UBS believes the APAC tech main theme still lies in the AI and memory upcycle

Overview

This UBS APAC tech strategy report revolves around the key industry debates in April 2026, including whether the AI rally should be exited, whether the memory semiconductor upcycle will continue, and the scale of the downturn in smartphone and PC demand. Overall, the report believes that AI infrastructure investment and demand from the AI semiconductor supply chain will continue to drive growth, especially for HBM, server DRAM, server/storage SSDs and the related equipment and materials chain; however, traditional end-market demand is under pressure, so assets tied to PCs, smartphones and display panels warrant greater caution.

Core views

The core views are: first, the AI theme remains durable, with capex by leading hyperscalers, AI accelerators and HBM demand still trending up; second, the memory cycle will extend into 2027, with DRAM supply unlikely to catch up with demand for some time, while NAND also benefits from AI storage demand; third, AI/data center storage demand can offset memory demand destruction from traditional applications such as smartphones and PCs; fourth, the report prefers Memory, WFE, MLCC and OSAT, while avoiding or remaining cautious on PCs, smartphones and displays.

Analysis framework

The report combines industry strategy with single-name coverage: it first judges the sector direction based on AI capex, hyperscaler investment, AI accelerator shipments, HBM bit demand and memory pricing cycles under macro uncertainty, and then maps those views to memory makers, equipment suppliers, materials companies, OSAT and display companies, while assigning ratings and target prices to core covered names such as Samsung Electronics, SK Hynix, DI Corporation, Eugene Tech and Hanmi Semiconductor.

Methodology notes

• industry cycle analysismemory upcycle framework

  Use DRAM, NAND, HBM demand, supply constraints, pricing and WFE capex to judge where the memory cycle sits.

  UBS compares AI server, HBM, server DDR and SSD demand within the same supply-demand framework as traditional phone and PC demand, concluding that the AI-side increment outweighs traditional demand destruction and that DRAM shortages may persist through 4Q27.

• capex chain analysishyperscaler capex transmission framework

  Infer demand for AI accelerators, HBM, server storage and semiconductor equipment from cloud providers' capex.

  The report focuses on capital expenditures by the top 11 hyperscalers, the cooperation and investment relationships among Anthropic, Amazon, Microsoft, NVIDIA and Google, and the pull-through effect of AI accelerator platform iterations on HBM capacity and bandwidth.

• relative allocation strategysub-industry preference ranking

  Compare the risk-reward of AI beneficiaries versus traditional end-market chains within the same tech-hardware framework.

  The report explicitly prefers Memory, WFE, MLCC and OSAT because they benefit from AI and the memory cycle, while staying cautious on PCs, smartphones and displays because demand forecasts have been cut and margin pressure is more visible.

Asset mapping & comparison

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

• Samsung Electronics

  Beneficiary of the memory upcycle and HBM execution improvement

  Strengths

  Strong position in DDR/NAND, improving HBM4 samples and customer qualification, expected narrowing of foundry/LSI losses, and a clear shareholder return policy.

  Weaknesses

  HBM execution still needs to catch up with the leading peer, and the smartphone business is not immune to margin pressure.

  Comparison

  Versus SK Hynix, Samsung is stronger in traditional DDR/NAND scale but weaker in HBM leadership.

  Risks

  HBM qualification falls short of expectations, further weakening in traditional end-market demand, or too-rapid supply recovery from WFE expansion.

• SK Hynix

  HBM leader and core beneficiary of AI storage

  Strengths

  Expected HBM bit share of 51%/44% in 2026E/2027E, high long-term ROE, and support from both AI and traditional server memory demand.

  Weaknesses

  Adjustments to its manufacturing footprint in China could create cost and execution pressure.

  Comparison

  Stronger than most peers in HBM share and execution, but customer multi-sourcing strategies may create share pressure.

  Risks

  HBM demand volatility, costs from China capacity adjustments, peers catching up, and customer de-risking procurement.

• DI Corporation

  Beneficiary of HBM4 test equipment

  Strengths

  Could gain a large share in SK Hynix's HBM4 KGSD burn-in testing; HBM test tools have higher ASP and margins than traditional DDR test tools.

  Weaknesses

  Customer concentration is relatively high, and earnings are sensitive to the HBM4 ramp schedule.

  Comparison

  Compared with global back-end test equipment peers, it trades at a valuation discount but has higher EPS growth.

  Risks

  Changes in SK Hynix's supplier strategy, delays in HBM4 mass production, or intensifying competition in test equipment.

• Eugene Tech

  Beneficiary of DRAM WFE expansion and process upgrades

  Strengths

  Benefits from DRAM WFE TAM growth, rising etch tool penetration, and greater share in PEALD and plasma products.

  Weaknesses

  Needs key customer qualification and share gains to materialize.

  Comparison

  The report sees a clear valuation discount versus peers, and the positives are not yet fully reflected.

  Risks

  Weaker-than-expected WFE recovery at Samsung and SK Hynix, delayed customer qualification, and fluctuating equipment investment cycles.

• Hanmi Semiconductor

  Part of the HBM TCB equipment chain, but with a negative view

  Strengths

  Incremental HBM4 capacity in 2026 can still support some TCB tool demand.

  Weaknesses

  Orders come more from upgrades than from new builds, the probability of Samsung orders is low, and customer de-risking could lead to share losses.

  Comparison

  Versus competitors such as ASMPT and Hanwha Vision, the report believes Hanmi's long-term share could be diluted.

  Risks

  If Samsung orders or TCB demand surprise to the upside, the negative view could face upside risk; otherwise, market-share erosion would pressure valuation.

• PC and smartphone supply chain

  Assets under pressure from traditional end-market demand

  Strengths

  Shipments may return to modest growth in 2027, and AI PCs or high-end handsets could create localized structural opportunities.

  Weaknesses

  Both PC and smartphone shipments for 2026 have been revised down, with larger pressure in low-end and Android demand.

  Comparison

  Compared with the AI server and memory chains, the traditional end-market chain lacks equally strong demand drivers.

  Risks

  Further macro deterioration, OEM inventory adjustments, price competition and margin pressure.

Key data

• Top hyperscaler capex2026E YoY +64%, with strong growth continuing in 2027EUsed to support the view that AI infrastructure demand has not shown a meaningful slowdown yet.
• AI accelerator shipments2027E YoY +47%; 2026E YoY +30%The report's preliminary estimate for the 2027 AI semiconductor supply chain.
• HBM bit demand2026E YoY +80% to 31bn Gb; 2027E YoY +73% to 54bn GbHBM is the core driver of AI storage demand.
• DRAM+NAND industry revenue2026E US$803bn; 2027E US$1.21tnWell above the prior cycle peak of US$153bn in 2018.
• DRAM cycle viewExpected to remain in shortage through 4Q27Driven by HBM displacing DDR capacity, the traditional server refresh cycle and limited wafer capacity expansion.
• NAND cycle viewExpected upcycle through 3Q27Supported by AI-related server and storage SSD demand.
• Smartphone shipments2026E YoY -10%, 2027E YoY +2%Downside pressure is more concentrated in the low-end and Android markets.
• PC shipments2026E YoY -4%, 2027E YoY +2%Some OEMs have told suppliers that 2026 shipments could decline by double digits.
• DDR contract price forecast1Q26 +95% QoQ, 2Q26 +37% QoQNear-term memory price elasticity is significant.
• NAND contract price forecast1Q26 +80% QoQ, 2Q26 +40% QoQRising prices support memory makers' earnings.
• Samsung ElectronicsBuy, PT Won266,000Beneficiary of DDR/NAND supply constraints, HBM execution improvement and narrowing foundry/LSI losses.
• SK HynixBuy, PT Won1,700,000Expected HBM bit share of 51%/44% in 2026E/2027E, maintaining a leading position.
• DI CorporationBuy, PT Won45,000Benefits from HBM4 test equipment demand and a larger share in SK Hynix's supply chain.
• Hanmi SemiconductorSell, PT Won100,000The report worries that HBM TCB orders, Samsung opportunities and market share could be diluted over the long term by competitors.

Impact & implications

For portfolios, the implication is that AI capex and tight memory supply-demand conditions remain the most important positive drivers for APAC tech; memory makers, HBM equipment, test equipment, materials and parts of the packaging chain have the most attractive allocation value. In contrast, companies dependent on PC, smartphone and display panel end-demand face demand downgrades and margin pressure. If long-term supply agreements expand, memory price peaks may be lower, but earnings and cash-flow visibility could improve.

Risks

• If hyperscaler capex slows, the assumptions for AI accelerators, HBM and server storage demand would weaken.
• The industry's revenue and pricing forecasts imply strong cyclical elasticity; if supply comes on faster than expected, prices and profits could fall short of forecasts.
• PC and smartphone demand declines could be worse than assumed in the report and spill over into components, displays and traditional memory demand.
• The competitive landscape for HBM suppliers may change, and customer multi-sourcing strategies could compress leading players' shares or pricing.
• Long-term supply agreements may lower memory price peaks; while they improve cash-flow visibility, they would limit upside in the cycle.
• China manufacturing footprints, EUV equipment availability and geopolitical restrictions may affect capacity planning for companies such as SK Hynix.
• Single-name target prices are highly dependent on 2026E-2027E earnings and P/BV or P/E assumptions; if ROE, CoE or EPS CAGR changes, the valuation conclusions will change.

What to watch

• Whether agentic/reasoning AI and AI coding deployments can generate real application demand.
• Whether capex from leading hyperscalers is revised further upward in 2026-2027.
• HBM bit demand, HBM ASP and the mass-production qualification progress for HBM4/HBM4E.
• Whether the DRAM shortage really lasts through 4Q27 and whether the NAND upcycle lasts through 3Q27.
• Samsung's HBM4 qualification progress, and second-sourcing progress at OpenAI, Google, AWS and Meta.
• Whether SK Hynix's 2026E/2027E HBM share remains near 51%/44%.
• Whether PC and smartphone shipments continue to undershoot forecasts, especially in low-end and Android demand.
• Whether WFE lead times, engineering installation resources and equipment supply-chain bottlenecks limit supply recovery.
• The pricing bands, locked-volume ratios and cycle impact of long-term memory supply agreements.