1. NVDA controls ~80% of the AI accelerator market via the Blackwell ecosystem and CUDA lock-in — no short-term replacement in sight.
2. AVGO's custom ASICs (Google TPU, Meta MTIA) are becoming a structural disruptor that could reshape the market from 2027.
3. INTC's foundry pivot was first validated by the Google win, but a 2–3 year gap to profitability remains the key challenge.
This Week's Three Market Signals
The semiconductor sector delivered textbook-breaking moves over the past week. On June 4, AVGO plunged -12.59% in a single session after earnings. On June 8, INTC surged +11.19% on a single piece of news — Google's foundry contract. On June 11, MU gained +11.66%, ARM +11.32%, INTC +9.27%, and AMD +7.97% as the entire sector rallied.
These three moves are not just news reactions. They are the market beginning to price in a structural shift in the AI accelerator landscape.
| Date | Ticker | Move | Trigger |
|---|---|---|---|
| Jun 4 | AVGO | -12.59% | Q2 earnings — software segment growth concerns |
| Jun 8 | INTC | +11.19% | Google 18A foundry contract (3M+ TPUs) announced |
| Jun 11 | MU | +11.66% | HBM4 demand surge expectations + sector-wide AI rally |
| Jun 11 | INTC | +9.27% | Continued foundry order momentum |
| Jun 11 | AMD | +7.97% | MI350 shipment expectations + AI demand confirmation |
NVDA — The Structural Dominance of the Market Leader
NVIDIA holds approximately 80% market share in AI accelerators. Transitioning from H100 to its Blackwell (B200, GB200) architecture, NVDA's data center revenue reached $39.3 billion in FY2026 Q4 — up 93% year-over-year.
NVDA's moat is not just chip performance. The CUDA ecosystem — 3M+ developers, hundreds of thousands of libraries, 5+ years of optimization layers — is the true barrier to entry. Even if a competitor builds a faster chip, the cost of rewriting existing CUDA codebases keeps customers locked in.
· Blackwell B200: 4x training, 30x inference vs. H100
· NVLink switch-based DGX SuperPOD — unmatched for large cluster configurations
· CUDA / cuDNN / TensorRT ecosystem — 5+ year head start
· Rubin architecture (R100) roadmap already announced for H2 2026
· Hyperscalers accelerating custom ASIC development (Google TPU, Meta MTIA, Amazon Trainium)
· Export controls — China revenue share under ongoing pressure
· Valuation — ~45x 2026 PE; sharp correction risk if growth decelerates
· HBM supply chain concentration risk (heavy dependence on SK Hynix)
AVGO — The ASIC Revolution Behind the Shock Drop
AVGO's -12.59% on June 4 was driven on the surface by software (VMware) segment deceleration. But looking at the AI hardware business alone, the story is entirely different.
In FY2026 Q2 (May fiscal quarter), AVGO's AI revenue reached $10.8 billion (+44% YoY), and Q3 guidance pointed to $16 billion (+200% YoY). This growth is driven by explosive demand for custom ASICs — Google TPU v5, Meta MTIA, and ByteDance chips.
General-purpose GPUs (NVDA) work for any workload, but for large-scale inference on specific models, custom ASICs deliver 30–50% better performance per watt. That's why Google invests tens of billions in TPUs. AVGO is the primary beneficiary of ASIC design and packaging — a fabless model via TSMC that captures AI infrastructure investment returns without massive capital expenditure.
The June 4 sell-off is widely interpreted as a staged buying opportunity for long-term investors. Q3 AI guidance of $16B significantly beat market expectations, and annualized AI revenue of $60B+ by end of 2026 is being discussed.
INTC — Google Win: What the Foundry Pivot Really Means
Intel spent two years in relative obscurity amid restructuring and market share losses. Then on June 8, news broke that Google had signed a contract to procure 3M+ TPUs using Intel's 18A process node by 2028 — sending INTC up +11.19% in a single session.
Intel 18A combines two innovations: the gate-all-around transistor architecture RibbonFET and the backside power delivery system PowerVia. This combination is considered Intel Foundry's first node genuinely competitive with TSMC N2/N3.
✅ 18A process maturity — first validated by Google TPU win
✅ US CHIPS Act grants pending (~$8.5B)
✅ Geopolitical tailwind — US domestic foundry demand growing
⚠️ Foundry profitability target — 2027–2028
⚠️ PC and server CPU market share continuing to cede to AMD
⚠️ 18A yield stabilization is the critical unknown
The Google win is significant beyond a technology proof-of-concept. Google is strategically nurturing Intel Foundry as a TSMC alternative to diversify geopolitical supply chain risk (Taiwan concentration). This single contract gave Intel Foundry its first credibility proof point with enterprise customers.
Three-Way Comparison
| Category | NVDA | AVGO | INTC |
|---|---|---|---|
| Core AI Strategy | General-purpose GPU (Blackwell) | Custom ASIC design | AI chip foundry |
| 2026 AI Revenue | Data center ~$180B+ (annualized) | AI segment ~$60B+ (annualized est.) | Orders in progress (revenue from 2027+) |
| Market Share | AI accelerator ~80% | Custom ASIC #1 | Foundry ~2% (growing) |
| Key Strength | CUDA ecosystem + NVLink | Fabless + TSMC partnership | US domestic foundry monopoly |
| Key Risk | ASIC transition acceleration | VMware software growth slowdown | 18A yield and profitability timeline |
| 2025 P/E | ~45x | ~35x | N/A (loss-making) |
| Dividend | Minimal (0.03%) | ~1.4% | ~1.5% |
| Near-term Momentum | Bullish | Neutral (recovering) | Bullish (foundry order optimism) |
HBM Analysis — Memory is the AI Bottleneck
AI accelerator performance is increasingly limited not by compute but by memory bandwidth. LLM inference requires reading enormous model weights from GPU memory at high speed. HBM (High Bandwidth Memory) addresses this bottleneck — and the HBM supply chain is the hidden battleground of the AI arms race.
| Company | HBM Share | Key Customers | Status |
|---|---|---|---|
| SK Hynix | ~62% | NVDA (exclusive) | HBM3E in mass production; HBM4 samples H2 2025 |
| Micron (MU) | ~21% | NVDA, AMD | Surpassed Samsung; HBM3E yield improving rapidly |
| Samsung | ~17% | AMD, INTC | HBM3E yield issues reduced NVDA supply share |
The key development is that Micron has overtaken Samsung. Samsung's HBM3E yield problems have significantly reduced its share of NVDA supply, and Micron is rapidly filling the gap. MU's +11.66% on June 11 reflects the market re-pricing this structural benefit.
HBM4 targets 2x the bandwidth and 1.4x the capacity of HBM3E. SK Hynix began sample shipments in H2 2025, with mass production planned for 2026. If HBM4 is integrated into Blackwell Ultra (B300) and Rubin (R100), SK Hynix's exclusive NVDA supply position is likely to strengthen further.
Portfolio Strategy for Global Investors
Three angles to consider when approaching this sector as an international investor:
① Direct Stock Exposure
NVDA remains the clearest AI infrastructure play, but at ~45x PE, even modest growth deceleration could trigger a sharp pullback.
AVGO's June 4 sell-off brought it to a technical support zone (~$170), making a staged entry worth considering.
INTC is a speculative position — short-term momentum (foundry order expectations) coexists with a long-term restructuring story.
MU offers HBM upside + semiconductor cycle recovery as a Samsung alternative play.
② ETF Exposure
For sector exposure without single-stock risk:
SOXX (iShares Semiconductor ETF) or SMH (VanEck Semiconductor ETF) are the standard choices.
For concentrated AI accelerator exposure: NVDL (2x NVDA leveraged, for short-term trading only) — high volatility.
"Semiconductors are a cyclical sector. But AI infrastructure demand is structural growth riding on top of the cycle. Whether you see short-term corrections as fear or as entry opportunities — that judgment will define 2026 returns."
— MarketBrief Editorial
- Broadcom (AVGO) FY2026 Q2 실적 발표 — SEC 8-K 공시 (2026-06-04)
- Intel 18A Foundry 구글 TPU 수주 — TrendForce (2026-06-09)
- NVIDIA Blackwell Architecture Overview — NVIDIA 공식 페이지
- HBM 공급사 동향 — TrendForce (2026-03-09)
- Yahoo Finance — AVGO, INTC, MU, NVDA, AMD 주가 데이터
- Bloomberg Intelligence — 반도체 섹터 밸류에이션 리포트 (유료 구독 리포트)
※ 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.