AI demand drives a deeper ABF substrate shortage, and Morgan Stanley raises earnings and target prices

Overview

This report focuses on the ABF substrate industry chain within Greater China technology hardware. Morgan Stanley believes demand for AI-related GPUs, ASICs, CPUs, and networking chips continues to be revised upward, while new effective capacity takes a long time to come online, intensifying the ABF substrate supply-demand imbalance. The report raises its 2030 ABF substrate supply-demand gap forecast from about 15% to 22%, and based on that increases assumptions for prices, margins, EPS, and target prices for major suppliers.

Core views

The core view is that ABF substrates are still in the early stage of an upcycle, and near-term supply expansion cannot fully match AI demand growth. Unimicron benefits from its supply share in AI ASICs and server CPUs; Nan Ya PCB benefits from networking-chip exposure and an active pricing strategy; and Zhen Ding has rerating potential thanks to Chinese AI chips, BT substrates, AI server PCBs, and potential Google TPU substrate certification.

Analysis framework

The report uses a bottom-up ABF substrate supply-demand model to reassess demand for GPUs, ASICs, CPUs, and networking chips, while incorporating the capacity expansion cycle, customer certification, pricing assumptions, and supplier earnings leverage. On valuation, Unimicron and Nan Ya PCB are assessed using a residual income model, while Zhen Ding's target price is derived with reference to CY27e/CY28e P/E and mid-term growth rates.

Methodology notes

• Supply-demand analysisABF substrate supply-demand model

  ABF substrate supply-demand gap

  By comparing total ABF substrate supply with substrate demand from end markets such as GPUs, ASICs, CPUs, and networking, the report assesses future shortage severity and upward price pressure.

• Valuation methodResidual income model

  Residual income valuation

  The report uses a residual income model to value Unimicron and Nan Ya PCB, with key assumptions including cost of equity, risk-free rate, equity risk premium, beta, mid-term growth rate, and terminal growth rate.

• Scenario analysisBase case / bear case / bull case

  Risk-return scenarios

  The report provides upside and downside scenarios for individual stocks using different demand, valuation multiple, and earnings assumptions to assess the risk-return profile of target prices and ratings.

Asset mapping & comparison

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

• Unimicron (3037.TW)

  Core beneficiary name, maintain Overweight and raise target price to NT$1,225.

  Strengths

  Strong share in AI ASIC and server CPU supply, better ability to relieve capacity bottlenecks, and earnings returning to growth from 2026.

  Weaknesses

  The share price has already significantly outperformed the TAIEX, and the valuation is sensitive to earnings delivery and supply-demand gap assumptions.

  Comparison

  Compared with Zhen Ding, Unimicron has a longer track record in ABF and commands a higher valuation multiple.

  Risks

  Weaker-than-expected PC, general server, or AI server demand; capacity expansion faster than expected; tighter T-glass supply; ABF substitution by CoWoP.

• Nan Ya PCB (8046.TW)

  Core beneficiary name, maintain Overweight and raise target price to NT$1,275.

  Strengths

  Strong share in networking-chip-related ABF substrates, and its active pricing strategy allows it to benefit more from shortages.

  Weaknesses

  Sensitive to high-end demand and price assumptions; if supply-demand tightness eases, margin leverage may decline.

  Comparison

  Like Unimicron, it benefits from the ABF shortage, but the report assumes higher price growth for NYPCB in CY28.

  Risks

  AI and networking demand falls short; customer bargaining pressures prices; added capacity eases the shortage.

• Zhen Ding (4958.TW)

  Rating upgraded to Overweight; target price NT$570.

  Strengths

  Rising share in Chinese AI chip ABF substrates, BT substrate share gains, AI server PCB and optical module PCB all provide growth drivers; if Google TPU substrate certification advances, it could bring a valuation rerating.

  Weaknesses

  Its ABF business track record is shorter than Unimicron's and NYPCB's, so its valuation multiple still trades at a discount.

  Comparison

  Its implied CY27e/CY28e P/E multiples are below those of Unimicron and Nan Ya PCB, but if expansion into high-end ASIC/GPU customers succeeds, there is potential for the multiple to converge toward peers.

  Risks

  Failure to win high-end customer certification; Chinese AI chip demand below expectations; Shenzhen and Kaohsiung capacity ramping below expectations.

Key data

• 2030 ABF supply-demand gap22% undersupplyThe prior assumption was about 15%; this widens on the back of higher demand assumptions.
• CY26 ABF substrate price growth15-20% y/yDriven mainly by cost inflation and supply-demand imbalance.
• CY27 ABF substrate price growth20%+ y/yThe report expects the worsening shortage to continue pushing prices higher.
• CY28 ABF substrate price growth25-40% y/yPrice elasticity differs by customer and supplier; NYPCB assumes a higher level.
• 2030 CPU TAM base caseUS$125bnIncludes a US$79bn orchestration CPU TAM and a US$45bn host/cloud CPU TAM.
• 2030 CPU TAM bull caseUS$283bnBased on assumptions for AI data center installed GW and a higher CPU:GPU ratio.
• 2030 shipment assumption for other ASIC chips~9.2M unitsUp from about 3.9M units previously, including Microsoft, other U.S. companies, and Chinese AI chips.
• Unimicron target priceNT$1,225Approximately 49% upside; CY26-28 EPS raised by 19%, 21%, and 41%, respectively.
• Nan Ya PCB target priceNT$1,275Approximately 62% upside; CY26-28 EPS raised by 37%, 59%, and 81%, respectively.
• Zhen Ding target priceNT$570Approximately 42% upside; rating upgraded to Overweight.

Impact & implications

If the report's view is correct, ABF substrate suppliers will benefit from price increases, margin recovery, and earnings upgrades during 2026-2030, especially manufacturers with exposure to high-end AI ASICs, server CPUs, networking chips, or Chinese AI chips. The implication for the supply chain is that customers will more actively expand their supply ecosystems, creating certification opportunities for smaller suppliers and new entrants, while also increasing the importance of tracking capacity expansion, material bottlenecks, and alternative packaging routes.

Risks

• Weaker-than-expected demand for PCs, general servers, and AI servers.
• Meaningful capacity expansion by major suppliers, leading to a easing of the ABF shortage.
• T-glass supply becomes tighter than expected, affecting available output and margins.
• Technology routes such as CoWoP replace ABF substrates.
• The stocks involved have already rallied significantly, raising valuation downside risk if earnings upgrades fail to materialize.

What to watch

• Whether the ABF substrate supply-demand gap continues to widen toward 22% or higher before 2030.
• Whether demand revisions for Amazon Trainium, Google TPU, Chinese AI chips, and other ASICs are realized.
• Price changes, gross margins, and CY26-28 EPS revisions for Unimicron, Nan Ya PCB, and Zhen Ding.
• Whether Zhen Ding secures certification from Google TPU or other high-end ASIC/GPU customers.
• The pace of new capacity plans, because large-scale new capacity usually takes at least two years to affect supply and demand.
• Whether end-demand for PCs, servers, AI servers, and networking chips slows.