Amid macro uncertainty, UBS believes the AI and memory upcycle is still continuing

Overview

This UBS APAC Technology Strategy monthly sector note centers on three questions: whether AI investment has peaked, whether the memory semiconductor upcycle can continue, and how sharply demand for PCs and smartphones may decline. The conclusion is constructive: AI infrastructure, cloud capex, HBM, and server/storage SSD demand are still expanding enough to offset weakness in traditional end markets, allowing the memory industry to sustain exceptionally strong conditions from 2026 through 2027.

Core views

UBS's core view is: first, the AI cycle is not over, with hyperscaler capex still growing rapidly in 2026 and the AI semiconductor supply chain still seeing solid growth in 2027; second, the DRAM and NAND upcycles will last longer than previously expected, with DRAM supply-demand tightness potentially lasting until 4Q27; third, AI server and data center memory demand is materially stronger than the demand decline in traditional markets such as PCs and smartphones; fourth, memory makers are increasing capex and WFE spending, but equipment lead times, supply-chain constraints, and engineering-resource bottlenecks mean supply cannot quickly catch up with demand; fifth, at the stock level, UBS is more positive on memory leaders, HBM test equipment, materials, and selected equipment names, while remaining wary of valuation and share risks in some back-end equipment and traditional end-market chains.

Analysis framework

The report combines top-down industry demand modeling with bottom-up coverage of supply-chain companies: it first tracks cloud capex, AI accelerator shipments, HBM bit demand, the server DRAM mix, and DRAM/NAND prices, and then maps those inputs to market share, profitability, valuation multiples, and target prices for Samsung Electronics, SK Hynix, DI Corporation, Eugene Tech, Hanmi Semiconductor, and others.

Methodology notes

• Industry cycle analysisSupply-demand gap and price cycle framework

  Use AI servers, traditional end markets, WFE spending, and capacity bottlenecks to judge the length of the DRAM/NAND upcycle.

  The report offsets growth in HBM, server DDR, and server/storage SSD demand against declines in PCs and smartphones, and combines equipment lead times, engineering resources, and wafer-capacity migration to judge how quickly supply can respond.

• Capex trackingHyperscaler capex framework

  Use capex from the top 11 hyperscalers as a leading indicator of AI infrastructure demand.

  The report notes that capex from the top 11 cloud providers is expected to grow 64% YoY in 2026 and argues that strong growth will continue in 2027, which is the main basis for continued AI semiconductor and memory demand.

• Valuation methodsP/BV, P/E, and long-term ROE/CoE valuation

  Apply price-to-book or price-to-earnings multiples to different covered companies, combined with long-term ROE, CoE, and EPS growth assumptions.

  For example, Samsung Electronics is valued at 2.09x forward 12-month P/BV, SK Hynix at 2.86x forward 12-month P/BV, while DI Corporation and Eugene Tech are valued using 2027E P/E.

Asset mapping & comparison

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

• Memory / HBM / DRAM / NAND

  Core beneficiary

  Strengths

  Strong demand from AI servers, HBM, server DDR, and server/storage SSDs, with DRAM supply-demand tightness potentially lasting until 4Q27.

  Weaknesses

  Cycle durability depends heavily on hyperscaler capex and AI application adoption, while traditional end-market demand is still declining.

  Comparison

  Compared with PCs, smartphones, and display panels, memory benefits more directly from the expansion of AI infrastructure.

  Risks

  If cloud providers cut capex, AI monetization falls short of expectations, or supply expands faster than expected, prices and earnings could retreat.

• Samsung Electronics

  Memory leader and HBM chaser

  Strengths

  Leadership in DDR/NAND, improving HBM execution, narrowing Foundry/LSI losses, and large room for shareholder returns.

  Weaknesses

  Its HBM competitive position still needs further validation, and the smartphone business is not immune to margin pressure.

  Comparison

  Compared with SK Hynix, Samsung has a larger advantage in traditional memory scale, but its HBM leadership is still in catch-up mode.

  Risks

  HBM qualification progress, advanced DRAM design execution, capex timing, and weakening traditional end-market demand.

• SK Hynix

  HBM leader

  Strengths

  UBS expects its HBM bit share to be 51%/44% in 2026E/2027E, and stronger traditional memory demand should support the stock.

  Weaknesses

  Adjustments to its China manufacturing footprint may create cost pressure and capacity-migration friction.

  Comparison

  It leads Samsung and other suppliers in HBM, but customer three-supplier strategies may gradually dilute share.

  Risks

  HBM demand volatility, DRAM capacity adjustments in China, EUV equipment availability, and intensifying competition.

• DI Corporation

  Beneficiary of HBM test equipment

  Strengths

  Seen as the Korean memory test company best positioned to benefit from the HBM4 transition, with a relatively high expected share of SK Hynix HBM4 KGSD testing.

  Weaknesses

  Customer concentration is high, and dependence on SK Hynix orders is obvious.

  Comparison

  Compared with traditional DDR test equipment, HBM4 KGSD burn-in test tools have higher ASPs and margins.

  Risks

  Customer ramp timing, competition from first-tier suppliers such as Advantest, and HBM4 test demand coming in below expectations.

• Eugene Tech

  Beneficiary of DRAM WFE and process equipment

  Strengths

  Benefits from growth in DRAM WFE TAM, higher etch-tool attach rates, and share gains in PEALD, plasma, and related products.

  Weaknesses

  Highly exposed to the capex cycle of memory manufacturers.

  Comparison

  Compared with some back-end equipment names, Eugene Tech is more tied to front-end process expansion and advanced DRAM process upgrades.

  Risks

  Capex spending by Samsung and SK Hynix falling short of expectations, customer qualification delays, and changes in advanced process roadmaps.

• Hanmi Semiconductor

  Risk name in HBM TCB equipment

  Strengths

  It can still benefit from some TCB tool demand tied to new HBM4 capacity at SK Hynix and Micron.

  Weaknesses

  UBS worries that it may not win Samsung HBM4 TCB orders, and customers may diversify suppliers to reduce risk.

  Comparison

  Unlike the more constructive views on memory equipment and test-chain names, UBS rates Hanmi a Sell.

  Risks

  Market-share losses, Samsung's in-house preference, and stronger competition from ASMPT and Hanwha Vision.

• PC / Smartphones / Displays

  Cautious area

  Strengths

  There may be a low-single-digit recovery in 2027, but near term there is no demand support comparable to the AI memory chain.

  Weaknesses

  PC and smartphone shipments are expected to decline in 2026, and improvement in display panel profitability remains uncertain.

  Comparison

  Compared with AI servers and the memory chain, traditional end-market demand is less elastic.

  Risks

  Further weakening of end-user demand, price competition, OEM order downgrades, and delayed panel supply-demand repair.

Key data

• Capex of the top 11 hyperscalers2026E YoY growth of 64%; still strong growth in 2027EUsed to support the view that AI infrastructure demand is still expanding.
• AI accelerator shipments2027E YoY growth of 47%; 2026E YoY growth of 30%The report says preliminary 2027 AI semiconductor supply-chain estimates still point to solid growth.
• HBM bit demand2026E YoY growth of 80% to 31bn Gb; 2027E YoY growth of 73% to 54bn GbHBM is the key driver of the memory upcycle.
• DRAM and NAND revenue2026E at US$803bn; 2027E at US$1.21tnSignificantly above the prior cycle peak of US$153bn in 2018.
• Server DRAM shareIn 2027E, server DRAM (DDR+HBM) will account for about 50% of total DRAM bits, versus 33% in 2025Shows a shift in demand structure toward data centers.
• Smartphone shipments2026E YoY decline of 10%; 2027E YoY growth of 2%Weakness is more pronounced in the low-end Android segment.
• PC shipments2026E YoY decline of 4%; 2027E YoY growth of 2%Some OEM guidance to suppliers suggests low- to mid-teens unit declines in 2026.
• DDR contract pricesExpected to rise 95% QoQ in 1Q26; expected to rise 37% QoQ in 2Q26Reflects strong near-term price elasticity.
• NAND contract pricesExpected to rise 80% QoQ in 1Q26; expected to rise 40% QoQ in 2Q26NAND prices are also showing a strong recovery.
• SK Hynix HBM share51% in 2026E; 44% in 2027EUBS expects SK Hynix to remain the HBM leader, although a three-supplier strategy is forming.

Impact & implications

For investors, the report strengthens the preference for APAC tech exposure to AI compute, HBM, DRAM/NAND, semiconductor equipment, and materials. If UBS's forecasts play out, the earnings and cash flow visibility of memory leaders should improve materially, and equipment and test-tool suppliers should also benefit from HBM4, HBM4E, and advanced DRAM expansion. By contrast, PCs, smartphones, and display panels still face demand pressure and margin headwinds, so those assets require a clearer distinction between structural growth and cyclical rebound.

Risks

• If hyperscaler capex comes in below expectations, demand for AI semiconductors, HBM, and server memory will weaken.
• If AI application deployment, especially agentic/reasoning AI and AI coding, is slower than expected, demand durability could be affected.
• If DRAM, NAND, and HBM supply expands faster than expected, peak pricing could come in below the report's assumptions.
• PC and smartphone demand could fall more sharply than currently forecast, dragging down the traditional memory and hardware chains.
• A three-supplier strategy for HBM customers could dilute the share of incumbent leaders.
• China manufacturing adjustments, EUV equipment constraints, and advanced-node migration could raise execution costs for SK Hynix and others.
• Long-term supply agreements may cap peak memory prices and alter the distribution of earnings leverage.

What to watch

• The direction of 2026 and 2027 capex revisions for the top 11 hyperscalers.
• Changes in AI accelerator shipments, HBM bit demand, and SOCAMM penetration.
• Whether the DRAM supply-demand gap really lasts through 4Q27.
• The persistence of DDR and NAND contract prices after 1Q26 and 2Q26.
• Samsung's HBM4 qualification progress and its position in the AMD, OpenAI, Google, AWS, and Meta supply chains.
• SK Hynix HBM capacity expansion, Yongin campus progress, and capacity adjustments in China.
• Whether OEM order guidance for PCs and smartphones is revised down further.
• The pace at which testing equipment, WFE, and materials orders are released as HBM4/HBM4E ramps.