Semiconductor Sector H2 2026 Outlook — Where Is the AI Cycle?

Global Semiconductor Market
$1.3T
2026 BofA Forecast

NVDA FY2026 Revenue
$215.9B
+65% YoY

AVGO AI Revenue (Q2)

$10.8B

+143% YoY

HBM Price Surge

+50%

Deloitte, mid-2026

📌 3-Line Summary

⚡ Key Takeaways

1. Big Tech CapEx of $500B+ is sustaining the semiconductor demand cycle — the expansion phase remains intact.
2. While NVDA dominates, AVGO's custom AI ASICs and AMD's MI400 platform will reshape the H2 2026 landscape.
3. HBM supply shortages persist through 2026 → MU and SK Hynix remain top beneficiaries; Samsung's HBM4 yield recovery is the pivotal wildcard.

🔭 AI Demand — Peak or Act II?

In 2026, combined AI infrastructure CapEx from the Big Four (Amazon, Microsoft, Google, Meta) is projected at $500B–$600B, up more than 30% year-over-year. This goes far beyond GPU demand, stimulating the entire semiconductor ecosystem — power and cooling infrastructure, network switches (Arista, Marvell), and custom silicon (TPU, Trainium).

Rather than the "AI bubble peak" scenario the market fears, inference demand is exploding as the new growth engine. Training GPU demand is becoming more saturated, but surging usage of ChatGPT, Claude, and Grok is driving exponential inference server investment.

🟢 Bull Case — Inference Cycle as the New Demand Engine

Inference requires chips far more frequently than training. With ChatGPT daily queries exceeding 1 billion, inference server investment is forecast to expand through 2027. NVDA's next-gen architecture after Blackwell (Rubin) is specifically designed to capture this demand.

🏢 Big 3 Earnings Breakdown — NVDA, AMD, AVGO

Company	2026 Revenue (Est.)	Growth	AI Highlight	Risk
NVIDIA (NVDA)	$215.9B	+65% YoY	Blackwell platform at full capacity; Q1 FY27 guidance $91B	US export controls tightening; H20 sales restrictions
Broadcom (AVGO)	$88B+ (annual est.)	+84% YoY (Q3)	AI semiconductor Q2 $10.8B; Q3 guidance $16B+	Concentration risk with Google/Apple; xPU competition intensifying
AMD	Data center $28.7B	+73% YoY	MI400 (CDNA Next, 3nm) ramp scheduled H2 2026	Software ecosystem (CUDA) gap vs. NVDA

💾 HBM — The Hidden Star of the Semiconductor Cycle

High Bandwidth Memory (HBM) is essential for AI GPUs — NVDA's H100, H200, and Blackwell architectures all require it. Deloitte forecasts HBM prices will rise over 50% by mid-2026, and Micron (MU) has already sold out its entire 2026 HBM allocation.

Supplier	HBM Market Share	Status	Investment Angle
SK Hynix	~50%	HBM3E in mass production; primary NVDA supplier	Listed on KRX (KRX: 000660) — accessible to Korean investors
Micron (MU)	~25%	2026 allocation fully sold out	US-listed; most direct access for retail investors
Samsung	~25%	HBM3E yield issues; HBM4 in development	Yield improvement = market share upside catalyst

💡 Investor Insight

For direct US exposure to the HBM theme: Micron (MU) is the most straightforward play. SK Hynix trades on the Korea Exchange (KRX: 000660). For broader sector exposure, SOXX (iShares Semiconductor ETF) or SMH (VanEck Semiconductor ETF) cover the full landscape.

🇰🇷 Samsung & SK Hynix — Positioned in the Global AI Cycle

The AI semiconductor cycle benefits both US equities (NVDA, AMD, MU) and Korean equities (Samsung, SK Hynix). SK Hynix holds the clearest direct AI cycle exposure with its ~50% HBM market share. Samsung's HBM4 yield trajectory is the critical H2 2026 catalyst to watch.

⚠️ Key Risk Factors

US Export Control Tightening: The Trump administration is maintaining and expanding restrictions on China-bound AI chip exports. NVDA's H20 (China-specific downgrade chip) is under discussion for additional restrictions.
CapEx Growth Deceleration Risk: If Big Tech earnings disappoint, AI infrastructure investment could be paced back — reigniting peak-demand concerns.
TSMC Production Bottleneck: Leading-edge node (3nm/2nm) capacity is concentrated in AI, pushing PC/smartphone chipsets lower in priority. Long-term, Intel Foundry and Samsung Foundry could absorb some overflow.
Chinese Local Semiconductor Growth: Huawei Ascend 910B/C and other domestic AI chips are accelerating their displacement of NVDA in the Chinese market.

🔴 Bear Case — Demand Peak-Out Scenario

If hyperscaler CapEx comes in below expectations (recession, rate pressure), GPU inventory could build and NVDA's valuation could compress. Historically, semiconductor cycle downturns have produced drawdowns of -40% to -60%.

📊 Major Semiconductor ETF Comparison

ETF	Issuer	Expense Ratio	Top Holdings	Characteristics
SMH	VanEck	0.35%	NVDA · TSMC · AVGO · AMD	Broadest global semiconductor exposure
SOXX	iShares	0.35%	NVDA · AVGO · AMD · MU	US-focused; near-equal weighting
SOXS/SOXL	Direxion	0.76%	SOX 3× leveraged/inverse	Short-term trading only; not for long-term holding

📋 marketbrief 투자 의견

Growth Momentum

Valuation

C+

Risk

B-

Investor Accessibility

The AI semiconductor cycle is expected to remain in expansion mode through H2 2026. Rather than concentrating in NVDA alone, a SMH or SOXX ETF core position for broad sector exposure — supplemented by MU as a direct HBM satellite holding — offers a better risk/reward profile. Given event risks (export controls, CapEx misses), dollar-cost averaging into positions is recommended.

📎 주요 출처

※ This report is provided for informational and educational purposes only and does not constitute a recommendation to buy or sell any security.

New research, when it lands

Subscribe and the next deep dive comes to you, along with the daily market brief — free, unsubscribe anytime.