AI Hot Radar¶
Core of This Page
Currently active 7 AI investment themes — each includes: 1-sentence description + beneficiary ticker + monitoring signals. Monthly refresh.
Theme Definition
"Theme" = a cross-company, medium-to-long-term (6-24 months) investment thesis with specific catalysts. Not a single stock hype.
Current 7 Major Themes (Ranked by Heat + Execution Speed)¶
🔥 #1 Stargate Beneficiary Chain Expansion (2025-2028)¶
Description: $500B US AI infrastructure → UAE Phase 2 ($100B) → Saudi (HUMAIN $40B) → India / Brazil sovereign AI follow-up.
Beneficiary Tickers: - Direct: ORCL (Stargate primary cloud) · CRWV (Neocloud) · NBIS - Supporting: VRT · CEG · GEV · ETN · EQIX - Upstream: NVDA (GPU primary supplier)
Monitoring Signals: - UAE / Saudi sovereign AI progress (Bloomberg / Reuters announcements) - Sovereign AI announcements from various countries (India / Brazil / South Korea) - New power PPA signing pace
Related Patterns: #7 Sell Shovels > Gold Rush · #13 Energy is the Wildcard
🔥 #2 HBM4 Mass Production (2026 H2)¶
Description: Next-gen HBM, SK Hynix ramps early / Micron follows / Samsung lags. Historical HBM3e cycle repeats (P4-C2 case study).
Beneficiary Tickers: - Primary Beneficiaries: SK Hynix (000660.KS) · MU - Capped: NVDA (Rubin H2 2026 uses HBM4) - Second-Order: TSM CoWoS 2.5D packaging (TSM)
Monitoring Signals: - SK Hynix Q2/Q3 earnings HBM4 ramp timeline - Micron earnings HBM4 mass production update - Samsung qualify NVDA pass/fail (continued failure = lost market share) - NVDA Rubin shipment timing capped by HBM4
Related Patterns: #6 Micro Divergence ±150%
🔥 #3 Agentic AI Inference Demand Explosion (2025-2027)¶
Description: Regular ChatGPT ~1K tokens per query; agentic (Claude Code / Cursor / Devin) single session 50K-500K tokens = 100-500x inference compute. New growth curve.
Beneficiary Tickers: - Inference GPU: NVDA (Blackwell inference optimized) - Inference Cloud: CRWV · NBIS · MSFT Azure - Agentic Companies (Tier 1): Anthropic Claude Code · Cursor · Cognition Devin
Monitoring Signals: - Anthropic + OAI revenue growth rate (quarterly update) - Cursor / Devin / Cognition user growth (TechCrunch / The Information) - NVDA earnings inference compute breakdown
Related Patterns: #15 Agentic New Growth Curve
🔥 #4 US Nuclear Renaissance (2024-2030)¶
Description: AI data centers 1 GW each → US annual new electricity demand 80 GW vs current 10 GW → 8x gap. Nuclear / gas renaissance.
Beneficiary Tickers: - Nuclear: CEG (Three Mile Island restart) · VST · TLN (Talen, AWS PPA) - Gas: GEV (gas turbine backlog soaring) - Distribution: ETN · HUBB - REIT: EQIX · DLR
Monitoring Signals: - New PPA signings (MSFT / AWS / Stargate) - Trump Energy Order implementation progress - Nuclear plant approval / restart timeline
Related Patterns: #13 Energy Wildcard But Long Cycle
⚡ #5 China Domestic AI Substitution (2023-2030)¶
Description: H100 not sold to China → Huawei Ascend 910C / SMIC 7nm / YMTC HBM rising. China market vs US market segmentation.
Major Players (mostly private or illiquid): - Huawei Ascend 910C (private) - SMIC (HK listed, illiquid) - YMTC (private) - DeepSeek (private, algorithmic efficiency breakthrough) - Alibaba / ByteDance / Tencent (each with self-developed chips + applications)
Indirect Beneficiaries: - After domestic substitution, **NVDA loses 25% of China** (already priced in) - US friend-shoring beneficiary: TSM Arizona / Korea / Japan / EU
Monitoring Signals: - DeepSeek new version (V4 / R2) performance breakthrough - Huawei Ascend mass production progress - US export control expansion (e.g., 5nm equipment)
Related Patterns: #5 Regulatory Lag + Substitution Effect · #17 Dual-Track Development
⚡ #6 CPO (Co-Packaged Optics) Mass Production (2025-2027)¶
Description: Optical modules shift from pluggable to co-packaged with switch chips. Power consumption -50%, bandwidth +2x. But CPO yield is difficult, true mass production in 2026-2027.
Beneficiary Tickers: - Packaging: TSM (CoWoS-L / SoIC) - Switch: AVGO (Tomahawk + Jericho) - Optical: COHR · LITE (next gen) - New Challenger: Marvell
Monitoring Signals: - TSM CoWoS mass production + yield - AVGO CPO product launch (Tomahawk ⅚) - 800G → 1.6T optical module transition
Related Patterns: #13 Physical Bottlenecks
⚡ #7 Scaling Laws Data Wall + Synthetic Data Risk (2026+)¶
Description: Long-term wildcard. High-quality human text ~40T tokens, GPT-6 needs ~100T → data wall. Synthetic data (Meta / OAI trying) risks model collapse.
Impacted Tickers: - NVDA long-term (2027+) thesis adds new invalidation - OAI / Anthropic training ROI questioned - Data companies (Reddit / Stack Overflow / publishers) bargaining power ↑
Monitoring Signals: - Foundation model public benchmarks continue exponential improvement OR start plateauing - GPT-5 / Gemini 2 / Claude 5 actual capability vs training cost - Synthetic data papers (Meta / Anthropic / Google) model collapse evidence
Related Patterns: #9 Data Wall Risk
Relationship Between 7 Themes and Previous 5 Parts¶
| Theme | Uses Which Previous Part |
|---|---|
| Stargate Beneficiary Chain | P4-C3 Stargate Forensic |
| HBM4 Mass Production | P4-C2 Samsung HBM + P3-C3 Find Bottlenecks |
| Agentic | P1-C8 Application Layer + P1-C4 LLM Intuition |
| Nuclear Renaissance | P1-C5 Nuclear + P1-C9 Geopolitics |
| China Domestic Substitution | P4-C4 Export Controls + P1-C9 Geopolitics |
| CPO Mass Production | P1-C5 Hardware Stack |
| Data Wall Risk | P1-C2 Scaling Laws + P1-C1 AI Winter Base Rate |
Theme is Not a Buy Signal
Each theme gives you direction, not when to enter / exit. You still need: - Catalyst Calendar (timing) - Thesis YAML (size + invalidation) - Part 3 Process (research + sanity check)