Forget the Chips for a Second. AI’s Real Bottleneck Is Power — and Korea Quietly Owns Pieces of It.

This article is for informational purposes only and is not financial advice. TheGatBull may earn a commission from some links at no cost to you — see our disclosure and full disclaimer.

The entire market is buying one AI story: the chips. Nvidia, HBM memory, the GPU arms race. It’s a real story. But step back and ask a dumber, more physical question — what actually makes a data center run? The answer isn’t silicon. It’s electricity, and right now electricity is the thing that’s running short. This is a map of the other AI trade: the power supply chain that feeds the data centers — and the surprising number of links in it where Korean manufacturers quietly sit near the front of the line. This is not financial advice.

The short answer: AI’s bottleneck is moving from chips to electrons

You can stack GPUs to the ceiling, but with no power to run them and no cooling to keep them from melting, they’re paperweights. That’s why the money is quietly flowing past the chip and into the generators, transformers, cables, storage and cooling that stand behind it. US data-center electricity demand is projected to more than double — from roughly 180 TWh in 2025 to about 391 TWh by 2030 — and that single curve is dragging an entire industrial chain into a super-cycle. Think of it as the “picks and shovels” version of the AI rush: instead of the people mining the gold (chips), the people selling them the power. (Information, not advice.)

Mr. Gat, TheGatBull's Korean-market mascot

🎩 Under the Gat — Wall Street counts GPUs. Korean factories are counting transformer, gas-turbine and cable orders. The real queue for AI isn’t at the chip fab — it’s on the power grid. And standing near the front of that line, with decades of heavy-industry muscle, is Korean manufacturing. Watch the order backlog, not the headlines.

Read the map two ways: the vertical axis and the horizontal axis

This ecosystem only makes sense if you read it along two axes.

The vertical axis — “the journey of electricity.” The upstream-to-downstream chain that gets a single electron from a power plant to a GPU: generation → transmission & transformers → cables/connection → storage (ESS) → the last mile (in-data-center distribution and cooling). This is the axis of depth.

The horizontal axis — “the layers that cut across everything.” The cross-cutting layers that support every vertical stage: materials & components, engineering/EPC/construction, and operations/O&M/grid software. This is the axis of breadth.

The vertical map — Korea’s stocks, link by link

Mr. Gat pointing at the AI data center power supply chain map

Link (the electron’s journey) What it does Representative Korean names Global anchor
1. Generation Gas turbines, SMRs, nuclear, fuel cells Doosan Enerbility, KEPCO E&C, KEPCO KPS GE Vernova, Siemens Energy, Mitsubishi (JP)
2. Transmission & transformers Ultra-high-voltage transformers, switchgear Hyosung Heavy Industries, HD Hyundai Electric, LS Electric Hitachi Energy, Siemens Energy, GE Vernova
3. Cables / connection Extra-high-voltage, subsea and HVDC cable Taihan Cable & Solution, LS Eco Energy Prysmian (IT), Nexans (FR)
4. Storage (ESS) Load balancing, backup power LG Energy Solution, Samsung SDI CATL (CN), Tesla Megapack
5. Last mile In-data-center power distribution & cooling (no large pure-play — Korea’s weak link) Vertiv, Eaton, Schneider

Not a perfect parallel — most of these Korean names are not “pure AI stocks.” They are existing heavy-industry companies onto which AI demand has been bolted, so what matters is the AI/data-center share of each order book, not the label.

The horizontal map — the layers that cut across

Layer What it does Representative Korean names
A. Materials & components Electrical steel, windings, bushings; battery materials POSCO (electrical steel); EcoPro BM, L&F, POSCO Future M (battery materials)
B. Engineering / EPC / construction Data-center construction, grid build-out Hyundai E&C, GS E&C, Samsung E&A
C. Operations / O&M / software Plant maintenance, grid & data-center management KEPCO KPS and others (software layer is globally dominated)

Where Korea is strongest: transformers and power generation

Transformers (link 2) are Korea’s best square on the board. The “big three” — Hyosung Heavy Industries, HD Hyundai Electric and LS Electric — posted record results on the back of a global rush to reinforce aging grids and connect new data centers. Industry reporting puts their combined order backlog above ₩22 trillion, with Hyosung holding the top share of the US 765 kV ultra-high-voltage market and HD Hyundai Electric estimated around 15–20% of the US market. Rather than be cornered by US tariffs, they are answering with on-shore expansion — new and expanded plants in Alabama, Tennessee and Texas.

Generation (link 1) runs through Doosan Enerbility. It has signed deals to supply a US big-tech data center with four 370 MW steam turbines and four generators, is developing a mid-size 90 MW gas turbine aimed squarely at hyperscalers, and carries an SMR (small modular reactor) story on top — having become the first in Korea and fifth in the world to independently develop a large-scale gas turbine.

Mr. Gat presenting which links of the AI power chain Korea is strongest in

🎩 Under the Gat — Korea is famous for chips, but its older muscle is heavier: building big, hard things fast — ppalli-ppalli. Transformers, generators, cables, reactors. The AI era suddenly made “the ability to make and move electricity” valuable again, and that old muscle is back in the spotlight. Just remember: not every square on this board is equal.

The cables and storage links — real, but read the fine print

Cables / HVDC (link 3) are booming as data centers and renewables both demand more high-voltage connection: Taihan Cable & Solution and LS Eco Energy ride subsea and HVDC demand, and Hyosung has broken ground on a new HVDC transformer plant in Changwon. Storage (link 4) is the battery makers’ second act: with EV demand soft, LG Energy Solution and Samsung SDI are converting lines to ESS and chasing US data-center power deals — LG ES alone signed a $1.6 billion, 6 GWh LFP storage contract tied to a hyperscaler buildout. We give that battery pivot its own dedicated episode in this series.

Be honest about the weak link: cooling and the last mile

A fair map shows the gaps too. The last mile inside the data center — power distribution, UPS and especially advanced cooling — is dominated by global players like Vertiv, Eaton and Schneider, and Korea has no large pure-play there. As power density per rack explodes, cooling is becoming its own bottleneck — and it’s the one square on this board where Korea is, for now, a spectator rather than a leader.

Mr. Gat looking analytical while weighing the risks in the AI power supply chain

🎩 Under the Gat — The “AI power” theme is real, but don’t treat every link as equally good. These are cyclical, capital-heavy businesses; when expectations get priced in ahead of the order book, the pullbacks are sharp. Separate the squares where Korea is genuinely #1 or #2 in the world (transformers, gas turbines) from the squares where it’s just along for the ride. Even picks and shovels come in different grades.

Both sides, fairly — bulls vs. skeptics

The bull case. The power bottleneck is structural and multi-year, not a one-quarter fad. Korea sits in the exact links the world is short of — heavy electrical equipment and dispatchable generation — and is building US capacity to clear tariff walls. The order backlogs are real and already booked.

The skeptic case. Much of this is a cyclical industrial trade dressed in an AI narrative; the “AI share” of some order books is smaller than the story implies. Tariffs and policy can swing both ways, valuations have already re-rated, and the weakest links (cooling, software) leave Korea exposed to importing the highest-margin pieces.

Mr. Gat, arms crossed, summing up the AI power supply chain map

🎩 Under the Gat — Here’s how I read the whole board: the chip trade is crowded, the power trade is the same bet one layer down. The signal to watch isn’t a stock price — it’s the order backlog and how much of it is genuinely data-center-driven. Map first, then pick your square.

How to use this map (a view, not advice)

This is a theme spread across several industries, not a single ticker. Read it vertically to follow the electron, horizontally to see who supplies every stage, and then judge each company on one question: how much of this order book is actually AI? This article is the map. In the episodes that follow, we zoom into each link — starting with the square Korea is strongest in, the transformer makers, and including a dedicated look at the battery/ESS pivot. “AI power” does not mean “automatic gains,” and tickers and figures should always be checked against primary sources as of your trade date. (Not a recommendation to buy any specific stock.)

The bottom line: the market is fixated on the chip, but the data center’s hard limit is the power behind it — generation, transformers, cables, storage and cooling. Korea doesn’t own every link, but it owns some of the best ones. Follow the electrons, and you’ll find a quieter, more industrial slice of the AI trade hiding in plain sight.

— Mr. Gat 🐂

This is not financial advice. Past performance does not guarantee future results, and all figures should be checked against primary sources as of your trade date.

Frequently Asked Questions

Why focus on power instead of AI chips?

Because the physical bottleneck for AI is increasingly electricity, not silicon. You can buy all the GPUs you want, but they are useless without power to run them and cooling to keep them alive. US data-center electricity demand is projected to more than double from roughly 180 TWh in 2025 to about 391 TWh by 2030, and that surge is pulling the entire power supply chain — generation, transmission, transformers, cables and storage — into a multi-year up-cycle. This is not financial advice.

Which link of the chain is Korea strongest in?

Transformers and heavy electrical equipment. Korea’s “big three” — Hyosung Heavy Industries, HD Hyundai Electric and LS Electric — hold leading positions in the US ultra-high-voltage market, and power generation (Doosan Enerbility’s gas turbines and SMRs) is the other standout. The weakest link is data-center cooling and power management, where global names like Vertiv and Eaton dominate and Korea lacks a large pure-play.

How should an investor read this map?

Two ways. Vertically, follow the journey of electricity from generation to the server rack. Horizontally, look across the layers that cut through every stage — materials, engineering/construction, and operations/software. Most Korean names here are not “pure AI stocks”; they are existing industrial companies that AI demand is now lifting, so the key is checking how much of each order book is actually AI-driven. Not a recommendation to buy any specific stock.

Is this a single stock to buy?

No. The AI-power supply chain is a theme spread across many companies and several different industries. This article is the map; each link deserves its own close-up, which is what the rest of this series does. Always verify tickers and figures against primary sources as of your trade date.

This article is for informational purposes only and is not financial advice. TheGatBull may earn a commission from some links at no cost to you — see our disclosure and full disclaimer.

Leave a Comment