Goldman Sachs believes the explosion in demand for AI servers is pushing the entire electronic components and materials supply chain into a new cycle of supply-demand tightness, with the situation continuing to intensify. A report published on April 22 by Goldman Sachs' Daiki Takayama team indicates that the demand surge driven by AI servers since the beginning of this year is exacerbating an already tight supply chain into a more severe shortage. The report covers 11 critical segments, from semiconductors and passive components to specialty materials, with significant upward revisions to supply-demand forecasts for five key areas. From MLCCs, ABF substrates, PCBs/CCLs, and memory to indium phosphide (InP) substrates, supply-demand gaps have continued to widen over the past four months, a trend that is unlikely to reverse before 2027.
**MLCCs: Potential for Long-Awaited Price Hikes, Capacity Expansion Faces Constraints**
Multilayer Ceramic Capacitors (MLCCs) are one of the categories in the report where supply-demand changes are drawing significant market attention. Goldman Sachs points out that demand for MLCCs from AI servers is projected to grow approximately 4.3 times from 2025 to 2030. Combined with persistently strong demand from automotive applications, the entire MLCC industry is entering a phase of pronounced supply-demand tightening. Notably, even customers in weaker demand segments like smartphones and PCs have begun proactively seeking long-term supply contracts with MLCC manufacturers. This is not due to robust demand in their own sectors, but rather concerns about securing sufficient supply in the future once production capacity is absorbed by AI server orders. This "defensive ordering" behavior is itself further intensifying the supply-demand strain. Regarding capacity, Goldman Sachs believes the MLCC industry's annual capacity growth is capped at just over 10%, constrained by limitations in equipment and materials, making expansion extremely slow. If these limited capacity additions are primarily absorbed by AI servers and automotive applications, the current tight supply-demand dynamic could persist throughout the cycle. On pricing, reports of MLCC price increases emerged in mid-April 2026. While their authenticity awaits confirmation during earnings season, Goldman Sachs has revised its 2026 price change forecast for similar products from flat previously to an increase of 0% to 5%. If average price increases reach 5%, Goldman Sachs estimates this could boost the operating profit of relevant leading manufacturers by 13% and 37%, respectively, in fiscal year 2027.
**ABF Substrates: AI Chip Design Specifications Continue to Evolve, No Significant Supply Expansion Expected Before 2027**
Ajinomoto Build-up Film (ABF) substrates are a core material for AI chip packaging. Goldman Sachs maintains a positive outlook on the ABF substrate supply-demand picture for the next 18 to 24 months, driven by two factors: First, the expansion of AI chip shipments. Second, rapid upgrades in the substrate area and layer count per chip. The expected price increase forecast for 2026 has been raised from 20-30% to 30-35%. The mainstream substrate design specification for current AI chips is 75mm×85mm with 14/16 layers. Goldman Sachs predicts this will rise to 120mm×150mm with over 20 layers by 2027, and further to 200mm×250mm with over 20 layers by 2030. Larger sizes not only directly reduce production yields but also increase panel utilization losses, effectively decreasing the actual useful supply from a given unit of capacity. Capacity at major Taiwanese ABF substrate manufacturers is largely booked through the end of this year, and Goldman Sachs expects no significant capacity expansion from ABF substrate suppliers before the second half of 2027. Extended lead times imply sustained pricing power.
**PCBs/CCLs: Technical Specifications Leap Forward, AI Servers Drive Average Selling Prices Up Over 30% Annually**
Regarding AI server demand for Printed Circuit Boards (PCBs) and Copper-Clad Laminates (CCLs), Goldman Sachs observes two parallel upgrade paths: First, the grade of CCLs is rising: from the mainstream M7 grade in 2023 to M8 in 2025, and projected to reach M9 by 2027. Second, the number of PCB layers is increasing: from the mainstream 22 layers in 2023 to 28 layers in 2025, and projected to exceed 36 layers by 2027. Goldman Sachs attributes these trends primarily to demands for higher data transmission speeds and improved computing performance, forecasting that the average selling price for AI CCLs/PCBs will maintain year-on-year growth exceeding 30% annually for the foreseeable future. Regarding supply-demand balance, Goldman Sachs estimates that leading PCB/CCL suppliers are expanding capacity by about 20% to 30% annually, a pace notably slower than the growth rate of AI server demand, which TSMC previously forecast could see a compound annual growth rate exceeding 50%. Excess demand will spill over to second and third-tier suppliers. However, based on 2025 data, the production yields of these lower-tier suppliers are generally about 20% lower than those of leading manufacturers, making it difficult for them to truly capture market share even by initiating price wars. Consequently, leading PCB/CCL suppliers are expected to continue enjoying advantages in both supply-demand dynamics and pricing.
**Memory: Supply-Demand Gap Widens Sharply in 2026, Goldman Sachs Significantly Raises DRAM/NAND Price Forecasts**
Memory is the category with the most substantial forecast revision in the report and one of the segments with the most profound impact on the entire supply chain. Goldman Sachs has significantly raised its 2026 DRAM price increase forecast from approximately 150% previously to a range of +250% to +280%. The NAND price increase forecast has been raised from about 100% to a range of +200% to +250%. Three key drivers of the tightness are: Strong demand from server applications. Limited new capacity additions industry-wide, with new capacity prioritized for High Bandwidth Memory (HBM), compressing the supply growth for conventional memory. Exceptionally low inventory levels across the industry, further limiting supply elasticity. More notably, Goldman Sachs has extended its expectation of supply tightness into 2027. Previously, the firm anticipated the memory market would return to balance in 2027, but the latest assessment suggests tight supply conditions are unlikely to ease that year, with the overall market expected to remain tight.
**Indium Phosphide (InP) Substrates: Core Material for AI Optical Interconnects, Tight Supply to Persist, Capacity Expansion a Multi-Year Process**
Indium Phosphide (InP) substrates are an emerging scarce material highlighted in the report. Goldman Sachs expects the tight supply-demand situation for InP substrates to continue in 2026, forecasting a price increase of approximately 15%. Regarding capacity expansion plans, major suppliers have announced multiple expansion phases in July and October 2025, and February 2026, with continued investment. Despite clear expansion intentions, Goldman Sachs notes that the substantive commissioning of new capacity is not expected to begin until around fiscal year 2028 at the earliest, meaning the capacity bottleneck is unlikely to be fundamentally resolved in the near to medium term. Concurrently, industry feedback indicates robust market demand for InP substrates, with supply from multiple companies already insufficient to meet downstream demand. The supply shortage is expected to persist into the next fiscal year.
**Foundry: Tight Supply for TSMC's Advanced Processes to Continue Until 2027 and Beyond**
In the foundry sector, Goldman Sachs' assessment for TSMC is that, driven by growing demand for AI inference, supply for advanced process nodes will remain tight at least until 2027. The additional capacity resulting from TSMC's current capital expenditure cycle is not expected to come online until 2028-2029. Goldman Sachs forecasts that TSMC's US dollar revenue will grow year-on-year by 35%, 30%, and 29% in 2026, 2027, and 2028, respectively. The outlook differs for mainland Chinese foundries. Although current capacity utilization is high, Chinese foundries are continuously expanding capacity, and the supply-demand situation is not tight. Goldman Sachs expects prices for similar products from mainland Chinese foundries to rise about 10% in 2026, driven by high utilization rates, recovering demand from industrial equipment end-markets, growth in AI-related product demand, rising raw material costs, and a relatively healthy competitive landscape.
Synthesizing the findings of the Goldman Sachs report, for investors, the significantly raised memory price forecasts, the persistent tightness of ABF/BT substrates, and the multi-year expansion cycles for InP substrates all point in the same direction: physical constraints on the supply side will underpin the profitability and pricing power of the entire computing hardware supply chain for a considerable period.
Comments