AI server demand intensifies tightness in the Asia technology supply chain

Overview

This report is Goldman Sachs' update on supply-demand balance and price outlooks for 11 categories of electronic components and materials in the Asia technology supply chain. It continues the January theme that AI server-driven demand is tightening supply and demand for niche components and materials, with prices stabilizing or rising, and notes that this trend has strengthened further over the past four months. The research scope covers MLCC, ABF substrates, BT substrates, PCB/CCL, T-Glass cloth, electrolytic copper foil, tantalum powder, InP substrates, memory, nearline HDD, and wafer foundry.

Core views

The core view is that AI server demand is spreading supply chain tightness from isolated segments to multiple key materials and component segments. MLCC capacity growth is only slightly above 10% annually, making it difficult to fully absorb AI server and automotive application demand; ABF substrates are expected to maintain favorable supply-demand dynamics for 18 to 24 months due to larger size, higher layer-count design upgrades, and yield loss; BT substrates are supported by memory demand and T-Glass shortages; PCB/CCL is driven by high-speed transmission and layer-count upgrades, with first-tier suppliers benefiting more clearly; the memory supply gap is more severe than previously expected; and demand for TSMC advanced process nodes may remain supply-constrained even beyond 2027.

Analysis framework

The report uses a combination of supply-demand heat maps, price change forecasts, capacity expansion pacing, application-demand breakdowns, and beneficiary company mapping to update each of the 11 categories of devices and materials. The analysis focuses on AI server shipments and rising content per unit, automotive and industrial demand, material bottlenecks, equipment and engineering resource constraints, fully booked capacity, yield loss, layer-count and specification upgrades, inventory levels, and pricing pass-through ability.

Methodology notes

• Supply-demand analysisSupply-demand heat map

  Compare the degree of tightness across different components and materials by quarter

  The report marks categories including DRAM, NAND, nearline HDD, wafer foundry, MLCC, T-Glass, tantalum powder, copper foil, InP substrates, ABF substrates, and CCL by quarter as very tight, tight, not tight, balanced, easing, or oversupplied, and compares the changes versus the January version.

• Price forecastingPrice change rate assumptions

  Use price change rates to reflect the transmission of supply-demand tightness into earnings

  The report raises its 2026 price assumptions for multiple segments, for example DRAM to +250% to +280%, NAND to +200% to +250%, MLCC to +0% to +5%, ABF substrates to +30% to +35%, and CCL to +35% to +40%.

• Company mappingBeneficiary mapping

  Map component and material segments to related Buy-rated stocks

  The report lists related companies across chains such as MLCC, ABF substrates, BT substrates, PCB/CCL, materials, memory, and wafer foundry to identify assets more directly affected by AI server demand and rising prices.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• MLCC: Murata Mfg., Taiyo Yuden, Samsung Electro-Mechanics

  AI server and automotive applications are boosting demand for high-load products, and customers are beginning to seek long-term contracts.

  Strengths

  Capacity growth is constrained, pricing has shifted from prior expectations of declines to flat to modest increases, and price hikes have high earnings elasticity.

  Weaknesses

  Some news on price hikes still needs confirmation from company earnings calls, and price increases may cover only part of the product portfolio.

  Comparison

  Both Murata and Taiyo Yuden are considered beneficiaries, with Taiyo Yuden showing higher operating profit elasticity under the theoretical price-hike scenario.

  Risks

  If AI server demand is below expectations or new capacity comes onstream faster than expected, supply-demand tightness and the magnitude of price increases may fall short of expectations.

• ABF substrates: Ibiden, NYPCB

  Growth in AI server IC shipments, larger sizes, and higher layer-count designs are driving ABF substrate demand.

  Strengths

  The supply-demand outlook is favorable for the next 18 to 24 months, Taiwan manufacturers' capacity is close to fully booked, and high-end switch chip and AI ASIC substrates offer growth opportunities.

  Weaknesses

  Complex designs reduce yields and increase panel loss, while the capacity expansion cycle is long.

  Comparison

  NYPCB is viewed as a key supplier of high-end switch chip substrates, while Ibiden is seen as a leader in the technological evolution of high-end GPU and ASIC substrates.

  Risks

  If customer designs change, yields improve, or competitors expand capacity more than expected, price and share assumptions could come under pressure.

• BT substrates: Kinsus, NYPCB, Unimicron

  Memory-related demand accounts for about 50% to 60% of BT substrate demand in 2026 and is supported by T-Glass shortages.

  Strengths

  Prices already rose in 4Q25 and are expected to continue rising QoQ in 1Q26 and 2Q26.

  Weaknesses

  Part of revenue growth depends on passing through material costs; if end demand weakens, sustainability could be affected.

  Comparison

  Kinsus has the highest BT substrate exposure among Taiwan manufacturers, followed by NYPCB and Unimicron.

  Risks

  An easing in T-Glass shortages or cooling memory demand could weaken pricing momentum.

• PCB/CCL: EMC, TUC, GCE

  High-speed transmission demand from AI servers is driving upgrades in CCL grades and PCB layer counts, and increasing PCB value per server.

  Strengths

  Capacity expansion by first-tier suppliers is constrained by equipment and high-end materials, so supply growth is slower than AI server demand growth, leaving room for sustained ASP increases.

  Weaknesses

  Second- and third-tier suppliers may take overflow demand, but their yields are lower and competitiveness is limited.

  Comparison

  EMC benefits from adoption of M9-grade CCL, TUC is seen as gaining CCL share in ASIC AI servers, and GCE is seen as a key PCB supplier for ASIC AI servers.

  Risks

  If changes in AI server architecture reduce PCB/CCL usage, or if second- and third-tier suppliers rapidly improve yields, industry pricing and market share dynamics may change.

• Materials: T-Glass cloth, electrolytic copper foil, tantalum powder, InP substrates

  High-end packaging, PCB/CCL, optical communications, and AI server-related demand are tightening key materials.

  Strengths

  Nittobo is expanding capacity but a 20% to 30% price revision is still assumed; Mitsui Kinzoku is expanding VSP copper foil capacity and there is a 5% to 10% probability of further price increases; JX tantalum powder and InP substrate demand are strong.

  Weaknesses

  Many capacity expansions will only come onstream gradually by FY3/28 and beyond, leaving limited short-term supply elasticity.

  Comparison

  JX Advanced Metals is exposed to both tantalum powder and InP substrates, while Sumitomo Electric mainly benefits from InP substrate capacity expansion.

  Risks

  Capacity expansion progress, customer qualification, material substitution, and price acceptance are the main uncertainties.

• Memory: Samsung Electronics, SK Hynix

  Strong server demand, limited new industry capacity, and priority allocation to HBM are constraining supply growth for conventional memory.

  Strengths

  2026 DRAM and NAND price expectations have been revised up significantly, and supply tightness may continue into 2027.

  Weaknesses

  The magnitude of price increases is very high, implying strong dependence on supply discipline and demand sustainability.

  Comparison

  Priority allocation of new capacity to HBM tightens conventional DRAM/NAND further, benefiting the major memory manufacturers.

  Risks

  New capacity additions, inventory recovery, slower server demand, or changes in customer procurement pace could lead to price declines.

• Wafer foundry: TSMC, SMIC, Hua Hong

  Both advanced nodes and China's domestic mature nodes are affected by AI-related demand and high utilization, but their supply-demand conditions differ.

  Strengths

  TSMC advanced nodes are expected to remain supply-constrained through 2027 and beyond; China's domestic manufacturers are supported by high utilization, recovering industrial demand, and rising raw material costs, with prices for comparable products expected to rise 10% in 2026.

  Weaknesses

  China's domestic wafer foundries are continuing to expand capacity, so the report does not view their supply and demand as tight.

  Comparison

  TSMC shows more pronounced supply-demand tightness, while domestic Chinese foundries reflect modest price increases under high utilization rather than structural shortages.

  Risks

  The release of new advanced-node capacity, fluctuations in AI model demand, export restrictions, and excessively rapid expansion of domestic mature-node capacity could all affect the outlook.

Key data

• AI server demand growthabout 4.3xThe report estimates that the AI server demand opportunity will grow about 4.3x from FY25 to FY30.
• Murata FY3/27 MLCC sales forecast¥1.055 tn, up 13% YoYIf an average 5% price increase is achieved, there would theoretically be about +13% upside to the FY3/27 operating profit forecast.
• Taiyo Yuden FY3/27 MLCC sales forecast¥286.0 bn, up 13% YoYIf an average 5% price increase is achieved, there would theoretically be about +37% upside to the FY3/27 operating profit forecast.
• MLCC industry capacity growth constraintannual growth slightly above 10%Equipment and materials are primarily internally produced and constrained by internal engineering resources.
• ABF substrate high-end design upgrade75mm*85mm*14/16 layers to over 120mm*150mm*20 layers by 2027The report expects this to rise further to 200mm*250mm*20+ layers by 2030.
• BT substrate pricesup 15% in 4Q25, with another 15%/20% QoQ increase expected in 1Q26/2Q26Mainly supported by memory-related demand and T-Glass shortages.
• T-Glass shortage ratiocould reach 30% to 40%+ in 2026Higher material costs may further push up BT substrate prices.
• PCB/CCL specification upgradeCCL upgraded from M7 to M8 and expected to upgrade to M9 in 2027; PCB increased from 22 layers to 28 layers and is expected to reach 36+ layers in 2027Driven by data transmission speed requirements resulting from higher AI server computing power.
• AI CCL/PCB ASP30%+ YoY increase annually for the foreseeable futureThe report believes specification upgrades and supply constraints support price increases.
• JX Advanced Metals tantalum powderFY3/26 total sales volume +161% YoY, AI server-related sales volume +580% YoYThe report assumes about a 10% price revision in 2026.
• DRAM/NAND 2026 price expectationsDRAM +250% to +280%; NAND +200% to +250%A significant upward revision versus previous forecasts of about +150% and about +100%.
• TSMC revenue forecast2026E to 2028E USD revenue growth of 35%/30%/29% YoYBased on the view that demand for advanced process nodes will remain supply-constrained.

Impact & implications

The investment implication is that bottlenecks in the AI server supply chain may continue to support prices, order visibility, and earnings elasticity for key components and materials. More direct beneficiaries include high-end MLCC, ABF substrates, BT substrates, AI server PCB/CCL, high-end materials, DRAM/NAND, advanced wafer foundry, and nearline HDD. At the same time, divergence is widening: TSMC advanced process nodes are seen as remaining tight, while China's domestic wafer foundries, despite high utilization, are not viewed as being in a tight supply-demand situation because of ongoing capacity expansion.

Risks

• AI server shipments or capital expenditures come in below expectations, causing component and material demand to fall short of the report's assumptions.
• Supplier capacity expansion, yield improvements, or the introduction of substitute materials proceed faster than expected, easing supply-demand tightness.
• Aggressive price assumptions could suppress customer purchasing or trigger redesigns, substitution, and inventory adjustments.
• Export controls, sanctions, the BIS Entity List, and cross-border regulatory restrictions may affect some China-related companies and supply chain transactions.
• The report includes disclosure of potential business relationships between Goldman Sachs and covered companies, and investors should be mindful of conflict-of-interest risk.

What to watch

• Confirmation from Murata Mfg. and Taiyo Yuden earnings calls on the validity and scope of MLCC price increases.
• The extent to which Taiwan ABF substrate manufacturers remain fully booked and the pace of capacity expansion implementation after 2H27.
• Start-up timing, qualification progress, and actual price revisions for expansion projects in T-Glass, VSP copper foil, InP substrates, and tantalum powder.
• Whether DRAM and NAND prices deliver sharp increases in 2026, and whether supply tightness continues into 2027.
• TSMC N3/N5 utilization, timing of new capacity releases, and growth in AI token consumption.
• The impact of AI server architecture upgrades on PCB/CCL layer count, grade, midplane PCB adoption, and value per server.