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 market is chasing the “AI trade” in chipmakers and a handful of US reactor startups. Korean insiders look one link down — at who will actually build the small reactors meant to power AI data centers. And here is the catch outsiders keep missing: the most-hyped names design the reactor, but they do not make it. This is not financial advice.
This is Episode 5 of our series on the AI data center power supply chain — a close-up on the nuclear value chain that designs, builds, and maintains the reactors. Generation hardware itself was covered in Ep.2; here the focus is the chain of companies behind it. See the full map in our intro: Korea’s AI power supply chain (Ep.0).
The short answer: design is American, the factory is Korean
The hardest power problem for an AI data center is not how much electricity, but uninterrupted, around-the-clock clean electricity — the kind solar and wind cannot supply on their own. That search has revived nuclear, and specifically the small modular reactor (SMR): a compact reactor whose modules can be factory-built and assembled near the data center. The United States has no shortage of SMR designers — NuScale, X-energy, TerraPower, Oklo — but it has very little capacity to actually forge the reactors. Korea built the other half: a chain that designs (KEPCO E&C), manufactures (Doosan Enerbility), and maintains (KEPCO KPS). So the first decision is which link you want — and the three do not earn money at the same time.
Why now — AI wants power that never blinks
This series runs on one line: AI eats electricity, and that electricity has to be firm. Training and inference clusters cannot pause, yet solar and wind are intermittent, and covering the gap with batteries gets expensive fast. So Big Tech went looking for firm clean baseload and arrived at nuclear. A large reactor takes a decade to build and cannot sit beside a data center; an SMR can. Through 2026 the deals piled up — Google contracted with Kairos Power for power from reactors due online between 2030 and 2035; Oklo signed for up to 12 GW with data center operator Switch and 500 MW with Equinix; Terrestrial Energy paired with Riot for up to 4 GW.[verify]
Here is the line outsiders miss: nearly all of these are deals for design and power, not manufacturing. NuScale, X-energy, TerraPower and Oklo are closer to fabless design houses, and the heavy-forging capacity to mass-produce reactor modules is thin in the US.
🎩 Under the Gat — Think semiconductors. America is strong at design; the actual making happens at Asian foundries. SMRs are splitting along the same line — the US holds the blueprints, Korea is pouring the factory. Wall Street cheers the Oklo and NuScale share prices, but the question that matters for you is simpler: whose plant does the first module come out of, and at what margin?
The value chain — who earns, and when
Korean nuclear isn’t one company doing everything. It splits into three links, and for an investor that split is the whole point — because cash arrives at different times and in different forms.
- Design — KEPCO E&C (KRX: 052690). Reactor systems design and engineering, and the lead on Korea’s home-grown i-SMR (innovative SMR). It sits at the front of the chain, so momentum arrives early — but revenue is small and the share price swings hard on a single policy headline or order.
- Manufacturing — Doosan Enerbility (KRX: 034020). Reactor modules, large forgings and EPC. The core of the foundry story. Capital-heavy, with revenue recognized at the build-and-deliver stage (the medium term). It was the star of Ep.2; here it plays the “factory” role.
- Maintenance — KEPCO KPS (KRX: 051600). Commissioning, repair, and operations-and-maintenance (O&M). The back of the chain — revenue lands late, but it compounds as the fleet of running reactors grows. The least exciting link, and the steadiest: its Q1 2026 operating profit jumped about 375% year on year to roughly ₩37 billion (≈$27M) on revenue up about 22%.[verify]
🎩 Under the Gat — Lump these three together as “the nuclear trio” and you’ll get the investing wrong. KEPCO E&C is the beta (volatility), Doosan is the body (manufacturing leverage), and KEPCO KPS is the bond-like tail (recurring revenue). To ride the power-crunch momentum aggressively, look to the front of the chain; to sit in it defensively, the back. Same cycle, different seats.
The bull case — the gap between US design and US manufacturing
The bull case is a manufacturing bottleneck. Few places on earth can forge large reactor pressure vessels, and many US SMR designers have no plant of their own. So Doosan Enerbility took the contractor’s seat: it has invested in NuScale and X-energy, completed a main-equipment contract for TerraPower’s Natrium reactor, and joined the X-energy project tied to Amazon.[verify] In late 2025 it committed about ₩806.8 billion (≈$590M) to a dedicated SMR plant in Changwon, targeting capacity of roughly 20 units a year by 2031.[verify] A tailwind followed when the US NRC put its streamlined SMR licensing framework, Part 53, into effect.[verify]
Domestic policy backs it too. Korea passed an SMR special act in February 2026, confirmed Gijang (Busan) as the candidate site for the country’s first i-SMR on June 17, 2026, and stood up SMR manufacturing-support centers in Changwon, Busan and Gyeongju plus a roughly ₩100 billion (≈$73M) nuclear-industry growth fund.[verify] KEPCO E&C designs that i-SMR; KEPCO KPS would later maintain it.
🎩 Under the Gat — What Korea is really selling is not new science — it’s speed. While the US argues over licensing and design, Korea broke ground on the factory first — ppalli-ppalli (Korea’s “hurry-hurry” execution culture). The meaning for you: whoever lays down mass-production capacity first is best placed to win the first orders. But “best placed to win” and “won” are not the same sentence.
The bear case — the year “2035,” and the thin-margin trap
Now the other side, fairly. The bear case is three points.
First, timing. Most SMR-for-data-center deals target startup in 2030–2035, and Korea’s own i-SMR targets commercial operation around 2035 (standard design approval around 2028).[verify] Today’s memoranda and reservation contracts do not show up in this year’s profit.
Second, the foundry paradox. “Design American, build Korean” is both the bull case and the bear case. A foundry depends on someone else’s design clearing its licensing and converting to orders — and contract manufacturing usually earns thinner margins than design and brand. TSMC’s unusually high margins come from a near-monopoly on leading-edge technology; Doosan, with zero commercial-SMR units produced so far, has not proven anything like that status.
Third, it’s priced in — and it swings. KEPCO E&C in particular saw violent moves in 2026 (a one-day jump of roughly 20% in early June, followed soon after by double-digit drops).[verify] A design house earns little revenue, so a single policy line or order can whip the stock — local broker Mirae Asset opened coverage with a target of ₩121,000 (≈$88).[verify] Doosan has already re-rated hard (about ₩96,700, ≈$71, on June 19, 2026), and because the business is cyclical and policy-driven, an over-eager price corrects sharply on small disappointments.[verify]
🎩 Under the Gat — The weak link in the foundry model is margin. Cutting steel is impressive, but the side that owns the design, the license, and the brand usually keeps the larger slice of the profit. The one number to track: the operating margin Doosan’s SMR unit actually prints on its first production run. That figure decides whether this is a “TSMC” or just a subcontractor.
The comparison — Korea’s would-be BWXT, with TSMC ambitions
For readers meeting it for the first time: Doosan Enerbility is, roughly, the BWX Technologies (BWXT) that Korea wants to build — with ambitions pointed toward an SMR foundry. BWXT has supplied US Navy reactors and components for decades.
| Doosan Enerbility (KRX: 034020) | BWX Technologies / BWXT (US anchor) | |
|---|---|---|
| Business | Reactor main-equipment maker + power EPC, building an SMR foundry | Naval reactors and components; government and commercial nuclear parts |
| Track record | Large-reactor construction, yes; commercial SMR units produced: zero | Decades of proven Navy manufacturing |
| Market cap | ~₩62 trillion (≈$45B, at ₩96,700, Jun 19 2026)[verify] | ~$18B (Jun 2026)[verify] |
| Customers | Fabricating for NuScale, X-energy, TerraPower + domestic i-SMR | US government (Navy) and commercial nuclear |
| Decisive difference | Its edge is capacity laid down first — but unproven | Its edge is proven manufacturing trust — but capacity expansion is conservative |
Not a perfect parallel — BWXT is a verified manufacturer hardened by government contracts, while Doosan is a challenger with no commercial SMR units yet. Buying Doosan isn’t buying a company that has become an SMR foundry; it’s betting on the chance that it becomes one. The “TSMC of small reactors” tag gets used a lot,[verify] but TSMC-grade margins and monopoly remain a hypothesis, not a fact.
So how do you frame it — seats by risk appetite
Not a recommendation — a distinction. The same SMR-and-data-center theme wears three different faces.
- Aggressive (momentum): the front link, KEPCO E&C — first and largest to react to design orders and policy news, with the most volatility and priced-in risk.
- Core (leverage): Doosan Enerbility — the direct beneficiary of manufacturing-capacity expansion, running on two clocks at once, gas turbines (near term) and SMR (long term), as covered in Ep.2. The cost is capex and time lag.
- Defensive (recurring): the back link, KEPCO KPS — the “boring but steady” seat whose maintenance revenue compounds as more reactors run. The weakest momentum.
Which seat fits is a question of risk tolerance, not of the ticker — and all three ride the same two variables: not the technology, but nuclear-policy continuity and a revenue timeline that lives in the 2030s.
Disclaimer: This article is for information only and is not financial advice. Figures marked for verification, market caps and small- and mid-cap prices should be re-checked at publish time against the KRX and DART. FX conversions use roughly ₩1,370 per US dollar as of June 2026.
Frequently Asked Questions
Are small modular reactors really being used for data centers, or is it hype?
The direction is real, but the timing is distant. Big Tech and data center operators — Google, Oklo, Terrestrial Energy and others — have signed nuclear deals, but most target startup between 2030 and 2035. For now these are contracts, not running reactors. This is not financial advice.
Why look at a Korean manufacturer instead of the US designers?
US names like NuScale, X-energy, TerraPower and Oklo are effectively design houses, and large-scale module manufacturing capacity is limited in the US. Doosan Enerbility has positioned itself to fabricate the core components for several of them — a split that rhymes with the chip industry’s design-versus-foundry divide.
Which of the three is the purest SMR bet?
For manufacturing exposure, Doosan Enerbility (KRX: 034020) is the headline name, though gas turbines and other businesses mean it is not a pure SMR play. KEPCO E&C (KRX: 052690), the design arm, is purer exposure but far more volatile. This is not a recommendation to buy any security.
What is the biggest risk?
Four things: revenue that lands mostly in the 2030s, the thin margins a foundry can earn on someone else’s design, the sharp share-price swings of a small design house like KEPCO E&C, and the historical on-off nature of Korean nuclear policy. Not financial advice.
Can foreigners buy these stocks?
Yes. Doosan Enerbility, KEPCO E&C and KEPCO KPS are all KRX-listed large or mid caps, reachable through brokers that offer Korean market access (an integrated account at IBKR and similar). Account and tax steps are covered in a separate guide.
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.