AI Data Center Power Bottleneck Map: After GnCenergy, VinaTech and Seojin System, Who Is Next?

Sat, 04 Jul 2026 13:40:00 +0900

Related context This post follows AI data-center CapEx and Korean bottlenecks, GnCenergy and AI data-center backup power, VinaTech and Bloom Energy power buffering, SK Gas vs SK Oceanplant, and the H1 2026 AI infrastructure postmortem. The relevant hubs are the Exclusive Analysis Hub and the Korean Semiconductor Value Chain Hub.

TL;DR

AI data-center power is not one trade. It breaks into several layers: generation and backup power, fuel cells and distributed power, ESS and UPS, power conversion and transformers, cables and bus ducts, cooling and EPC. Each layer maps to a different listed Korean supplier.

The first screen is straightforward.

Conclusion	Read-through
Strong direct evidence	GnCenergy, VinaTech and Seojin System have the clearest supply evidence. GnCenergy maps to Korean data-center backup generators. VinaTech maps to Bloom Energy-linked supercapacitors. Seojin System maps to ESS enclosures and racks.
Evidence still weak	HD Hyundai Marine Engine and Hanwha Engine have optionality in data-center engine power, but listed-entity-level supply evidence remains weak. Separate group-level orders from the listed company’s earnings capture.
Underpriced sub-layers	Backup generators, supercapacitors, ESS enclosures, internal power conversion, cable bus and bus ducts may still be less systematically priced than large transformer winners.

The first diligence priority is GnCenergy, VinaTech, Seojin System, Sanil Electric, Iljin Electric, Gaon Cable and SK Gas. Large electrical-equipment names are high quality, but a lot of the rerating is already known. The question now is which sub-bottlenecks turn into repeat revenue.

1. Why Power Is The Bottleneck

The IEA’s Energy and AI report sees global data-center electricity consumption rising from about 415 TWh in 2024 to around 945 TWh by 2030. In the United States, data centers may account for nearly half of electricity-demand growth through 2030.¹

The issue is not only generation. Grid equipment has become a bottleneck. The IEA notes that procurement can now take two to three years for cables and up to four years for large power transformers, about twice as long as in 2021.²

For investors, this means the opportunity set is wider than transformer makers. Backup power, ESS, UPS, internal distribution, cable systems and cooling can all become bottlenecks when AI data centers scale.

2. Value Chain Map

Layer	Products and services	Bottleneck	Listed Korean candidates
Backup and onsite generation	Diesel and gas generators, medium-speed engines, LNG/LPG power	Reliability, tier certification, delivery, fuel supply	GnCenergy, SK Gas, Doosan Enerbility, Hanwha Engine, HD Hyundai Marine Engine, SGC Energy
Fuel cells and distributed power	SOFC, PEMFC, fuel-cell BOP, supercapacitors	Load response, heat use, PPA structure, auxiliary power	VinaTech, Doosan Fuel Cell, SK Eternix, S-Fuelcell, Bumhan Fuel Cell
ESS, UPS and BBU	LFP/NCA batteries, racks, enclosures, UPS, supercapacitors	Fire safety, customer qualification, U.S. manufacturing, container design	Samsung SDI, LG Energy Solution, SK Innovation/SK On, Seojin System, VinaTech
Power conversion and distribution	PCS, DC solutions, switchgear, breakers, special transformers	Hyperscaler approval, density, efficiency	LS ELECTRIC, Sanil Electric, HD Hyundai Electric, Hyosung Heavy Industries
Grid equipment	Power transformers, GIS, reactors, breakers	Production slots, North American certification, lead times	HD Hyundai Electric, Hyosung Heavy Industries, LS ELECTRIC, Iljin Electric, Cheryong Electric
Cables and bus ducts	High-voltage cables, cable bus, bus duct, internal wiring	Copper, certification, installation	Gaon Cable, LS Eco Energy, Taihan Cable, Iljin Electric
Cooling and mechanical	HVAC, immersion cooling, CDU, cleanroom	PUE, heat density, coolant and refrigerant	Shinsung E&G, GST, LG Electronics, S-Oil
EPC and operations	Data-center construction, electrical work, power operations	Permitting, schedule, integrated design	Samsung C&T, HanmiGlobal, Hyundai E&C, DL E&C

3. Evidence-Based Ranking

Rank	Company	Ticker	View	Why
1	GnCenergy	119850	Watchlist / conditional Buy candidate	The most direct Korean data-center backup-generator supplier. Evidence includes Naver, kt cloud, LG CNS and other IDC projects.³
2	VinaTech	126340	Watchlist	A direct Bloom Energy AI data-center supercapacitor supply contract, with overseas revenue around 90% according to the company.⁴
3	Seojin System	178320	Watchlist	ESS enclosures, racks and containers. Samsung SDI America and U.S. ESS supply-chain exposure are the key evidence points.⁵
4	Sanil Electric	062040	Wait / Watchlist	Strong Bloom Energy special-transformer evidence, but valuation may already reflect the theme.⁶
5	LS ELECTRIC	010120	Wait	A complete power-distribution, DC, ESS and transformer platform. The issue is price.⁷
6	HD Hyundai Electric	267260	Wait	Very strong North American data-center equipment evidence, but the large-cap rerating is already advanced.⁸
7	Iljin Electric	103590	Watchlist	Both heavy electrical equipment and cables. The Malaysia data-center cable order is direct evidence.⁹
8	Gaon Cable	000500	Watchlist	U.S. AI data-center cable supply and U.S. capacity expansion are direct signals.¹⁰
9	SK Gas	018670	Watchlist	Ulsan GPS and SK’s Ulsan AI data-center power architecture make it a low-valuation infrastructure candidate.¹¹
10	Samsung SDI	006400	Watchlist	Data-center ESS, UPS and BBU exposure is real, but EV batteries still dominate the equity story.¹²

4. Main Ideas

GnCenergy: Backup power pure-play

GnCenergy is the clearest direct supplier in this screen. AI data centers cannot tolerate downtime, so backup generation is not an old-economy afterthought. It is the last reliability layer behind compute capacity. The risk is project lumpiness, margin volatility and potential environmental regulation against diesel backup generators.

VinaTech and Sanil Electric: Bloom Energy supply chain

The Bloom Energy supply chain is not only about fuel cells. If onsite SOFC power becomes more common for AI data centers, power smoothing and conversion hardware also matter. VinaTech supplies supercapacitors for AI data centers through Bloom Energy, while Sanil Electric has Bloom-linked special-transformer evidence. The key test is whether these become repeat orders rather than one-off projects.

Seojin System: ESS picks-and-shovels

AI data centers need ESS, UPS and BBU systems, but the cell supplier is not the only beneficiary. Enclosures, racks, containers, thermal design and fire-safety structures can be manufacturing bottlenecks. Seojin System is interesting for that reason. The issue is not revenue growth alone. Working capital, net debt and cash-flow quality must also improve.

Cables and bus ducts

Iljin Electric, Gaon Cable and LS Eco Energy sit in a less glamorous but important layer. Data centers need not only external grid connections but also dense internal power distribution. Cable bus, bus ducts and certified high-voltage cables can matter more as rack density rises. Copper pass-through and margin discipline are the main risks.

Energy campus

SK Gas, SK Eternix and Doosan Fuel Cell map to a broader energy-campus thesis. Data centers may value secured power more than cheap power. LNG/LPG generation, renewables, fuel cells and ESS may be bundled into site-level power solutions. SK Gas looks the most investable in this group because Ulsan GPS and the SK Ulsan AI data-center structure are more concrete, while SK Eternix and Doosan Fuel Cell still need direct PPA or commercial-order evidence.

Engine optionality

HD Hyundai Heavy Industries’ U.S. data-center HiMSEN engine order proves that medium-speed engines can be part of the onsite-power solution.¹³ But that does not automatically make HD Hyundai Marine Engine or Hanwha Engine direct beneficiaries. Listed-entity-level orders, production roles and margin capture must be confirmed first.

5. Entry, Catalysts and Invalidation

Group	Entry condition	Catalyst	Invalidation
GnCenergy	Data-center backlog grows, margins hold, valuation stays reasonable	LG Uplus Paju AIDC follow-on, Naver or telecom data-center expansion	Two quarters of slowing orders, margin collapse, diesel backup demand weakens
VinaTech	Bloom revenue recognition, module supply, visible path to profitability	Bloom data-center power expansion, follow-on purchase orders	One-off Bloom order, ongoing losses, no customer diversification
Seojin System	ESS enclosure growth and operating cash-flow improvement together	Samsung SDI America, Fluence or Powin-linked ESS expansion	Working-capital deterioration, debt pressure, low-margin OEM trap
Sanil Electric	New customers beyond Bloom, margin stability, valuation reset	Follow-on special-transformer orders	One-off Bloom order, margin compression, customer concentration
Iljin Electric and Gaon Cable	Repeat data-center cable orders and copper pass-through	U.S. and Southeast Asia data-center cable wins	Copper spike, certification or installation delays, commodity cable mix
SK Gas	More detail on SK/AWS Ulsan power economics	Ulsan AI data-center power contract, UGPS stable operation	Gas spread pressure, power-market regulation, weak PPA economics
Hanwha Engine and HD Hyundai Marine Engine	Listed-entity data-center engine order	North American onsite-generation orders	No order, unclear group margin capture, stock rises only on expectation

Final View

The strongest screen today is GnCenergy, VinaTech and Seojin System. They are not the obvious large transformer winners. They sit in lower-level bottlenecks inside data-center power infrastructure.

Sanil Electric, Iljin Electric and Gaon Cable have direct order evidence, but valuation and margin proof matter. SK Gas is not a pure-play, but Ulsan GPS and SK’s AI data-center power structure make it a low-valuation infrastructure candidate. Samsung SDI and LG Energy Solution have data-center ESS exposure, but EV battery cycles still dominate their equities.

Hanwha Engine and HD Hyundai Marine Engine remain stories ahead of evidence. Engine-based onsite power is a real theme, but listed-company revenue and margin capture must be proven.

The practical stance is not “buy every power stock.” It is to monitor repeat contracts, backlog quality, margin capture and cash flow, then enter selectively when the evidence improves.

UPS on Korea Invest Insights