crucial role in decoupling, filtering, and buffering current. The closer they are to the GPU, the shorter the current path, the lower the parasitic inductance, and the more effective the suppression of voltage fluctuations.

As high-computing-power chips become more high-frequency, high-density, and high-bandwidth, MLCCs must provide higher capacitance values and stronger reliability within smaller areas.

This transforms the value logic of MLCCs: general consumer electronics prioritize total volume and price, whereas AI servers emphasize specification combinations and positional value.

Positions near the GPU, CPU, VRM, power modules, and network chips often require products with higher capacity, smaller size, lower ESL, higher temperature resistance, and better consistency.

As server power supplies evolve from 12V to 48V and towards even higher voltage architectures, demand for high-voltage MLCCs will shift upwards accordingly.

While low-end MLCCs may remain in abundant supply, high-end server-grade MLCCs could face structural tightness—this is where the current MLCC market trends diverge from those of the smartphone era.

AI servers have propelled MLCCs to the forefront of power integrity; they remain small but now determine whether the system can operate stably under high loads.

PCB Upgrades: The High-Speed Infrastructure for Computing Clusters

High-end PCBs now serve comprehensive functions, including high-speed signal transmission, power path organization, thermal management coordination, and system structural connections.

Systems like GB200 NVL72 have organized computing power at the rack level, requiring large-bandwidth, low-latency communication between GPUs. The network side must accommodate evolutions to 800G and even 1.6T, with switching chips, network cards, optical modules, backplanes, midplanes, UBBs, OAMs, and other structures all raising the technical threshold for PCBs.

The upgrade direction for PCBs is clear: more layers, denser routing, lower dielectric loss, stricter impedance control, more complex backdrilling and microvia processing, and higher requirements for thermal stability and warpage control.

Prismark data indicates that driven by demand for AI infrastructure, the global bare-board PCB market will grow from $73.6 billion in 2024 to $85.8 billion in 2025, a 16.7% increase.

Leading AI server and network PCB suppliers are expected to see revenue growth exceeding 50% in 2025, while most PCB suppliers will experience sales growth below 10%.

This divergence is critical: the real profit elasticity in the PCB industry is concentrated among high-end manufacturers capable of meeting demand for AI servers, high-speed switching, and advanced packaging.

The more computing power moves towards clusters, the more PCBs resemble the infrastructure of a highway—they don't determine the brand of the car but do determine whether the car can run at high speed, safely, and with low loss.

Copper Foil and Electronic Fabric Price Hikes: Low Loss as a Prerequisite for Participation

Delving further upstream into PCBs, one of the core materials is copper-clad laminate, composed of resin, electronic fabric, and other reinforcing materials along with copper foil—the foundational material for PCBs.

With AI servers imposing higher demands on PCBs, pressure naturally transmits to copper foil and electronic fabric.

The key to copper foil lies not just in conductivity; when high-speed signals transmit in high-frequency environments, copper foil surface roughness affects insertion loss.

The higher the signal frequency, the more pronounced the skin effect becomes, with current concentrating on the conductor surface and making it harder to ignore the signal perturbations caused by rough surfaces.

The value of high-end HVLP and VLP copper foils stems precisely from their lower roughness, better signal integrity, and compatibility with high-frequency, high-speed copper-clad laminate systems—they have carved out an independent growth curve driven by AI servers and high-speed networks.

The changes in electronic fabric are even more pronounced. Ordinary glass fabric primarily provides mechanical support and insulation, whereas electronic fabric for AI servers must simultaneously meet low dielectric loss, low thermal expansion, high dimensional stability, and high consistency.

Low-Dk electronic fabric is utilized in high-speed transmission scenarios like AI server motherboards, switches, and routers, while low-CTE T-glass is more commonly used in AI chip packaging substrates to mitigate thermal expansion mismatches between large-sized chips and substrates.

As AI chip substrate areas expand—from Hopper to Blackwell to Rubin—both substrate area and layer count are increasing.

AI server motherboard layer counts are also evolving, from around 20 to 28 layers in 2024–2025 to 24 to 40 layers in 2026–2027.

The combination of increased layer counts, expanded area, and higher speeds is simultaneously amplifying electronic fabric consumption alongside material grade upgrades.

Certification-Based Scarcity Replaces Scale-Based Expansion

Bottlenecks in the AI computing power industry chain have extended to system-level material capabilities. Advanced processes and advanced packaging remain core, but the deliverability of computing power increasingly relies on the consistent synergy of power, interconnection, substrates, board materials, and other components.

Once materials enter mainstream platforms, their lifecycle and stickiness are often stronger than those of ordinary consumer electronics components—this is "certification-based scarcity."

Its scarcity arises not just from capacity but also from process windows, customer validation, yield ramp-ups, and system synergy.

Low-end capacity can expand rapidly, whereas high-end capacity must be defined by platforms, validated by customers, and repeatedly confirmed by production data.

The market tends to evaluate copper foil, electronic fabric, PCBs, and MLCCs within the framework of traditional cyclical commodities, but AI servers are propelling them into an intermediate zone between semiconductor materials and core system components.

From a domestic industry perspective, this hidden driver is equally significant. China holds advantages in PCB output value and manufacturing scale, but competition in the high-end AI server chain is not merely about expanding capacity.

High-end HDI, high-layer-count high-speed boards, high-frequency low-loss copper-clad laminates, HVLP copper foil, Low Dk/Q glass fabric, and low-CTE packaging materials—each segment requires joint validation by material companies, equipment companies, board manufacturers, server customers, and chip platforms.

This also explains why overseas leaders are raising prices, expanding capacity, and witnessing customers compete for capacity simultaneously—cloud providers, GPU vendors, packaging firms, and board manufacturers are all vying for production certainty in the next generation of AI platforms.

Conclusion

The rise of AI computing power brings far more than just an increase in usage volume; it systematically elevates performance standards across the entire industry chain, unifying previously dispersed niche segments under the mainline of computing power upgrades.

This AI-driven industry chain is redrawing the focus of industrial value.

Partial reference sources: CITIC Securities: "Review of the Electronic Fabric & Copper-Clad Laminate Industry Chain: Bottlenecks in Upstream Computing Power Materials Become Prominent, Tightening Supply-Demand Dynamics for Specialty Electronic Fabrics," Citi Securities: "Global PCB & CCL Industry Outlook: AI Server Architecture Iteration Drives Comprehensive Value Growth for High-Frequency High-Speed Copper-Clad Laminates," IDC China: "2026 Global AI Server Shipment Forecast and Computing Power Hardware Industry Chain White Paper"