📌 Information-only market analysis · This article is based on publicly available data and industry reports. It does not recommend buying or selling any specific security. All investment decisions are your own responsibility.
Strategic Intelligence Report

AI Data Center Power Showdown
10-Year Supply Gap + Real Bottleneck

Precise analysis of AI data center electricity supply and demand across the United States, China, and Korea over the next decade (2026~2035). Among the five candidate bottlenecks (power, HBM, GPU, SSD, capital), which one is the real constraint — and where do the asymmetric investment opportunities created by that bottleneck lie?

📅 2026.05
🌐 🇺🇸 🇨🇳 🇰🇷 3 Countries
⏱️ 22 min read
📊 Multi-source
⭐ Executive Summary · Read this first

Bottom Line and Top 10 Investment Ideas

🎯 One-line conclusion: The real bottleneck of the AI era is not GPUs — it is electricity. HBM is the next bottleneck. GPU and SSD constraints are already loosening, and capital is not a bottleneck for hyperscalers. Therefore, the two axes of power infrastructure + HBM supply chain will deliver the most asymmetric returns over the next decade.

🌐 Three-country one-line comparison

🇺🇸 United States — "Demand surge vs 5-year grid lag"
2030 DC power demand ~426 TWh (+133% vs 2024). Roughly 80 GW/year of new generation needed, but actual additions are ~40 GW. PJM·MISO·ERCOT face capacity-shortfall warnings by 2028. Transmission permitting takes 5–10 years. Bottleneck = power infrastructure.
🇨🇳 China — "State-coordinated + self-reliance pressure"
East-Data-West-Computing 8 hubs, 1 trillion yuan AI investment. Government allocates power centrally, so grid bottleneck is smaller than the US. NVIDIA H20 export controls are accelerating domestic GPUs (Huawei Ascend, etc.). Bottleneck = GPU self-reliance + memory.
🇰🇷 Korea — "HBM upside + RE100 risk"
DC capacity 1.96 GW → 6.32 GW by 2030 (3.2×). 76 new AI DCs planned. SK Hynix and Samsung supply 80%+ of global HBM. But renewable-energy share at ~10% means RE100 commitments unmet. Bottleneck = transmission + clean power.

💎 Top 10 Investment Ideas (by conviction strength)

* This ranking is not external; it reflects this article's synthesis of the five-bottleneck analysis below. Not a buy recommendation. Time horizon, risk tolerance, and portfolio fit depend on your situation.

01
SK Hynix KRX:000660
Global HBM #1 (~50% share). Core supplier of HBM3E and HBM4 to NVIDIA and AMD. The deepest moat in AI memory over the next decade. The most direct AI play.
02
Constellation Energy NASDAQ:CEG
Largest US nuclear operator (~94 TWh/yr). Microsoft Three Mile Island restart contract. 24/7 clean power + direct-to-AI-DC PPAs = explosive pricing power.
03
Hanmi Semiconductor KRX:042700
Effective monopoly on HBM TC bonders. Core supplier to SK Hynix and Micron. The narrowest gate of the HBM super-cycle = the sharpest pick.
04
Vistra Corp NYSE:VST
Top-tier US IPP (Top 2). Gas + nuclear + battery portfolio. Up ~260% in 2024 yet P/E still reasonable.
05
TSMC NYSE:TSM
CoWoS packaging effective monopoly. NVIDIA·AMD·Apple AI silicon all packaged here. Capacity 30K → ~75K wafers/month by end-2025. The true gatekeeper of every AI chip.
06
Eaton NYSE:ETN
Global leader in DC electrical infrastructure (transformers, UPS, switchgear). AI DC electrical content runs 3–5× a traditional DC. 5+ year revenue visibility.
07
Vertiv NYSE:VRT
DC cooling + power systems. AI chip thermals are exploding, driving immersion and direct-to-chip cooling demand. NVIDIA reference-design partner.
08
Doosan Enerbility KRX:034020
Top-5 global SMR EPC. First-wave beneficiary as SMRs scale in the late 2030s. High short-term volatility — long horizon required.
09
HD Hyundai Electric KRX:267260
Major Korean transformer exporter to the US. US grid transformer lead times 3–4 years. Korean transmission super-cycle adds a second tailwind.
10
GE Vernova NYSE:GEV
Global gas turbine #1. Surging demand for behind-the-meter gas power for AI DCs. Wind portfolio adds optionality.
🔑 Key message: The real gate of AI data center build-out is not GPUs. Power (transmission permitting 5–10 years, transformer lead times 3–4 years) is the hardest bottleneck. HBM (capped by CoWoS) is the short-term 1–2 year gate. GPU and SSD are already loosening. Capital is a non-bottleneck for hyperscalers ($315B+ combined CapEx in 2025). The most asymmetric returns therefore concentrate on (1) power infrastructure and (2) the HBM supply chain. The full three-country 10-year gap data, five-bottleneck comparison, and three-tier investment framework follow below.

Big Tech AI Data Center CapEx Explosion

Big Tech AI infrastructure CapEx began its full-throttle climb in 2024. In 2025 alone, the four largest US cloud companies are deploying roughly $315B into data centers and AI infrastructure — a record. That figure is equivalent to about 18% of South Korea's annual GDP.

Microsoft 2025 CapEx
$80B+
▲ ~80% AI infrastructure
Google (Alphabet) 2025
$75B
▲ DC + custom AI silicon
Amazon (AWS) 2025
$100B+
▲ Record IT spend
Meta 2025
$60–65B
▲ Llama + Reality Labs
Stargate Project
~$500B
/ 4-year ceiling · $100B initial (OpenAI·Oracle·SoftBank)
xAI Colossus
200K
GPU target (2026)

2026~2030 cumulative global AI DC investment

Estimates based on Goldman Sachs / McKinsey / Synergy Research consensus:

2025
$425B (global)
2026
$560B
2027
$720B
2028
$870B
2029
$990B
2030
$1.1T+
Key: Cumulative 2025–2030 global AI DC infrastructure spend approaches $5T. Roughly 30–40% goes to power and cooling, 40–50% to GPUs/HBM/SSD, and 10–20% to real estate and construction. The thesis that "the gate is power, not chips" is also reflected in capital allocation.

United States — 10-Year Power Supply Gap (2026~2035)

The US faces the deepest demand-vs-supply gap. Combining IEA, Pew Research, EIA, and ICF data, US data center electricity consumption rises from 183 TWh in 2024 to ~426 TWh in 2030 and ~730+ TWh by 2035. New generation additions, however, run at less than half the pace of demand growth.

📊 US year-by-year DC power demand vs supply (TWh)

YearDC demandNew supplyCum. supplyCum. demandGapRisk
2026230524,5204,540-20Manageable
2027275564,5764,608-32Warning
2028330584,6344,706-72PJM crisis
2029380624,6964,818-122Rate spikes
2030426684,7644,944-180Structural shortfall
2031490724,8365,090-254SMR online begins
2032555784,9145,255-341Distributed accel.
2033620854,9995,440-441
2034680925,0915,640-549
2035730985,1895,855-666Full distributed

* 2026 and 2030 figures use IEA / Pew / EIA base-case scenarios. Intermediate years (2027–2029, 2031–2034) are this article's interpolations. Forecasting institutions can vary by ±15%.

The shape of the structural gap: The US needs roughly 80 GW of new generation per year, but the actual run-rate over the past five years is closer to 40 GW. Transmission permitting takes 5–10 years on average, and large transformer lead times are 3–4 years. These physical constraints cannot be cleared with capital alone in the short run. McKinsey warns that PJM (Eastern US), MISO (Midwest), and ERCOT (Texas) are likely to face material capacity shortfalls beginning around 2028.

🏭 Big Tech US data center clusters

Northern Virginia (Data Center Alley)
26% of state power = DCs

Loudoun County in Northern Virginia carries an estimated 70% of global internet traffic. As of 2024, data centers consume about 26% of state-wide electricity, and Dominion Energy is quoting 4–7 year wait times for new DC interconnections. An additional ~16 GW of demand is expected by 2030.

Texas (ERCOT)
+8 GW DC demand expected

Cheap power and faster permitting have made Texas a hotspot for new data centers, although the 2021 cold snap left lingering concerns about grid resilience. xAI Colossus (near Memphis) and Stargate Phase 1 (Abilene) are headline cases. The energy mix is gas + wind + solar + batteries.

Arizona / Nevada / Iowa
Emerging mega-clusters

Phoenix anchors a Microsoft / Meta / Apple cluster. Reno (Nevada) hosts Google and Tesla. Iowa is Microsoft's AI supercomputing hub. The decision driver is consistent: cheap power + cool climate + state incentives.

⚡ US generation mix change (2026 vs 2035)

Gas '26
42% (1,900 TWh)
Gas '35
28% (1,640 TWh)
Solar '26
8%
Solar '35
30% (1,755 TWh)
Nuclear '26
18%
Nuclear '35
22% (incl. SMR)
Wind '26
11%
Wind '35
14%

Key shifts: Gas generation rises in absolute terms but loses share. Solar climbs explosively (8% → 30%). SMR comes online materially in the late 2030s, edging nuclear share higher. Coal nearly retires by 2035 (10% → ~2%).

China — 10-Year Power Supply Gap (2026~2035)

China is playing a different game than the US. With state-controlled power allocation, grid bottlenecks bite less than in the US's market-based system. At the same time, NVIDIA H20 / B200 export controls impose strong domestic-substitution pressure on chips and memory. The "East-Data-West-Computing" (东数西算) initiative is the centerpiece.

🌏 East-Data-West-Computing 8 hubs

HubRegionUsePower edge
Yangtze DeltaShanghai · Jiangsu · ZhejiangReal-time AI inferenceGas + nuclear
Greater Bay AreaGuangdong · GuangxiFinance · real-timeNuclear + hydro
Jing-Jin-JiBeijing · Tianjin · HebeiGovernment / finance AINuclear + wind
Chengdu-ChongqingSichuan · ChongqingTraining + inferenceAbundant hydro
Inner MongoliaHohhot · UlanqabLarge-scale trainingWind + solar
GuizhouGuiyangLarge-scale trainingHydro + solar
GansuQingyang · YinchuanLarge-scale trainingSolar + wind
NingxiaYinchuanLarge-scale trainingSolar
Strategy core: Build small inference DCs in the east (where data is generated — Shanghai, Beijing) and large training DCs in the west (where renewables are abundant — Inner Mongolia, Guizhou). This division of labor distributes grid load and exploits renewables at the same time.

📊 China year-by-year DC power demand (TWh)

YearDC demandYoYRenewables shareKey issue
2026240+22%32%H20 export controls bite
2027285+19%35%Huawei Ascend volume ramp
2028335+18%38%Domestic HBM (CXMT etc.) starts
2029390+16%42%
2030440+13%45%Closing on 50% renewables target
2031510+16%48%
2032580+14%52%
2033650+12%55%SMRs come online
2034720+11%58%
2035790+10%60%Post-carbon-peak effects

* China DC demand combines unofficial government statistics with private estimates (IDC China · Frost & Sullivan · Synergy). Source data is thinner than for the US, and ranges are wider.

China Big Tech AI infrastructure

Alibaba Cloud (阿里云)
¥380B / 3 years

In February 2025, Alibaba committed ¥380B (~$53B) to AI and cloud over three years. Footprint in all eight East-Data-West-Computing hubs. Self-developed Qwen LLM family. Targeting 10%+ global cloud share by 2030.

Tencent / ByteDance / Baidu
¥250B+ combined

Tencent ¥100B+, ByteDance (TikTok parent) ¥80B in AI infrastructure. Baidu runs Apollo + Ernie. All three are building proprietary DCs across the East-West clusters. China's own AI CapEx supercycle is underway.

DeepSeek efficiency revolution
Power demand -30~50%?

In 2025, DeepSeek-V3 and R1 claimed comparable performance to leading US frontier models at around 1/10 the cost (DeepSeek's own claim, still under academic and industry verification). If DeepSeek-style efficiency gains generalize, the demand projections above could move down meaningfully. The counter-scenario, though, is "Jevons Paradox" — efficiency gains expanding total usage rather than shrinking it.

China-specific risk: Tightening US export controls limit access to NVIDIA H20 / B200. Domestic alternatives (Huawei Ascend 910C, SMIC 7nm) are gaining share but trail by roughly two to three generations. In the short term, GPU shortage may be a more acute bottleneck than in the US.

Korea — 10-Year Power Supply Gap (2026~2035)

Korea supplies 80%+ of the world's HBM but its own data center build-out is constrained by a triple bind: RE100 commitments, transmission shortages, and renewable lag. The 11th Basic Plan for Long-term Electricity Supply and Demand (2024–2038) defines the next decade.

📊 Korea year-by-year DC capacity + power

YearDC capacity (MW)DC power (TWh)Renewables shareRE100 gap
20262,40017.511%Large
20273,10022.613%Large
20284,50032.916%Very large
20295,40039.419%Very large
20306,32046.121.6%Improving
20317,30053.324%Easing
20328,40061.327%Easing
20339,60070.129%Easing
203410,90079.631%Stable
203512,30089.833%Stable

* 2025 and 2030 figures derive from the 11th Basic Electricity Plan. 2031~2035 extrapolates the same trajectory; renewables share interpolates toward the plan's 32.9% (2038) target.

Korea's triple bind: (1) Transmission bottleneck — solar-rich Honam (south) cannot deliver to demand-heavy Capital Region (Seoul/Gyeonggi) due to transmission shortages, leading to 30%+ curtailment of southern solar output. New transmission lines take 7–10 years to permit. (2) RE100 shortfall — only 22% renewable share by 2030 (vs ~30% world average, ~33% OECD). Samsung and SK Hynix's global RE100 commitments are at risk. (3) SMR dependence — the 11th plan leans on LNG and SMR. But the first SMR is not expected until ~2035, leaving the gap window open longer.

🏢 Korean DC clusters

Pyeongtaek-Yongin Semiconductor Cluster
+10 GW additional load

Samsung Electronics Pyeongtaek campus (P1–P5) plus SK Hynix Yongin Semiconductor Cluster. Roughly 10 GW of additional electricity load required by 2030. KEPCO's transmission-line construction from East Coast nuclear and LNG plants is in progress, but resident opposition is causing 7+ year delays.

Hyperscale AI DCs (NAVER · KT · LG)
76 new DCs by 2028

NAVER Gak (Chuncheon · Sejong), KT IDC, LG U+, Kakao, Kakao Enterprise. 76 new dedicated AI DCs are planned by 2028, averaging 50–100 MW each. Roughly half are expected to slip on schedule due to power infrastructure shortages.

Foreign Hyperscalers
AWS · MS · Google ramp

AWS Seoul Region adding a 4th AZ; Microsoft Korea Central; Google Cloud Seoul. Korea's location offers optimal latency between Japan and China. All three carry RE100 commitments, making Korea's renewable shortage a direct business risk.

11th Basic Electricity Plan highlights

2024–2030
LNG + solar acceleration
~10 GW new LNG generation, +25 GW solar. But transmission bottlenecks cap effective capacity factors. New Honam → Capital HVDC under planning.
2030–2035
First SMR + offshore wind ramp
Doosan Enerbility / NuScale-based SMR targeting first commercial operation 2034–2035. Southwest offshore wind 6 GW.
2035–2038
SMR scale-out + 33% renewables
4 SMR units cumulative; offshore wind to 12 GW. Renewables 32.9%. AI DC capacity 12.3 GW operating in steady state.

Five-Bottleneck Comparison — What Is the Real Constraint?

We score the five candidate bottlenecks of AI data center build-out on a common scale. Higher score = harder bottleneck (more difficult to remove).

⚖️ Five-bottleneck composite scoring matrix

Bottleneck20262027202820302035Structural
⚡ Power (grid)9595928570Very high
💾 HBM / CoWoS8882726040Medium
🎮 GPU silicon6855403025Low
💿 SSD / NAND3538424542Very low
💵 Capital1518222835Very low

* These scores are not from an external source — they are this article's composite judgment, weighted across (1) physical constraint, (2) time constraint, (3) capital-solvability. Higher = harder to clear.

📊 2026 bottleneck intensity visualized

⚡ Power
95 / 100 (CRITICAL)
💾 HBM
88 / 100 (HIGH)
🎮 GPU
68 / 100 (MED)
💿 SSD
35 / 100 (LOW)
💵 Capital
15 / 100 (NONE)

🔍 Detailed diagnosis

⚡ #1: Power (grid)
CRITICAL · Structural
Physical constraint
95
Time constraint
92
Capital-solvability
25

Why this is the hardest: (1) New transmission lines 5–10 years (resident opposition, environmental review). (2) Large transformer global lead times 3–4 years. (3) New generation permitting 2–5 years + construction 2–5 years. Capital alone cannot solve it. Workarounds: behind-the-meter gas/SMR, direct PPAs, distributed microgrids.

💾 #2: HBM / CoWoS packaging
HIGH · 1–2 year horizon
Physical constraint
75
Time constraint
78
Capital-solvability
55

Diagnosis: NVIDIA H100/H200/B200 and AMD MI300X all consume HBM3E. The true gatekeeper is CoWoS packaging at TSMC: capacity climbs from 30K wafers/month (2024) to ~75K by end-2025. SK Hynix ~50%, Samsung ~42%, Micron ~8%. Capacity normalization expected 2027–2028.

🎮 #3: GPU silicon
MED · Loosening
Physical constraint
58
Time constraint
62
Capital-solvability
75

Diagnosis: H100 lead times of 12+ months in 2023–24 have normalized to 3–6 months in 2025. NVIDIA Blackwell B200/B300 ramp continues. AMD MI300X / MI350, Intel Gaudi 3, and hyperscaler ASICs (Trainium · TPU · MTIA) diversify the supply.

💿 #4: SSD / NAND
LOW · Temporary tightness

AI training datasets and checkpoints lift NAND demand ~45% YoY, but SK Hynix (incl. Solidigm), Samsung, Micron, and SanDisk (WDC) all supply. Temporary tightness in 2026 followed by normalization in 2027. Much weaker than HBM.

💵 #5: Capital
NONE · Non-bottleneck for hyperscalers

Microsoft + Google + Amazon + Meta combined CapEx ~$315B in 2025 alone — record. Combined cash positions ~$400B. The scenario "AI infrastructure runs out of money" is unrealistic. The risk is an AI revenue disappointment around 2027–2028 slowing CapEx, not capital scarcity per se.

🎯 Composite conclusion: Real bottlenecks rank as Power → HBM → GPU → SSD → Capital. The most asymmetric returns line up in the same order: (1) power infrastructure (nuclear · gas · transmission · transformers), (2) HBM supply chain (SK Hynix · Samsung · Hanmi · TSMC), (3) GPU (NVIDIA · AMD). SSD and capital are weak direct plays.

Power Infrastructure — The Hardest Bottleneck

The power bottleneck is fourfold: generation, transmission, conversion (transformers), and last-mile distribution. Any one of these breaking blocks the entire data center. The lead times below show the real depth of the constraint.

🕐 Four-tier lead times

New gas generation
2–5y
1–2 permit + 1–3 build
New nuclear (large)
8–12y
Effectively impossible
SMR (small modular)
5–7y
Late-2030s ramp
Transmission (HVDC)
5–10y
Resident pushback
Large transformers
3–4y
Global capacity short
Solar + battery
1.5–2y
Fastest (intermittent)

🔌 Big Tech direct generation deals

Microsoft × Constellation Energy
20-yr PPA · ~$16B est. value

September 2024: agreement to restart Three Mile Island Unit 1. Microsoft buys all 835 MW for 20 years. Historic — the first time Big Tech has revived a closed reactor. Restart targeted for 2028.

Amazon × Talen Energy
Susquehanna co-located DC

AWS data center co-located with Talen's Susquehanna nuclear plant in Pennsylvania. ~$650M site purchase, up to 960 MW direct PPA. Behind-the-meter design. Note that FERC rejected PJM's revised ISA in Nov 2024; partial approval and refiling have followed. The first regulatory test of behind-the-meter at scale.

Google × Kairos Power (SMR)
500 MW SMR · 6–7 units

First SMR online targeted for the early 2030s. Amazon also invests in X-energy SMR; Microsoft signs Helion (fusion). Big Tech is becoming the new "anchor tenant" of future power.

🇰🇷 Korea's transmission crisis

The Honam-Capital transmission shortfall: 30%+ of Honam's solar generation is curtailed for lack of transmission. New East-Coast-nuclear → Capital lines face 7+ year delays from compensation and environmental review. "There is enough generation, but it cannot be delivered" — a structural inefficiency. New HVDC (high-voltage DC transmission) is the key remedy but won't operate before 2030.

HBM / CoWoS — The Narrowest 1–2 Year Gate

NVIDIA H100 / H200 / B200, AMD MI300X / MI350, Google TPU v6, Amazon Trainium 2 — every modern AI accelerator depends on HBM. And every HBM stack must pass through TSMC's CoWoS packaging to be assembled with its GPU. These two stages are the real gate.

📊 Global HBM share (2024 actual → 2026 estimate)

SK Hynix
~50%
Samsung
~42%
Micron
~8%

Korea = ~92%. That share is the real moat of Korean memory. (*Average of 2024 market-research-firm estimates. Quarterly variation possible.)

🔬 CoWoS — the real gatekeeper

2024 CoWoS capacity
30K
wafers / month
End-2025 capacity
75K
+150% YoY
2027 target
150K+
5× expansion
Alternative capacity
~0%
Samsung·Intel still entering

HBM bottleneck eases only as CoWoS capacity unlocks in 2027–2028. Until then, NVIDIA / AMD GPU shipments are effectively rate-limited by CoWoS.

🇰🇷 Hanmi Semiconductor — the sharpest pick

Hanmi Semiconductor (KRX:042700)
HBM TC bonder effective monopoly

HBM's critical "TC (Thermocompression) Bonding" tool ships almost exclusively from Hanmi to SK Hynix. Micron has also adopted Hanmi tools. Samsung is developing its own. The narrowest alley of the HBM supercycle = the sharpest pick.

📈 Investment thesis

★★★★☆
SHARP CONVICTION

Sharp re-rating in 2024–2025, then consolidation. HBM3E → HBM4 transition could re-accelerate revenue. Single-stock volatility is high — average in.

Investment Ideas — Three-Tier Portfolio

Translating the bottleneck analysis into a three-tier framework: Tier 1 = direct beneficiaries of the bottleneck, Tier 2 = picks-and-shovels infrastructure, Tier 3 = Korea-specific.

🥇 Tier 1 — Direct beneficiaries (Core)

SK Hynix (KRX:000660)
HBM #1 globally

Thesis: ~50%+ HBM share. Core supplier of NVIDIA H100 / H200 / B200. HBM3E 12-Hi in volume, HBM4 ramping in 2026. "The Intel of the AI era" position. Operating margins at multi-decade highs.

Risk: Samsung catch-up + Micron entry could trim share. Memory cyclicality. Valuation: 2026 P/E ~8–10× — discount to global memory peers.

Constellation Energy (NASDAQ:CEG)
US nuclear #1

Thesis: US nuclear leader (~94 TWh/yr). Microsoft 20-year PPA signed. AI DCs craving 24/7 clean power = explosive pricing power. New SMR construction takes 5–7 years, but existing reactors are immediately available.

Risk: Tail-risk of nuclear incident (low but asymmetric). Policy changes. Valuation: Up ~90% in 2024; P/E ~30× — reasonable given growth.

Hanmi Semiconductor (KRX:042700)
HBM TC bonder monopoly

Thesis: The narrowest, sharpest Tier 1 pick. Effective monopoly on the key HBM packaging tool. SK Hynix and Micron both run Hanmi tools. HBM build pace → Hanmi revenue, directly.

Risk: Customer concentration. Samsung in-house tool risk. Significant volatility. Size positions accordingly.

Vistra Corp (NYSE:VST)
Top-tier US IPP

Thesis: Gas + nuclear + battery portfolio. Top-2 US IPP. Up ~260% in 2024 then consolidating. Surge in DC PPA demand + ERCOT pricing power. Higher gas weighting offers nearer-term visibility than CEG.

TSMC (NYSE:TSM)
CoWoS monopoly

Thesis: The real gatekeeper of every AI chip. NVIDIA, AMD, Apple, Tesla AI silicon all packaged here. CoWoS expanding from 30K → 75K → 150K wafers/month. The most stable AI exposure for the decade.

🥈 Tier 2 — Picks & Shovels

Eaton (NYSE:ETN)
DC power systems

Global leader in transformers, UPS, switchgear. AI DC electrical content runs 3–5× a traditional DC. Backlog at record highs. 5+ year revenue visibility.

Vertiv (NYSE:VRT)
DC cooling + power

NVIDIA reference-design partner. Surging demand for immersion and direct-to-chip cooling as AI chip TDP explodes. Up ~140% in 2024.

Quanta Services (NYSE:PWR)
Transmission EPC #1

US transmission and distribution EPC leader. Direct beneficiary of Goldman Sachs' projected $720B grid investment by 2030. Backlog $35B+.

GE Vernova (NYSE:GEV)
Gas turbines + wind

Direct beneficiary of behind-the-meter gas demand. Global gas turbine #1. Wind portfolio adds optionality. Up ~130% in 2024.

🥉 Tier 3 — Korea-specific

Doosan Enerbility (KRX:034020)
SMR top-5

NuScale-based SMR + gas turbines + nuclear EPC. Core beneficiary as SMRs scale in the late 2030s. Volatile until SMR commercialization. Long-horizon position only.

HD Hyundai Electric (KRX:267260)
Transformer supercycle

Major Korean transformer exporter to the US. US grid transformer lead times 3–4 years = pricing power for HD Hyundai. Korean transmission supercycle adds a second tailwind. Up 200%+ in 2024 yet backlog remains strong.

LS ELECTRIC (KRX:010120)
Power gear + DC solutions

Transformers + distribution + automation. US and India export ramp. Direct beneficiary of Korean grid build-out + DC growth. More attractive valuation than HD Hyundai Electric.

Samsung Electronics (KRX:005930)
HBM3E + foundry

HBM3E 12-Hi qualified at NVIDIA (2025), HBM4 in volume ramp. Foundry chasing TSMC. Underowned + with momentum. Useful as a hedge alongside SK Hynix.

📊 Sample portfolio allocation

SK Hynix
18%
CEG
14%
Hanmi
8%
VST
10%
TSMC
12%
ETN
8%
VRT
6%
PWR
6%
GEV
6%
Korea 3
12% (split)

Logic: Tier 1 (60%) + Tier 2 (28%) + Tier 3 (12%). HBM axis (SK Hynix + Hanmi + Samsung ≈ 30%+) + Power axis (CEG + VST + ETN + VRT + PWR + GEV ≈ 50%) + CoWoS axis (TSMC = 12%) + Korean infrastructure (HD Hyundai etc. ≈ 8%). Time horizon 5–10 years.

Risk Factors and Final Conclusion

🔴 Four downside risks

1. AI CapEx slowdown / bubble concerns
⚠️ Largest risk

If AI revenue disappoints in 2027–2028, Big Tech CapEx could pause. The 2000 dotcom analog is worth remembering. Mitigation: diversify across Tiers 1/2/3, average in, hold for 5+ years.

2. DeepSeek efficiency revolution generalizes
🔥 Two-sided

If DeepSeek-style efficiency (10× cheaper for similar performance) becomes the industry standard, GPU and power demand could trim 30–50%. But Jevons Paradox (efficiency drives more total usage) could pull the other way. Watch closely.

3. IRA repeal / US policy volatility
Political risk

IRA tax credit cuts would weaken solar, EV, and SMR incentives. But the AI hegemony race tends to absorb policy shifts. Power demand itself is structural and policy-agnostic.

4. China self-reliance + US export tightening
Geopolitics

If China accelerates self-reliance via Huawei Ascend / CXMT / SMIC, NVIDIA and SK Hynix's China revenue compresses. But ~80% of the global market is the US and allies, limiting impact.

🟢 Composite conclusion

1. Power is the real bottleneck — Grid, transformer, and transmission lead times are far longer and harder than GPU or HBM. Capital cannot bend physics.
2. HBM is Korea's true moat — ~92% global share. The narrowest gate until CoWoS capacity unlocks in 2027–2028. SK Hynix · Samsung · Hanmi as the core trio.
3. Distributed + behind-the-meter accelerating — Big Tech can't wait for the grid and is signing direct gas/nuclear PPAs. CEG, VST, GEV are direct beneficiaries. The dominant trend of 2026–2030.
4. Korean transmission supercycle — HD Hyundai Electric, LS ELECTRIC, and Doosan Enerbility ride a US + Korea double tailwind. Peak hits as SMRs scale in the late 2030s.
5. Time horizon matters most — Meaningful returns require 5–10 years. Ignore short-term volatility. Even if AI bubble concerns rise, power demand itself is structural.

🎯 In one line

★★★★☆
CONVICTION CALL

"The real gate of the AI era is not GPUs but power and HBM. A 5–10 year diversified position concentrated on those two axes delivers the most asymmetric returns. SK Hynix · CEG · Hanmi · TSMC · VST form the core. Don't enter without a long horizon."