DESK-TREND-20260318-INFCHIP AI INFRASTRUCTURE MID-TERM
OW +23.75
OverweightSignal Weighted Score

The Hidden Component Revolution of Inference AI Chips

Beyond Packaging & Cache, 10 Key Components and Investment Opportunities Supporting AI Inference Infrastructure

AUTHOR Quantamental Hybrid Desk
DATE 2026-03-18
HORIZON 1-6M
STYLE Stacker
HYPERSCALER CAPEX
$650-700B
2026E
AI SERVER MLCC
30,000/unit
Based on GB300
LIQUID COOLING
76%
2026E Penetration Rate
SIGNAL SCORE
+23.75
→ OW

Executive Summary

Macro Environment

The Fed's benchmark interest rate has entered a stable trajectory at 3.50-3.75%. With the February CPI at 2.4%, the trend of slowing inflation continues, and expectations for an additional rate cut in June are forming. Interest rate stability is a key macroeconomic backdrop supporting the valuation multiple of growth stocks. FACT

Hyperscaler CAPEX is breaking historical highs at $650-700B. Approximately 75% of the total CAPEX is being invested in AI infrastructure, which means a concentration of demand for GPUs and server components in general. The combined 2026 CAPEX guidance of MSFT, GOOG, META, and AMZN alone amounts to $380B+. FACT

Brent oil price at $102/bbl due to the Iranian war is driving up energy costs. Paradoxically, the rise in data center power costs is acting as a catalyst to accelerate demand for power-efficient components (GaN/SiC power ICs, high-efficiency HVDC) and cooling solutions. NARRATIVE

The weak yen at ¥159.7 is simultaneously increasing the real export competitiveness and won-converted sales of Japanese component companies (Murata, Nittobo, Ajinomoto Fine-Techno). The exchange rate leverage is maximized in high value-added materials such as MLCCs, ABF films, and T-Glass, which are dominated by Japan. FACT

◆ Overall Macro Judgment
BULLISH. Interest rate stability + record CAPEX + accelerated demand for power efficiency are a triple tailwind for AI component investment themes. The Iran-driven oil price risk is paradoxically a catalyst for the power efficiency/cooling sector.

Analysis of 10 Key Components

1. ABF Substrate / T-Glass

28/30 BULLISH

Ajinomoto's ABF (Ajinomoto Build-up Film) dominates more than 95% of the global advanced semiconductor substrate market. ABF is essential for the organic substrates of all AI accelerators (GPUs, ASICs, FPGAs), and there is virtually no alternative. Since AI server substrate area requires 10 times more than smartphones, demand surges with the AI shift alone. FACT

Nittobo's T-Glass (ultra-low dielectric constant glass fiber) also maintains a monopolistic position. The supply gap for T-Glass, which is essential for high-speed signal transmission substrates of 56Gbps or higher, is currently about 20%, and the lead time is the longest in the industry at 35 weeks or more. FACT The industry consensus is that normalization is difficult until 2027, linked to the expansion speed of TSMC CoWoS, despite Ajinomoto and Nittobo's combined ¥150B expansion investment. INFERENCE

ABF/T-Glass is the most serious 'hidden bottleneck' in the AI chip supply chain. Although it does not receive media attention like CoWoS or HBM, the actual supply constraints are greater for these materials. Beneficiary companies: Ajinomoto, Nittobo, Ibiden, Shinko Electric. NARRATIVE

2. MLCC (Multi-Layer Ceramic Capacitor)

25/30 BULLISH

The number of MLCCs required per AI server is approximately 30,000 based on GB300. This is a 4-6x increase compared to existing general servers (5,000-8,000 units). Demand is concentrated on high-capacity (100μF+) and high-withstand voltage (100V+) MLCCs, and Murata leads the market with a global market share of 40%. FACT

Spot prices for high-performance MLCCs for AI servers have risen +20% year-on-year, and contract prices are also increasing by 5-8% quarterly. The compound annual growth rate (CAGR) of