SILICON NEXUS

Signals

Semiconductor supply-demand signals, ordered by the questions investors actually ask. Experimental signals labeled.

Is AI investment sustainable?

Cycle Pressure Index

AI Infra Investment Sustainability

+40

Critical Overheating

-50-250+25+50

Base Pressure

+25/25

CAPEX/OCF = 70.3% (avg 46.5%)

Top 0% historically · 33 windows

Reality Check

+15/25

GPU Rental-28.7%
DRAM Spot0.0%
DSI Inventory+4.3%

Reading: AI Hyperscaler 4 (Alphabet, Microsoft, Amazon, Meta) spending 70.3% of operating cash flow on capex. Historical avg is 46.5%. Currently at all-time high. (Custom composite indicator, not an industry benchmark)

CAPEX/OCF Trend (TTM, 4-Company Agg.)

As of Q1'26 · 2026-08-22

Is HBM short or oversupplied?

HBM Market Tightness

61
Tight

Combines hyperscaler CAPEX, memory maker HBM allocation, and TSMC packaging load to assess HBM market conditions.

0 Oversupply50 Balanced100 Tight
Hyperscaler CAPEX (35.7%)
81

5 companies, total CAPEX $132B, 3/5 direction up

HBM Supply Allocation (35.7%)
36

HBM rev $27B, 28% of DRAM (SK Hynix $12.2B, Samsung $10.3B, Micron $4.2B)

TSMC Packaging Load (28.6%)
68

TSMC rev $1270B, QoQ +12.0%, YoY +36%

$132B
5-Co CAPEX Total
$27B
3-Co HBM Revenue
28%
HBM Share of DRAM
3/5
CAPEX Direction Up
80+Very Tight — supply shortage
60-80Tight — demand exceeds supply
40-60Balanced — supply-demand stable
20-40Loosening — supply catching up
0-20Oversupply — investment caution

Based on earnings filings. Price-direction proxy in data accumulation (target: 2026-09). Excluded from score until validated. Base FX: ₩1,450/$.

How tight is AI infrastructure supply-demand?

AI Infrastructure Tension

Geometric mean of fact axes (GPU·DC) · CapEx sentiment separated as narrative

27
Expanding

Geometric mean of GPU rental prices and DC construction pipeline measures AI infrastructure supply-demand tension.

0 Excess Pipeline50 Balanced100 Overheating
GPU Rental Price
50

H100 $1.47 → $1.47 (+0.0%)

DC Pipeline
12

Construction 2173MW / Total 4388MW = 50% pipeline

Unclassified 70 entries · 10.0 GW (excluded)
Narrative (News Sentiment)
CapEx Sentiment
95

CapEx news: 102 positive, 10 negative out of 117 (86% positive)

Gap:Fact 27 vs Sentiment 95 (-67.1)
80+Overheating — construction lags demand
60-80Tight — supply short, build urgent
40-60Balanced — construction meets demand
20-40Expanding — construction decelerating
0-20Excess pipeline — massive build underway

US AI Data Centers — Construction Status

Circle=capacity, color=stage. Construction pipeline dwarfs operational

Under Construction 70 sites · 2.2 GW
Operational 33 sites · 2.2 GW
Proposed 44 sites ·

Some coords/capacity note-inferred · CONUS only (excl. Alaska & overseas) · Weekly refresh

US — Who Is Building

Private-led, no state procurement. Asset proxy, not a stock signal

Hyperscaler

Big-tech captive demand (Meta·Google·Amazon·Microsoft)

3.4 GW78%

52 facilities

Meta, Microsoft

Neocloud

AI-native GPU service (CoreWeave·xAI·Oracle/Stargate)

-0%

13 facilities

Independent

Non-big-tech developers — largest single pipeline

-0%

26 facilities

MW = construction-phase estimates · Mapping: manual rules (update when new companies added)

Sovereign AI — National Programs

Color=ownership model. US is private-led, the world is state-led

Procurement (2)
Subsidy (4)
Gulf SWF (2)
Announced = dashed (not yet executing)

Announced vs building distinguished, source-verified · GPU units heterogeneous (classified by model) · Semi-annual refresh

World — How States Acquire AI

Color = ownership model. SWF/procurement/subsidy patterns. 5 verified countries only

Gulf SWF
Saudi Arabia
HUMAIN
announced
18KGPU
Tier 2

사우디 파트너십 → 1차 18,000 GB300 인도, 3년 최대 600,000 발표(2025-11). 미 재무부 수출허가 throttle 下(2025-11-20 조건부). 500MW 목표. xAI가 첫 고객

United Arab Emirates
G42 / Stargate UAE
building
35KGPU
Tier 2

Stargate UAE 1GW 클러스터(1차 200MW, 2026 가동), 35,000 Blackwell 확보. Mubadala/MGX 국부펀드. Microsoft $15.2B 투자·美 수출허가 첫 케이스(주권인데 미국 의존 아이러니). G42 지분 MS 보유

Procurement
India
IndiaAI Mission
building
34KGPU
$1.24B
Tier 1

34,000 GPU 가동 중(Rs 115-150/GPU-hr, 시장가 -42%), 2026말 100,000 목표. 예산 ₹10,372cr($1.24B) 5년 승인이나 집행은 4% 수준(발표→ 진행). Tier 1 무제한 접근 우선

South Korea
K-Nvidia / National AI Computing
building
13KGPU
$0.99B
Tier 1

1.46조원($989M)로 13,000 GPU 조달(Naver/NHN/Kakao 3사), 1만+ 배포 중(2026.2~). 2028년 52,000 → 2030년 260,000 목표. 2026 "계획→실행" 전환

Subsidy
Canada
Sovereign AI Compute Strategy
building
$1.42B
Tier 1

$2B CAD($1.42B) 5년(Budget 2024). 공공 슈퍼컴 $1B + 민간 $700M. Cohere에 $240M 첫 집행(2024.12). SCIP 프로그램 가동

France
France 2030 / Mistral
building
18KGPU
$118B
Tier 1

€109B($118B) 동원(2025.2, France2030+민간+국제 이니셔). Mistral Compute 18,000 GB200(Essonne 40MW). ★Mistral 최대주주 ASML(11%) → 유럽 주권AI↔유럽 장비 연결. OVHcloud/Scaleway 주권 클라우드

Germany
EU InvestAI Gigafactory
announced
$21.5B
Tier 1

EU InvestAI €200B 동원/€20B 공공으로 4~5개 기가팩토리(각 ~100,000 칩, 기존 AI팩토리 4배). 77개 제안·16개국·60개 부지. 2027 1호 착공. ($21.5B≈€20B)

Japan
GENIAC
building
$10B
Tier 1

GENIAC 2024.2~ 운영(3차 24개 프로젝트, 2025.7). GPU 사용료 보조 구조. FY2026 AI/반도체 ¥1.5조($10B). "하이퍼스케일러 주도→국가규모 인프라로 전환"(IDC)

Context (unverified — next update)

USPrivate-led 28GW · No state procurement
ChinaSelf-reliance · Export-controlled · Not covered
UAE·France·Canada·Singapore — pending verification

Common pattern: SWF/budget + cloud + Nvidia GPU + data sovereignty clause. States now "buy" chips too

Is stress building in the supply chain?

Supply Chain Tension Index

77
Elevated

Measures supply chain stress via customer-supplier temperature gap and price-volume divergence.

0 Loose50 Normal100 Tension
T1: Customer-Supplier Gap
61
HBMNVIDIA +5.6% vs 3-Co DSI +3.4d
61
FoundryTSMC +44.7% YoY vs Equip -1.1%
67
MemoryETF -0.2% vs DSI +3.4d
54
T2: Price-Volume Divergence
94
DRAM Spot (2wk)+2.66%
Export Volume (MoM)-14.15%
⚠ Price↑ + Volume↓ = Supply constraint

Are spot prices diverging from exports?

Experimental

DRAM Price vs Export Divergence

Aligned

When DRAM spot price movement diverges from export value movement, it signals a supply-demand imbalance.

Price and volume moving together — supply-demand balanced

DRAM Spot (DDR5 16Gb)
$54.10
MoM 6.1%
Semi Export (12)
$10.7B
MoM 4.0%
Divergence
+2.1pp
Price MoM - Export MoM

Does market narrative match the data?

Narrative vs Fact Divergence

-14.5
Aligned

Gap between news sentiment (positive ratio) and actual export/DRAM data. Excessive optimism vs reality signals overheating.

Sentiment
66.0
Fact
80.5
Divergence
-14.5
Sentiment vs Fact by Country
🇺🇸 US(1123)
60%sent49fact
+10.4 Aligned
🇰🇷 Korea(880)
62%sent100fact
-38.4 Fact Ahead
🇹🇼 Taiwan(1151)
76%sent90fact
-14.1 Aligned
🇯🇵 Japan(361)
59%sent81fact
-21.5 Aligned
Sentiment by Segment
HBM(247)
73%sent91fact
-17.4 Aligned
DRAM/NAND(768)
66%sent63fact
+2.3 Aligned
AI Server(312)
80%sent81fact
-0.4 Aligned

※ Fact axis is a quarterly/filing-updated metric. Segment fact (esp. HBM) = 3-company filing HBM-mention trend + SK Hynix inventory-days (DSI); it moves only when quarterly/half-year filings land — not daily. Sentiment axis updates daily.

Where are analysts pointing?

Experimental

EPS Revision Direction

Neutral

Direction of analyst EPS estimate revisions. Rising upgrade ratio = improving earnings outlook; falling = downgrade cycle.

• 7d upgrade ratio > 30d ratio → accelerating (recent revisions more positive)

• 7d upgrade ratio < 30d ratio → decelerating (recent revisions more negative)

7d Upgrade Ratio
77%
upgrades / total
30d Upgrade Ratio
77%
baseline
Direction
-1pp
7d - 30d
Based on 24 US/NL semiconductor stocks · 602 revisions in last 30 days

Is semiconductors leading the market?

Market context

SOX vs S&P 500 Relative Strength

Market Leading

Shows whether semiconductors are outperforming or underperforming the broader market. Rising ratio = semis outperform.

SOX 1M
-0.7%
S&P 1M
+3.5%
Relative 1M
-4.1%
Relative 3M
-6.3%

How hot is each value chain segment?

Experimental

Semiconductor Supply Chain Momentum

Stable

Shows how hot each semiconductor value chain segment is relative to its own history. Each is an independent thermometer, not a cross-segment comparison.

• 50 = neutral vs own history. 65+ = Hot, 35- = Cold.

• KR 89 > TW 41 means "KR momentum is hotter," not that KR is bigger than TW in absolute terms.

• Score = price momentum + earnings/export YoY. Weights and axes vary by segment.

• Materials score updates less frequently (quarterly earnings).

🇯🇵 JP Materials
24.4
Cold
β
2/5 TDnet · mixed
Wafer Volume(🇯🇵→🇰🇷)
+75.6%YoY kg
HS3818 · UN Comtrade · 2026-03
2-3 mo lag · Aux signal
Accumulating (Jul-Aug fix)
🇯🇵 JP Equipment
60.0
Warm
🇹🇼 TW Foundry
68.6
Hot
🇰🇷 KR Memory
63.6
Warm

Which supply chain node is hottest?

Experimental

Accumulating sentiment data...

What signals show up in hiring first?

Experimental
Loading...

This data is for informational purposes only, not investment advice. Correlation does not imply causation.