AI Demand Spillover: From Memory Chips to PCB Materials, the Electronics Supply Chain is Experiencing a New Round of Structural Tightness

Introduction

The impact of AI on the semiconductor industry has long been limited not just to GPUs or HBM. Recently, information from multiple electronic manufacturing service providers and component distributors shows that AI demand is penetrating every corner of the supply chain: prices for DRAM, NAND, HBM, controllers, PMICs, and other chips are broadly rising, with some categories seeing double-digit increases; passive components and packaging substrates are also facing price revaluation; and even conventional FR-4 PCB materials are beginning to experience supply tightness.

This is not an isolated price hike, but rather a systemic chain reaction caused by the reallocation of upstream capacity under the wave of AI infrastructure investment. As giants such as TSMC, Samsung, and SK Hynix prioritize advanced processes and memory capacity for AI chips, the resources available to traditional consumer electronics, automotive electronics, and industrial electronics are being squeezed out.

This article will analyze how AI demand is reshaping the electronic component supply chain from the dimensions of technology roadmaps, supply chain transmission, competitive landscape, and regional impact, and answer the question: is this round of tightness a short-term fluctuation or a structural inflection point?

Background: The Siphon Effect of AI Capital Expenditure

The growth in demand for AI servers and accelerator cards has changed the flow of resources in the semiconductor industry. Data center GPUs, AI ASICs, HBM memory, and supporting power management and networking chips occupy the highest-quality capacity at wafer fabs and packaging and testing facilities.

According to observations from industry research institutions TrendForce and SemiAnalysis, global AI server shipments will grow by more than 60% year over year in 2025, while the wafer consumption growth rate of AI chips is far higher than that of traditional logic chips. This demand intensity leads upstream manufacturers to prioritize high-margin, high-certainty AI orders when allocating capacity.

At the same time, the memory industry is experiencing significant divergence. As the most critical companion memory for AI chips, HBM capacity continues to be prioritized by Samsung, SK Hynix, and Micron, squeezing the supply of conventional DRAM such as DDR5 and LPDDR. NAND, on the other hand, faces a different supply-demand landscape due to sluggish consumer electronics recovery and capacity expansion. This divergence has caused memory prices to experience a tale of two extremes.

Deep Analysis: How AI Demand Is Transmitted to the Entire Supply Chain

1. Memory Chips: HBM Squeezes Ordinary DRAM, and Prices Begin to Move in Tandem

The average DRAM capacity installed in AI servers is 6 to 8 times that of ordinary servers, with HBM accounting for most of the increase. In order to meet HBM capacity demand, memory manufacturers have shifted part of their DRAM wafer capacity to HBM, limiting the incremental supply of standard DRAM such as DDR5 and LPDDR5.

Recently, spot prices for standard DRAM have risen for several consecutive weeks, with some niche DRAM segments seeing double-digit increases. As for NAND, since AI servers mainly rely on large-capacity HDDs and QLC SSDs, the pull from incremental demand is relatively limited. Combined with manufacturers' capacity expansion in 2025, NAND prices have instead tended to stabilize or even weaken.This divergence means that the storage supply chain is no longer in an overall boom, but is showing clear structural differences. For buyers, risks need to be assessed based on specific product models, rather than making decisions based on a blanket "memory going up or down" view.

2. Power Management ICs and Controllers: AI Power Architecture Increases Demand

AI chip power consumption continues to climb, with GPU power per card evolving from 350W to over 1000W. This directly drives demand for high-performance power management ICs (PMICs), DrMOS, multi-phase controllers, and server VRMs.

Since these chips mostly use mature process nodes, but have high design complexity and long certification cycles, the influx of AI orders has led IDM and Fabless manufacturers to prioritize capacity for server power chips, thereby extending lead times for PMICs needed in consumer electronics and industrial applications.

Industry sources indicate that lead times for some PMICs and controller ICs have extended from the normal 6-8 weeks to 12-16 weeks, and some suppliers have even suspended accepting new orders, only able to serve existing customers. This change in prioritization has a particularly noticeable impact on small and medium-sized OEMs.

3. Passive Components and Substrates: Demand Transmitted Upstream

AI servers require 2-3 times the number of MLCCs used in ordinary servers, with greater demand for high-capacitance, high-temperature-resistant specifications. Capacity allocation at manufacturers such as Murata and Samsung Electro-Mechanics is also tilted toward AI, leading to slight price increases and lead-time fluctuations for conventional MLCCs and chip resistors.

Packaging substrates are another bottleneck. AI chips commonly use 2.5D/3D advanced packaging, requiring 3-5 times the amount of ABF substrate compared to traditional chips. Although ABF substrate capacity expanded in 2024, AI demand is growing faster, making supply of high-end substrates for non-AI applications tight.

4. PCB Materials: The "Hidden Shortage" Spilling Over from AI

The most easily overlooked segment is PCB upstream materials. AI accelerator platforms typically require high-density HDI or high multi-layer boards with more than 40-50 layers, far exceeding the 16-20 layers of traditional servers. A single AI server consumes several times the copper-clad laminate area and glass fabric of a traditional server.

The three basic materials—glass fabric, resin, and copper foil—have limited production capacity themselves. After the surge in AI demand, some copper foil manufacturers have prioritized electrolytic copper foil for high-frequency/high-speed laminates, tightening supply for standard FR-4 copper foil. Multiple copper-clad laminate manufacturers have issued price increase notices for FR-4 materials, with lead times extending from 2 weeks to 4-6 weeks.

This is not a shortage of AI components themselves, but a spillover effect from AI demand crowding out upstream basic material capacity. For OEMs producing industrial equipment, automotive electronics, and consumer electronics, even if the ICs, connectors, and passive components they need are all in stock, they may be unable to complete assembly due to PCB material shortages.

Industry Chain Analysis: Complete Industry Chain Impact

Upstream: Materials and Equipment## Industry Chain Analysis: Impact Across the Full Industry Chain

Upstream: Materials and Equipment

• Silicon wafers/chemical materials: AI chips increase demand for high-purity silicon wafers and specialty gases, but overall capacity is sufficient. Impact is minimal.
• Copper foil, fiberglass cloth, resin: Driven by AI server PCB consumption, combined with demand from new energy and communications infrastructure, supply is tight. Price increases are being passed on to CCL and PCB manufacturers.
• Substrate materials (ABF): Demand for advanced packaging for AI chips is strong. ABF film suppliers such as Ajinomoto have limited expansion, and supply remains tight.

Midstream: Chip Manufacturing and Packaging/Testing

• Wafer foundry: TSMC and Samsung's advanced process capacity is occupied by AI chips, while mature processes are relatively loose, but some specialty process capacity used for PMICs and controllers is being squeezed.
• Memory manufacturers: Samsung, SK Hynix, and Micron are prioritizing capacity for HBM and DDR5, shrinking supply of niche DRAM.
• Packaging and testing: AI packaging (CoWoS, InFO, etc.) capacity is in short supply, while traditional packaging and testing capacity is relatively balanced. Manufacturers such as JCET and ASE are reallocating capacity between AI and non-AI products.

Downstream: OEM/ODM and Terminal Brands

• AI server manufacturers: NVIDIA, Supermicro, Dell, etc. benefit, but they have to absorb rising costs for HBM and power ICs.
• Traditional server/PC/phone OEMs: They face the risk of being squeezed out of key component supply, and price increases are either passed directly to end customers or squeeze gross margins.
• Automotive and industrial electronics: Tight supply of PMICs, MCUs, and PCB materials may affect production schedules.

Competitive Landscape: Suppliers Gain Stronger Bargaining Power

When supply tension persists, suppliers with capacity and inventory will have stronger bargaining power. Memory manufacturers, power IC giants (such as Texas Instruments and Infineon), and major CCL manufacturers (Shengyi Technology, Kingboard Laminates) all have the ability to improve profitability by adjusting their product mixes.

For small and medium-sized distributors and EMS companies, the ability to secure stable supply will become a core competitive advantage. Distributors with early positioning in AI-related components, such as Avnet and Arrow Electronics, may further expand their market share.

Conversely, OEMs that rely on a single supplier and lack alternatives will face out-of-control lead times and rising costs. Supply chain management complexity has escalated from capacity allocation during the "chip shortage" era to multi-material, multi-dimensional balancing.

Regional Impact: Rebalancing of Asian Manufacturing Hubs

Mainland China, as the world's largest PCB and electronics assembly base, will directly bear the impact of FR-4 material price increases, but local CCL companies' expansion plans may benefit. South Korea and Japan are the main suppliers of HBM and high-end materials (fiberglass cloth, ABF), respectively, and their suppliers' profits are expected to rise. New PCB capacity in Southeast Asia is under construction, but it will be difficult to fill the gap in the short term.The United States dominates AI chip design, but its supply chain is highly dependent on Asia. Export controls and tariff policies may further push up the cost of imported materials, prompting more American OEMs to bring procurement and design verification forward to cope with uncertainty.

Investment Perspective: New Focus of Capital Markets

This round of AI-driven supply chain strain has led investors to focus on the "pick-and-shovel sellers" of the pick-and-shovel sellers—that is, companies that supply upstream materials for AI hardware. Companies in areas such as copper-clad laminates, ABF substrates, HBM packaging and testing, and power management ICs are expected to see valuation uplift in 2026.

On the other hand, the inventory value of traditional electronic component distributors may be revalued. Companies with large inventories and priority supply agreements will benefit from rising prices. But beware: if demand falls short of expectations or capacity expands too quickly, the price decline could be just as sharp.

Long-Term Outlook: Structural Change Rather Than Short-Term Volatility

Over the next three years, AI demand will continue to dominate supply chain capital expenditure. Capacity expansion for HBM and advanced packaging takes time, and tight supply of standard DRAM and PMICs may persist until 2027. In the long run, as the AI server installed base grows, demand for basic materials will be sustained.

Coping strategies for the industry chain include:

• Diversified sourcing: OEMs need to qualify second and third suppliers, especially for PCB materials.
• Design adjustments: Use standard part numbers wherever feasible to reduce dependence on scarce materials.
• Early engagement: Communicate with suppliers at the R&D stage to avoid "procuring based on assumptions from a few months ago."

Over a 5-10 year horizon, the AI supply chain will form an ecosystem independent of consumer electronics. From specialized chips and high-bandwidth memory to efficient power architectures and specialty laminates, a stable, high-value supply chain cluster is expected to take shape. Traditional electronics manufacturing, meanwhile, will need to seek balance amid limited resources.

Conclusion

The impact of AI demand on the electronics supply chain has evolved from an "AI chip shortage" into a "reallocation of resources across the entire chain." From DRAM and PMICs to PCB base materials, a capacity shift in any link of the chain could leave non-AI product lines in a passive position.

For the industry, the most important judgment is this: this is no longer cyclical stock-outs, but a permanent structural change in the supply chain in the AI era. OEMs, distributors, and material suppliers that can anticipate and adapt to this change in advance will gain a competitive advantage in the future; those still seeking supply with an old map may fall behind in the next lead-time storm.