Skip to content

P1-C9 · US-China + Export Controls + Energy Geopolitics

Core One-Liner

Geopolitics is the biggest wildcard in the AI industry — if your thesis doesn't account for geopolitics, it's missing a dimension.

AI Industry Knowledge — History → Technology → Industry Chain → Business → Application → Geopolitics

P1-C9 (Part 1, Chapter 9). After this chapter, you can add 3 geopolitical dimensions (US-China / Energy / Capital) to your thesis, and no longer be caught off guard by "unexpected geopolitical news."

1. The Problem: DeepSeek Surprise / H100 Can't Be Sold to China / Three Mile Island Nuclear Restart — These Are All One Geopolitical Story

Several major news events from 2025-2026 you've seen:

  • 2025/01 DeepSeek V3 + R1 trained near GPT-4 level using H800 — NVDA fell 17% in a single day / -$593B
  • 2022-present H100/H200 export controls — NVDA China/HK revenue ~13% FY2025 10-K ($17.1B / $130.5B total); Data Center China dropped from 19% (2023) to 14% (FY2024), continuing to decline. Source: NVDA 10-K FY25
  • 2024 MSFT Three Mile Island nuclear PPA for 20 years — Constellation Energy rose 100%+
  • 2025/01 Stargate $500B — backed by Trump, OAI + Oracle + MSFT + SoftBank
  • 2026 China's domestic chip rise — Huawei Ascend 910C / SMIC 7nm

These are not "AI company earnings" — they are all geopolitical events. If your thesis doesn't reserve a geopolitical dimension, once an event occurs, your entire thesis is upended.

2. The Solution: 3 Geopolitical Dimensions

Geopolitical Dimension Impact How Your Thesis Should Respond
US-China Tech War Export controls → China's domestic substitution → market fragmentation Any NVDA / TSM / ASML thesis must add "China exposure + substitution risk" red_flag
Energy Security AI data centers 1 GW/unit → nuclear / natural gas revival CEG / VST / GEV are indirect theses. Any Hyperscaler capex must include "energy PPA" information
Capital Controls CFIUS / China sovereign funds / Taiwan Strait TSM must add "Taiwan Strait geopolitical invalidation_trigger"

These 3 dimensions intertwine: DeepSeek breakthrough → China's AI self-sufficiency → US tightens controls → China further self-sufficient → US pushes friend-shoring → global AI supply chain fragmentation.

3. How It Works: Detailed Breakdown of the 3 Geopolitical Dimensions

3.1 US-China Tech War — Export Controls + Domestic Substitution

2022/10 Round 1: US bans A100 / H100 from being sold to China — NVDA loses the Chinese market. 2023/10 Round 2: Bans H800 (customized, neutered version) + AMD MI300 — further compression. 2024-2026 Round 3: Restrictions on investments in China affiliates + equipment (ASML has never sold EUV to Chinese customers per CNBC 2024; Dutch licensing now blocks EUV + certain advanced DUV (2100i / 2050i / 2000i / 1970i / 1980i) to China; ASML China revenue is primarily DUV / mature-node).

China's Response: - Huawei Ascend 910B/C — self-developed NPU; per DeepSeek research (Tom's Hardware 2025/02) 910C reaches ~60% of H100 inference performance (training still lags), not officially benchmarked; process still trails at 7nm/5nm - **SMIC 7nm/5nm — per Semiecosystem / TD Cowen 2025 estimate 7nm yield has ramped to ~60-70% (from <40% at 2H23 launch), 5nm still pilot (<20%); still materially below TSM leading-edge but ramping - YMTC / CXMT — domestic HBM, mass production 2026+ - DeepSeek V3** — trained near GPT-4 level using H800 + algorithmic efficiency, proving "banning chips ≠ banning capability"

Investment Implications: - NVDA: China/HK is ~13% FY2025 ($17.1B / $130.5B per 10-K); Data Center China dropped 19%→14% from 2023→2024 (continuing decline). Other markets compensate. Long-term impact: Chinese market is invisible to NVDA - TSM: Add Taiwan Strait trigger (low probability but catastrophic). Any Taiwan Strait event = global AI chain halted for 6-12 months - ASML: Never sold EUV to Chinese customers (CNBC 2024). China revenue primarily DUV / mature-node; export restrictions affect EUV (continuously banned) + certain advanced DUV. Still a monopoly, not fatal. - China domestic substitution stocks: Primary market (Huawei / SMIC public, but low liquidity)

3.2 Energy Security — Nuclear + Natural Gas Revival

Fact: Some Stargate-scale clusters reach 1GW (e.g., Stargate UAE 1GW Abu Dhabi cluster, Phase 1 200MW online Q3 2026; Stargate Abilene / other site capacities not officially disclosed). US power demand: NERC 2026 summer assessment estimates aggregated peak demand grows >11GW vs 2025 (annual additions in the ~10-11GW range); BNEF estimates data-center power demand from ~35GW in 2024 → ~78GW in 2035 (10-year cumulative). (Note: "+80GW added in 2026" conflates BNEF's long-term cumulative figure with NERC's annual addition — these are two different metrics that should be kept separate.)

→ Electricity is the real bottleneck for 2026+ (you can buy GPUs, but you can't buy electricity).

Response (US + major jurisdictions): - **MSFT-CEG Three Mile Island Restart (2024) — 20-year PPA, nuclear - **MSFT-Vistra Texas — natural gas + solar - AWS-Talen Energy — nuclear PPA - Stargate Abilene Texas — natural gas + solar + storage hybrid - Trump Energy Order (2025) — accelerate nuclear approvals, drill baby drill

Investment Implications: - CEG / VST / TLN: Nuclear companies. Rose 100-300% in 2024. Continued capex catalyst. - GEV (GE Vernova): Gas turbines. Backlog soaring, awaiting realization in 2026+ - VRT / ETN / HUBB: Liquid cooling + electrical. Direct capex beneficiaries. - EQIX / DLR: Data center REITs. Priority for land acquisition + power PPAs.

Risk: Power construction cycles are long (nuclear 10 years, natural gas 3-5 years). Once AI demand plateaus, power companies face stranded assets (built but unused).

3.3 Capital Controls — CFIUS + Taiwan Strait + Sovereign Funds

CFIUS (Committee on Foreign Investment in the United States): - Since 2018+, tightened scrutiny of Chinese investments in US AI / semiconductors - TikTok case / Grindr case are typical examples

Taiwan Strait: - TSM is in Taiwan, 90% of global high-end chips depend on TSM - TSM is already building fabs in Arizona / Japan / Germany ("Fab 21"), but mass production is slow - Any Taiwan Strait event = global AI chain halted for 6-12 months (no overstatement)

Sovereign Funds: - UAE / Saudi heavily investing in OpenAI / Anthropic / xAI - Stargate UAE: 1GW Abu Dhabi cluster, Phase 1 200MW online 2026 (per OpenAI 2025/05/22; G42 / Oracle / NVDA / Cisco / SoftBank; no $100B UAE figure exists — $100B refers to US Stargate "immediate deployment", NOT UAE) - China sovereign funds vs US friend-shoring tug-of-war

Investment Implications: - TSM thesis must include "Taiwan Strait" red_flag trigger - Monitor OAI / Anthropic sovereign funding rounds (valuation anchor + geopolitical signal) - US CHIPS Act + Trump manufacturing reshoring → INTC / Texas Instruments / GlobalFoundries benefit

4. vs C8 What You Already Know

Dimension C8 Gives You C9 Adds for You
Application revenue ✓ Doesn't explain geopolitical impact
Geopolitical wildcard ✗ 3 dimensions + historical cases
Investment implications Knows which layer ROI is realized Knows application realization is also reshaped by geopolitics — DeepSeek-like breakthroughs can change NVDA by 17% in one day

C8 = internal economics. C9 = external wildcard. Without C9, your thesis is all micro, missing macro.

5. Try It: Assess the Impact of a DeepSeek-like Event on Your Thesis

Task (15 minutes): Assume tomorrow another "China breakthrough" news — DeepSeek V4 trained a reasoning model surpassing GPT-4 using domestic Ascend 910C. Answer:

Ticker Short-term Impact (1 week) Long-term Impact (1 year)
NVDA -? % ?
SK Hynix -? % ?
TSM -? % ?
OpenAI (private) Valuation -? ?
MSFT -? % ?

Hint: Short-term look at sentiment (how much panic is priced in), long-term look at real demand changes (Jevons Paradox — efficiency gains lead to increased demand).

Self-check (3 items checked → proceed to P1-C10):

  • You can add 1 Taiwan Strait invalidation_trigger + 1 China export control expansion red_flag to your NVDA thesis
  • You can explain why CEG / VST soared in 2024+ (energy bottleneck + long-term PPAs)
  • You can list 1 geopolitical event → your thesis add signal (e.g., Trump cutting CHIPS Act → INTC bearish trigger triggered)

6. What's Next

The first 9 chapters are all framework. Now apply them to 5 real cases — use the tools from the 9 chapters on actual events.

→ P1-C10 · 5 Real Cases to Build Intuition — The final chapter of Part 1.

7. Deep Dive (optional): Project Stargate UAE / Trump Energy Order / Sand-to-Silicon Stack

Click to expand 3 deep geopolitical cases

Project Stargate UAE (announced 2025/05/22): G42 / Oracle / NVDA / Cisco / SoftBank / OpenAI partner to build 1 GW data center in Abu Dhabi (UAE–US AI Campus, total 5 GW). Phase 1 200 MW online 2026 (per G42 update tracking Q3 2026 delivery). No $100B UAE figure was disclosed by OpenAI — the $100B number refers to US Stargate "immediate deployment", not UAE. G42 has no OpenAI equity in this project (per official announcement). → This is a new model of sovereign AI: national-level computing sovereignty. Saudi (HUMAIN $40B) / India / Brazil are following. → Investment implications: NVDA + OpenAI benefit. But geopolitics: UAE's close ties with Iran → conflict with US friend-shoring.

Trump Energy Order (2025/01): Accelerate nuclear approvals + drill baby drill + revoke some IRA renewable energy subsidies. → Short-term: Nuclear companies (CEG / VST) accelerate realization. Gas companies (TLN) benefit. → Long-term: Bottleneck for US AI data center expansion eases.

Sand-to-Silicon Stack — US friend-shoring: From silicon sand (Sumco / SiTime) → wafers (TSM Arizona) → equipment (ASML must export to friendly countries) → design (NVDA) → systems (Dell / HPE / SMCI assembled in US) → cloud (MSFT/GOOGL/AMZN). Trump pushes entire chain localization in the US (CHIPS Act + ETC). Actual progress: TSM Arizona Fab 21 Phase 1 mass production, Phase 2-3 2027-2028. → Investment implications: Long-term INTC + TSM Arizona + US domestic equipment companies (AMAT / LRCX) benefit. But cost: US manufacturing is 30-50% more expensive than Taiwan, customers complain.

8. Further Reading (this chapter — US-China + export controls + energy)

All free sources, aligned with P5 0-paid policy

Classic papers / primary government text:

Wikipedia (3-10 min):

Think-tank reports (free):

Videos / public lectures:

Podcasts:

  • Acquired — TSMC — Geopolitical risk section
  • ChinaTalk (Jordan Schneider) — Weekly podcast on US-China tech rivalry
  • The China in Africa Podcast — Geopolitical investment perspective

News sources (free):

Books (library):

  • Chris Miller "Chip War" (2022) — Must-read, 30 years of US-China semiconductor geopolitics
  • Dan Wang annual letters (free at danwang.co) — Insider view of Chinese tech industry

Pair with this chapter's self-check:

After "Chip War" + 1 BIS rule original + 1 CSIS report + DeepSeek Wikipedia, you should be able to answer "the 3 geopolitical lines" and "estimate DeepSeek-like event impact on your thesis."