Shenwan Hongyuan: AI compute inflation shifts to PCB and capacitors

🔑 Enhanced Key Takeaways

•AI servers increasingly rely on High-Density Interconnect (HDI) PCBs, which feature 20-30+ layers, microvias, and fine-line etching to support high-speed signal transmission and dense component integration for GPUs and other accelerators.
•Multilayer Ceramic Capacitors (MLCCs) are critical for maintaining power stability, regulating voltage, and suppressing noise in AI hardware, with next-generation platforms like Nvidia's Rubin architecture demanding a substantial increase in MLCC units per board (e.g., 12,000 units compared to 6,500 on the GB200).
•The surging demand for high-end MLCCs is creating a significant supply bottleneck, with leading manufacturers such as Taiyo Yuden and Murata Manufacturing reporting unprecedented demand levels that are straining production capacity and increasing the risk of supply chain disruptions.
•The global PCB market is projected for robust growth, with AI infrastructure driving a 13.9% increase in 2026, and the copper-clad laminate (CCL) market, a key material for PCBs, expected to exceed US$21.5 billion in 2026 with an annual growth rate of 34.2% due to AI-driven demand.
•Beyond PCBs and capacitors, other critical bottlenecks emerging in the AI supply chain for 2026 include High-Bandwidth Memory (HBM), power infrastructure, and optical networking, indicating a broader shift in investment focus across the AI hardware ecosystem.

🛠️ Technical Deep Dive

PCBs for AI:
- Require High-Density Interconnect (HDI) technology, characterized by microvias (typically less than 150 microns in diameter), fine lines and spaces (often below 100μm, sometimes less than 50 microns wide), and high layer counts (20 to 30+ layers).
- Must support extremely high-speed signals (exceeding 10 Gbps, with future demands for 800G/1.6T optical modules) and manage significant thermal challenges, as AI accelerators can dissipate 200-500W in compact footprints.
- Advanced materials, such as ultra-low loss copper clad laminates (e.g., NVIDIA's M10), are being developed to reduce signal loss by 30-40% compared to traditional FR-4 PCBs.
- Designs are increasingly blurring the line between traditional PCBs and IC substrates, sometimes requiring "PCB as interposer" functionality with 2-mil or finer traces and spaces.
- Incorporate intelligent thermal management solutions, including interfaces for liquid cooling systems and integrated temperature sensors.
Capacitors for AI:
- Multilayer Ceramic Capacitors (MLCCs) are essential for power supply decoupling, voltage stabilization, high-frequency noise filtering, transient current response, and electromagnetic interference (EMI) suppression.
- AI accelerators draw hundreds to thousands of amperes at very low core voltages (around 0.8 V), necessitating ultra-high capacitance MLCCs in very small case sizes (e.g., ≥47 µF in 0402, ≥100 µF in 0603) for effective close-in decoupling.
- Conductive polymer capacitors are valued for their low Equivalent Series Resistance (ESR) and high safety, providing stable capacitance for high-speed current delivery to processors.
- Supercapacitors (or ultracapacitors) are becoming crucial for AI data centers to manage highly dynamic power profiles, offering high power density and extremely fast charge/discharge cycles to buffer sudden energy spikes and smooth voltage fluctuations, complementing battery-based UPS systems for millisecond-scale transient suppression.
- AI servers often utilize 48V power distribution to minimize current and power loss, requiring capacitors capable of efficient voltage conversion.

🔮 Future ImplicationsAI analysis grounded in cited sources

The shift in investment focus to PCBs and capacitors will drive significant R&D and manufacturing capacity expansion in these component sectors.

The reported 'scary' levels of demand and projected market growth for these components indicate a clear need for greater supply and technological advancement to meet the escalating requirements of AI hardware.

AI hardware development will increasingly be constrained by the availability and technological limits of passive components and specialized PCB materials, rather than solely by leading-edge chips.

The article and search results highlight PCBs and capacitors as emerging bottlenecks, alongside High-Bandwidth Memory (HBM), suggesting that the entire hardware ecosystem needs to evolve to support AI's demanding computational needs.

The rising demand for advanced PCBs and capacitors will lead to increased pricing and potentially impact the overall cost and deployment timelines of AI infrastructure.

Reports indicate that MLCC manufacturers are prioritizing high-end supply, leading to price increases, and the inherent complexity of manufacturing high-density interconnect (HDI) PCBs also contributes to higher production costs.

⏳ Timeline

2021-2024

AI market characterized by HBM3 adoption, with supply largely meeting demand and primary focus on GPUs.

2025

Global copper-clad laminate (CCL) market reached US$16.02 billion, indicating early growth in PCB materials.

2025-2026

Structural shortage of memory and compute components, including HBM, emerged, shifting market focus to other bottlenecks.

2026-03

NVIDIA initiated supplier testing for M10, a new Copper Clad Laminate (CCL) material for its next-generation Rubin platform, signaling the PCB industry's entry into a high-frequency, high-speed era.

2026-05-20

Reports indicated Nvidia's next-generation Rubin architecture would require significantly more MLCCs (12,000 units per board) compared to previous platforms.

2026-05-29

Taiyo Yuden, a major MLCC manufacturer, reported 'scary' levels of demand for high-end AI server components, highlighting capacity strains.

2026-05-30

Goldman Sachs projected an unprecedented upcycle in multi-layer ceramic capacitors (MLCCs) extending through 2030, driven by AI infrastructure spending.

Shenwan Hongyuan: AI compute inflation shifts to PCB and capacitors

⚡ 30-Second TL;DR

Key Points

🧠 Deep Insight

🔑 Enhanced Key Takeaways

🛠️ Technical Deep Dive

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline

📎 Sources (29)

👉Related Updates

CATL Q2 earnings show growth despite margin pressure

Anthropic Launches Claude Opus 5 with Enhanced Performance

Professor Zeng Ming on AI-Native Business Strategy