Asia memory vs US AI semiconductor: two sides of the AI chip trade

The AI build-out runs on two semiconductor trades. Asia memory semiconductor owns the DRAM, NAND, and HBM supply chain of Korea, Japan, and Taiwan; US AI semiconductor owns the GPUs, custom accelerators, and dollar-listed foundry access that buy it. This page compares demand drivers, cyclicality, currencies, valuation, and how each side is investable today.

How do they differ?

DimensionAsia Memory SemiconductorUS AI Semiconductor
What you ownThe DRAM, NAND, and HBM makers of Korea, Japan, and Taiwan, plus the equipment and test suppliers that build their fabsUS-listed GPU, custom-accelerator, memory, networking-silicon, and foundry names across the AI compute stack
Representative tickers000660 (SK Hynix), 005930 (Samsung), 285A (Kioxia), 8035 (Tokyo Electron), 2408 (Nanya), 042700 (Hanmi)NVDA, AVGO, AMD, MU, TSM, ARM, MRVL, INTC, CBRS, SNDK
AI exposureSupply side of AI memory: SK Hynix holds roughly 62% of HBM shipments and completed the world's first HBM4 developmentDemand side of AI compute: NVIDIA's data-center revenue hit $75.2B in Q1 FY2027, up 92% year over year
Demand driverA memory supercycle: WSTS forecasts memory revenue jumping from $230.0B in 2025 to $803.9B in 2026, up 249.5%Hyperscaler capex of roughly $700B planned for 2026 across Alphabet, Meta, Amazon, and Microsoft
Cyclicality & pricing powerCommodity price cycles: Nanya's DRAM selling prices rose more than 70% in a single quarter, and can fall as fastPlatform pricing power: CUDA and Blackwell anchor margins, and Broadcom's AI revenue grew 143% to $10.8B in fiscal Q2 2026
Geography & currencyKRX, TSE, and TWSE listings in won, yen, and NT dollars; the whole chain sits in three export-dependent East Asian marketsNYSE and Nasdaq listings in US dollars, including TSMC's ADR; exposed to shifting US export policy toward China
Valuation profileEarnings-explosion pricing: SK Hynix's Q1 2026 operating margin hit 72%, so the multiple hinges on how long the cycle lastsGrowth-embedded pricing: years of AI capex are in the multiples, and newcomer Cerebras took 86% of 2025 revenue from two customers
How each is investableNo licensed tracking ETF yet; exposure means individual names such as SK Hynix, Samsung Electronics, or Tokyo ElectronIn Korea, the KODEX US AI Semiconductor TOP3 Plus ETF (0151S0) has tracked the concept since January 13, 2026

When does Asia memory semiconductor make sense?

Asia memory suits investors who want the supply side of AI, where shortage sets the price. WSTS forecasts global memory revenue jumping from $230.0 billion in 2025 to $803.9 billion in 2026, a 249.5% surge and the biggest driver of a $1.51 trillion chip market (WSTS). The chokepoint is high-bandwidth memory: SK Hynix completed the world’s first HBM4 development in September 2025 and holds roughly 62% of HBM shipments (SK hynix; Astute Group), while Samsung began the industry’s first mass sales of HBM4 for Nvidia’s Vera Rubin platform in Q1 2026 (Samsung).

HBM4, a symbolic turning point beyond the AI infrastructure limitations, will be a core product for overcoming technological challenges.

— Justin Kim, President & Head of AI Infra, SK hynix

The results of that shortage are already historic. SK Hynix posted a 72% operating margin in Q1 2026 on revenue up 198% (SK hynix), Kioxia rode AI data-center SSD demand to record fiscal-year revenue of 2,337.6 billion yen and reportedly sold out its 2026 NAND capacity in advance (Kioxia; DIGITIMES), and Nanya’s DRAM selling prices rose more than 70% in a single quarter (Nanya Technology). The trade-off is depth of cycle: the same quarter saw Hanmi Semiconductor’s revenue fall 65% as HBM equipment orders cooled (DART). This is a basket priced on how long a commodity shortage lasts, in three currencies, on three East Asian exchanges.

When does US AI semiconductor make sense?

US AI semiconductor suits investors who want the demand side, where committed hyperscaler spending funds the platform layer. For 2026, Alphabet guided capital spending to $180-190 billion, Meta to $125-145 billion, Amazon to about $200 billion, and Microsoft’s March 2026 quarter alone ran $31.9 billion, roughly $700 billion of combined investment (Alphabet Q1 2026 call; Meta; CNBC; CNBC). The anchor remains NVIDIA, whose Q1 FY2027 revenue was a record $81.6 billion, up 85%, with data-center revenue of $75.2 billion, up 92% (NVIDIA).

Agentic AI has arrived, doing productive work, generating real value and scaling rapidly across companies and industries.

— Jensen Huang, Founder and CEO, NVIDIA

The basket also holds the fastest-growing layers around the GPU. Broadcom’s AI semiconductor revenue grew 143% to $10.8 billion in fiscal Q2 2026 with $16.0 billion guided for fiscal Q3 (Broadcom 8-K), and Micron, the only US-based HBM maker, nearly tripled fiscal Q2 2026 revenue year over year to $23.86 billion (Micron 8-K). Access is also simpler: the KODEX US AI Semiconductor TOP3 Plus ETF (0151S0) has tracked the concept on the Korea Exchange since January 13, 2026, concentrating over 60% in NVIDIA, TSMC, and Broadcom (Samsung Asset Management).

Which risks separate the two sides?

The Asia side carries commodity-cycle risk first: memory prices that rose 70% in a quarter can fall as fast when supply catches up, and WSTS’s 249.5% growth forecast is a forecast, not a floor (WSTS). The whole chain also sits in three export-dependent markets exposed to US-China controls. The US side carries concentration and policy risk instead: a handful of hyperscaler capex budgets fund the demand, export rules toward China have shifted repeatedly (Council on Foreign Relations), and newly public names cut both ways, with two UAE-linked customers making up 86% of Cerebras’s 2025 revenue (Cerebras S-1). The failure modes differ: a memory glut would hit the Asia basket first, while an AI capex pause would hit both, starting with the US names that priced the growth in.

Verdict: which fits which investor?

Asia memory is the supply-side, deep-cycle play: 2026 earnings are historic, but prices that rose 70% in a quarter can fall as fast when supply catches up. US AI semiconductor is the demand-side platform play: roughly $700B of hyperscaler capex funds it, and a licensed ETF already tracks it. They are two layers of one AI stack, and many investors pair them.

Related concepts & themes

FAQ

Should I invest in Asia memory or US AI semiconductor stocks for AI exposure?

They are complements rather than substitutes: one supplies the memory every AI accelerator needs, the other supplies the accelerators themselves. The link is direct, since SK Hynix alone sold 7.78 trillion won of product to Nvidia in Q1 2026, about 14.8% of its revenue (SK hynix). The Asia side rides a memory supercycle that WSTS sizes at $803.9 billion of 2026 revenue, up 249.5% (WSTS), while the US side rides roughly $700 billion of 2026 hyperscaler capex (Alphabet Q1 2026 call). Many investors hold both layers and size them by cycle tolerance.

Which side is more cyclical?

Asia memory, by a wide margin. Memory is a commodity whose price is set by supply and demand each quarter: Nanya's DRAM selling prices rose more than 70% in Q1 2026 alone (Nanya Technology), and in the same quarter Hanmi Semiconductor's revenue fell 65% as HBM equipment orders cooled (DART). The US side is cyclical too, but through a different channel: its demand is concentrated in four hyperscaler capex budgets, while platform economics such as NVIDIA's CUDA ecosystem and Broadcom's custom-ASIC franchises cushion pricing between cycles (NVIDIA; Broadcom 8-K).

Do the two semiconductor concepts overlap?

At the edges, deliberately. Micron, the only US-based HBM maker, sits in the US basket even though it competes head-on with SK Hynix and Samsung; its fiscal Q2 2026 revenue nearly tripled to $23.86 billion on the same AI memory shortage that powers the Asia side (Micron 8-K). TSMC appears in the US basket through its NYSE-listed ADR while its fabs sit in Taiwan (TSMC 6-K). The cleanest distinction is functional: the Asia concept owns the memory supply chain, the US concept owns the compute and the dollar-listed access to it.

Is there an ETF for each side?

Only for the US side so far. The KODEX US AI Semiconductor TOP3 Plus ETF (0151S0) from Samsung Asset Management listed on the Korea Exchange on January 13, 2026, concentrates over 60% in NVIDIA, TSMC, and Broadcom, and charges a 0.49% total annual fee (Samsung Asset Management; newsroom). The Asia memory concept has no licensed tracking ETF yet; the Akros Asia Memory Semiconductor Top 10 Index (AAMST10), published May 26, 2026, frames its 10-stock universe (Akros Technologies), so exposure today means individual names such as SK Hynix (000660) or Tokyo Electron (8035). Always check fees, holdings, and risk before investing.

How do the two sides fit together in one portfolio?

A common framing is one demand sleeve and one supply sleeve of the same AI build-out. The US side gives dollar-denominated exposure to the platform layer, where the spending is committed in public capex guidance: Alphabet at $180-190 billion, Meta at $125-145 billion, and Amazon at about $200 billion for 2026 (Alphabet Q1 2026 call; Meta; CNBC). The Asia side adds the chokepoint that capex must buy through, HBM and DRAM, at deeper-cycle risk: SK Hynix printed a 72% operating margin in Q1 2026, a number that embeds the top of a cycle (SK hynix). Pairing them spreads one thesis across two failure modes.

Sources & references

  1. Global Semiconductor Market Surges Beyond $1.5 Trillion in 2026 (WSTS Forecast) · World Semiconductor Trade Statistics (WSTS), 2026-06-02
  2. SK hynix Reports Q1 2026 Business Results · SK Hynix Newsroom, 2026-04-23
  3. SK hynix Completes World's First HBM4 Development and Readies Mass Production · SK Hynix Newsroom, 2025-09-12
  4. Samsung Electronics Announces First Quarter 2026 Results · Samsung Electronics (Samsung Global Newsroom), 2026-04-30
  5. Consolidated Financial Results for the Fiscal Year Ended March 31, 2026 · Kioxia Holdings Corporation, 2026-05-15
  6. Nanya Technology Q1 2026 Results Press Release · Nanya Technology Corporation, 2026-04-13
  7. 한미반도체 분기보고서 (2026.03) · Hanmi Semiconductor / DART (Financial Supervisory Service), 2026-05-15
  8. SK Hynix holds 62% of HBM, Micron overtakes Samsung, 2026 battle pivots to HBM4 · Astute Group, 2026-01-15
  9. NVIDIA Announces Financial Results for First Quarter Fiscal 2027 · NVIDIA Corporation (Newsroom), 2026-05-20
  10. Broadcom announces second quarter fiscal year 2026 financial results (SEC 8-K, Exhibit 99.1) · Broadcom Inc. / SEC EDGAR, 2026-06-03
  11. Micron Technology, Inc. Reports Results for the Second Quarter of Fiscal 2026 (SEC 8-K, Exhibit 99.1) · Micron Technology, Inc. / SEC EDGAR, 2026-03-18
  12. TSMC Reports First Quarter EPS of NT$22.08 (Form 6-K, Exhibit 99.1) · Taiwan Semiconductor Manufacturing Company / SEC EDGAR, 2026-04-16
  13. Alphabet (GOOGL) Q1 2026 earnings call transcript · The Motley Fool, 2026-04-29
  14. Meta Reports First Quarter 2026 Results (SEC 8-K, Exhibit 99.1) · Meta Platforms, Inc. / SEC EDGAR, 2026-04-29
  15. Amazon (AMZN) Q1 earnings report 2026 · CNBC, 2026-04-29
  16. Microsoft (MSFT) Q3 earnings report 2026 · CNBC, 2026-04-29
  17. Cerebras Systems Inc. Form S-1 Registration Statement · Cerebras Systems Inc. / SEC EDGAR, 2026-04-17
  18. The Consequences of Exporting Nvidia's H200 Chips to China · Council on Foreign Relations, 2025-12-15
  19. KODEX 미국AI반도체TOP3플러스 ETF (0151S0) — product page · Samsung Asset Management, 2026-01-13
  20. KODEX 미국AI반도체TOP3플러스 ETF (0151S0) — newsroom listing article (1/13 상장) · Samsung Asset Management, 2026-01-06
  21. Akros Asia Memory Semiconductor Top 10 Index (AAMST10) · Akros Technologies, Inc., 2026-05-26
  22. Kioxia sells out 2026 NAND capacity as AI demand tightens global memory supply · DIGITIMES, 2026-02-23