AI compute expansion continues to strengthen the APAC semiconductor and advanced packaging chain

Overview

This material is a same-day summary of JPMorgan APAC Technology highlights, covering the news flow around META, MU, AMAT, TSMC, and the broader APAC technology supply chain. The central theme is that expanding AI data-center investment is driving higher demand across memory, advanced packaging, semiconductor equipment, electronic materials, and domestic AI chip supply chains. The material also notes that Nan Ya Plastics' second-quarter net profit and electronic materials margin exceeded expectations, Taiwan's drone exports grew rapidly, and lists various industry news items, research links, company events, and earnings calendars.

Core views

The report's most important conclusion is that AI compute capital expenditures are transmitting across multiple hardware segments: Meta's long-term supply agreements benefit Samsung, Sandisk/Kioxia, and the NAND chain; Micron's increased US capital expenditures and long-term supply agreement benefit front-end equipment and silicon wafer suppliers such as TEL and SCREEN; AMAT has longer-cycle visibility into SPE demand, benefiting equipment peers including TEL, SCREEN, Lasertec, Advantest, Disco, and Ebara; and TSMC and OSATs are accelerating CoWoS and similar capacity expansion, benefiting Chroma, ASE, GPTC, and the advanced packaging chain. For Chinese AI chips, supply-demand gaps are driving domestic substitution, while the training market is viewed as the next-stage opportunity after inference.

Analysis framework

The material uses an event-driven and industry-chain read-through approach, mapping news, management commentary, supply agreements, capital expenditure plans, capacity estimates, and second-quarter earnings updates to companies in upstream materials, equipment, foundry, advanced packaging, memory, and AI chips. The focus is not on a single-company financial model, but on identifying the marginal impact of AI compute expansion on APAC technology stocks.

Methodology notes

• Industry-chain mappingAI compute capital expenditure transmission

  Derive demand for memory, equipment, advanced packaging, and materials from cloud providers' compute deployments, long-term supply agreements, and capital expenditure plans.

  Information related to Meta, Micron, AMAT, and TSMC is used to cross-validate that AI data-center demand is driving multiple semiconductor supply-chain segments.

• Supply-demand analysisSupply-demand gaps and domestic substitution

  When overseas AI chip supply is constrained or deliveries are insufficient, domestic GPU and AI chip vendors gain substitution opportunities.

  Research on Shanghai GPU vendors indicates that domestic AI chips are progressing toward performance levels close to NVDA Hopper, and that the training or post-training market could become the next source of incremental growth.

• Capacity trackingCoWoS and OSAT advanced packaging capacity model

  Assess opportunities for advanced packaging equipment and outsourced semiconductor assembly and test providers through TSMC and OSAT CoWoS/CoWoS-like capacity forecasts.

  The report raises its FY26-28 CoWoS capacity estimates and points to AP7 Phase 2 expansion, outsourced interposers, and OSAT brownfield capacity releases.

Asset mapping & comparison

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

• META PLATFORMS INC

  Source of AI data-center compute demand

  Strengths

  Reportedly plans to deploy 7GW of compute in 2026 and approximately 14GW in FY27, and has signed long-term supply agreements, improving visibility for memory and the supply chain.

  Weaknesses

  The material does not provide Meta-specific financial modeling, valuation, or earnings-sensitivity analysis.

  Comparison

  Compared with a single hardware supplier, Meta is more of a demand-side driver, with its capital expenditures flowing through to Samsung, Sandisk/Kioxia, TSMC, and the advanced packaging chain.

  Risks

  Changes in the pace of AI capital expenditures, underperformance in executing supply agreements, and changes in the production ramp of internal AI chips.

• Micron Technology

  Driver of memory demand and semiconductor equipment capital expenditures

  Strengths

  Its US domestic capital expenditure plan has increased to $250bn and it has signed a 10-year supply agreement, benefiting front-end equipment and silicon wafer suppliers.

  Weaknesses

  High capital expenditures could create a risk of cyclical supply expansion, and the material does not provide return-on-investment estimates.

  Comparison

  Unlike equipment suppliers, Micron is both a capital expenditure driver and a beneficiary of the memory cycle.

  Risks

  Falling memory prices, policy or construction delays, and capital expenditure returns below expectations.

• Applied Materials

  Indicator of semiconductor production equipment demand visibility

  Strengths

  Management sees demand over the next eight quarters and has received directional outlooks from some customers through FY30; advanced packaging equipment is expected to continue growing rapidly.

  Weaknesses

  The material is based on media reports and management commentary and lacks quantitative breakdowns of orders, revenue, and profit.

  Comparison

  This provides a positive read-through for SPE peers including TEL, SCREEN, Lasertec, Advantest, Disco, and Ebara.

  Risks

  Changes in customer expansion plans, equipment delivery bottlenecks, and weaker-than-expected advanced process and packaging demand.

• TSMC

  Core CoWoS and advanced packaging capacity supplier

  Strengths

  TSMC CoWoS capacity is expected to reach 115k/190k/225k at the end of FY26/FY27/FY28, with AP7 Phase 2 expansion accelerating; OW and a PT of NT$3,100 are maintained.

  Weaknesses

  Capacity expansion requires capital expenditure and execution capabilities, while resource allocation between SoIC and CoWoS could affect the pace.

  Comparison

  Compared with OSATs, TSMC occupies the core advanced packaging bottleneck position; OSAT expansion primarily supplements CoWoS-like demand.

  Risks

  Fluctuations in AI accelerator demand, changes in customer product roadmaps, and advanced packaging yields and capacity ramps below expectations.

• SANDISK CORP

  Beneficiary of Meta's long-term supply agreement and NAND demand

  Strengths

  Reportedly signed a long-term supply agreement with Meta, benefiting its partner Kioxia and potentially supporting higher NAND demand.

  Weaknesses

  The material does not provide Sandisk-specific order size, pricing, or earnings contribution.

  Comparison

  Like Samsung, it benefits from the improved demand visibility provided by long-term agreements, although Kioxia's capital expenditures and capacity response are also key variables.

  Risks

  A reversal in NAND supply and demand, pricing pressure, and insufficient realization of customer demand.

• Nan Ya Plastics Corp

  Beneficiary of the AI server and electronic materials chain

  Strengths

  2Q net profit was NT$26.8bn, up 88% QoQ; electronic materials revenue grew more than 20% QoQ, and OPM could exceed 20%.

  Weaknesses

  Future margins depend on price increases, utilization, the T-glass ramp, and progress toward M10 certification.

  Comparison

  Compared with pure equipment or foundry companies, Nan Ya Plastics offers greater exposure to the earnings elasticity from electronic materials price increases and improved utilization.

  Risks

  Declines in CL, CCL, glass fiber, copper foil, or epoxy resin prices, or delays in certification.

• China AI chip fabless

  Domestic substitution and training-market opportunity

  Strengths

  Supply shortages are driving adoption of domestic AI chips; some local chips are approaching NVDA Hopper performance, and the training or post-training market is viewed as the next-stage opportunity.

  Weaknesses

  Commercialization and ecosystem maturity still require validation, while the training market has high competitive and software-adaptation barriers.

  Comparison

  Compared with the inference market, the training market is larger but has higher technical and ecosystem requirements.

  Risks

  Performance, yields, software ecosystem, customer validation cycles, and improved supply from overseas competitors.

• Taiwan drone supply chain

  Drone export growth theme

  Strengths

  Taiwanese drone exports reached $148m in the first four months, with the Czech Republic, Poland, and the United States as the top three markets; early success in Europe helps demonstrate the concept to Western buyers.

  Weaknesses

  The theme involves multiple companies, and the material does not provide individual revenue exposure or order visibility.

  Comparison

  Compared with the semiconductor AI chain, the drone theme is more driven by policy, exports, and defense applications.

  Risks

  Export controls, order sustainability, geopolitical changes, and customer concentration.

Key data

• Meta compute deployment plan7GW in 2026; approximately 14GW in FY27Bloomberg reported that Meta signed long-term supply agreements and plans to expand AI compute deployment.
• Micron US domestic capital expenditure planIncreased from $200bn to $250bn through FY35The report believes this provides a positive read-through for front-end semiconductor equipment suppliers and the raw silicon wafer supply chain.
• AMAT customer demand visibilityClear visibility over the next eight quarters; some customers provided directional outlooks through FY30Used to support the view of long-term demand for semiconductor production and advanced packaging equipment.
• TSMC CoWoS capacity estimateYear-end FY26/FY27/FY28: 115k/190k/225kJPMorgan further raised its FY26-28 CoWoS capacity estimates by 1%/9%/11%.
• OSAT CoWoS-like capacity estimateYear-end FY26/FY27/FY28: 15k/50k/85k wfpmThe report believes this benefits ASE and GPTC.
• CoWoS supply-demand gapApproximately 20%The report believes the supply-demand gap is widening, with demand from server CPUs, networking, LPUs, Trainium, ASICs, and other applications supporting capacity expansion.
• Nan Ya Plastics 2Q net profitNT$26.8bn, +88% QoQAbove JPMe of NT$20.5bn and the market consensus of NT$16.6bn; electronic materials OPM could exceed 20%.
• Electronic materials price increasesCL prices increased 15% from mid-MarchThe report considers this one of the important reasons for the expansion of Nan Ya Plastics' electronic materials margin.
• Taiwan drone exports$148m in the first four monthsExports accelerated significantly, with the Czech Republic, Poland, and the United States as the top three destinations.

Impact & implications

For portfolios, the report reinforces the logic of the AI hardware capital expenditure diffusion trade: beneficiaries extend beyond GPUs and cloud providers to include NAND, silicon wafers, front-end equipment, advanced packaging equipment, OSATs, electronic materials, and parts of Taiwan's drone supply chain. TSMC and OSAT CoWoS expansion provides high visibility for the FY26-FY28 advanced packaging chain, while Micron- and AMAT-related developments support a medium- to long-term constructive view on equipment and materials. It should be noted that the material states it comes from the Sales and Trading department and is not an independent research product of the J.P. Morgan Research Department.

Risks

• AI data-center capital expenditure growth falls short of expectations, leading to downward revisions to memory, advanced packaging, and equipment demand.
• CoWoS, SoIC, or OSAT CoWoS-like capacity ramps, yields, or customer qualifications fall short of expectations.
• Prices of cyclical products such as NAND, silicon wafers, and electronic materials decline, weakening margin improvement.
• Commercialization of the domestic AI chip training market progresses more slowly than expected, with ecosystem, software, and customer validation becoming bottlenecks.
• The material comes from the Sales and Trading department and states that it is not an independent research report from the J.P. Morgan Research Department; its investment-advice characteristics and compliance use should therefore be distinguished carefully.
• Much of the information comes from media reports, market news, and event notes, creating risks related to information updates, execution, and verification.

What to watch

• Actual execution of Meta's FY26/FY27 AI compute deployment plans and long-term supply agreements with Samsung, Sandisk, and others.
• The pace of implementation of Micron's increase in US domestic capital expenditures from $200bn to $250bn, and the transmission to orders for TEL, SCREEN, and silicon wafer suppliers.
• Whether AMAT's order visibility over the next eight quarters and advanced packaging equipment growth translate into revenue for equipment peers.
• Progress on TSMC AP7 Phase 2 expansion and CoWoS capacity reaching 115k/190k/225k, as well as whether ASE raises capital expenditures during its 2Q results.
• The customer mix behind OSAT CoWoS-like capacity expansion to 15k/50k/85k wfpm, particularly demand from server CPUs, networking, LPUs, Trainium, and ASICs.
• Whether Nan Ya Plastics' electronic materials OPM continues improving from approximately 20% toward higher levels, along with progress on T-glass and M10 certification.
• Customer validation, performance benchmarking, and commercial orders for Chinese GPU and AI chip vendors in the training or post-training market.
• Whether Taiwan's drone exports sustain the growth momentum implied by $148m in the first four months, and the order visibility of related companies such as Wistron, Mitac, AIDC, and Qisda.