South Korea's Power Crunch: Why 27GW Demanded a Nuclear Emergency

The number that broke the boardroom

In April 2025, South Korea’s power planners projected that the country’s peak electricity demand by 2040 would sit somewhere between 131.8 and 138.2 gigawatts. By June, they had to scrap that calculation entirely. The new range: 158.4 to 165.0GW. In four months, the country needed roughly 27GW of additional generation — an amount equivalent to building 19 large nuclear reactors, or replacing every single reactor currently operating in South Korea with fresh capacity.

The pivot was not driven by population growth or a resurgent manufacturing sector. It was driven by two things that English-language energy analysts have been slow to connect: the speed of South Korea’s semiconductor megaprojects, and the explosive, poorly accounted-for demand from AI datacenters.

A datacenter that sold itself before the foundation dried

Consider the site in Paju, Gyeonggi Province. In April 2025, LG Uplus was pouring concrete for what would become one of the largest AI datacenters in the Seoul metropolitan area. The complex spans an area equivalent to 21 soccer fields. Building one is already rising; building two and three is scheduled through 2028. The first building’s capacity — 200 megawatts — was already sold before construction finished.

Two hundred megawatts. That is enough to power 590,000 households continuously, or run tens of thousands of advanced AI semiconductor chips. There is no other facility in the capital region with that kind of secured grid connection. And it is only the beginning.

This is the hidden architecture of the AI arms race: not the chips themselves, but the grid connections that allow them to run. South Korea’s power grid is becoming the bottleneck and the battlefield.

The semiconductor shock no one priced in

The 27GW revision traces directly to the government’s announcement in late June of three so-called mega-projects. The first: the Yongin semiconductor cluster, completing ahead of schedule. The second: four new semiconductor fabs planned for the Honam region, with a former military airbase in Gwangju selected as the site. The third: AI datacenter demand that exceeded every prior forecast.

Samsung Electronics and SK Hynix together presented a demand potential of roughly 40GW for semiconductor operations alone. The planning committee incorporated 36.8GW of that — a figure that still raised eyebrows. The committee explicitly noted that not every announced investment plan was accepted at face value.

The distinction matters. Semiconductor capacity was treated as relatively certain because the investors are globally validated companies executing visible construction. Datacenter demand was treated with far more skepticism. Of the 20.8GW in announced datacenter plans, only 10.8GW made it into the revised forecast. The remaining 10GW was deferred pending market conditions and actual project progress.

Professor Yoo Seung-hoon of Seoul Science and Technology University, who chaired the subcommittee for the 9th power plan, offered a blunt historical check: South Korea’s entire datacenter sector currently draws just 1GW of power. The question of whether that could grow 15-fold in 15 years warranted caution. “In previous plans,” he noted, “we ended up reflecting only about half of what was requested. That turned out to be correct.”

The math that isn’t being shown

Here is where the analysis gets uncomfortable. The committee published its methodology for renewable energy supply — breaking down exactly how 45GW of offshore wind would be assembled from existing permits, new site proposals, and current capacity. But on the demand side, the formula stops.

Why 36.8GW of semiconductors and not 40? Why 10.8GW of datacenter demand and not 20.8? The inputs are not transparent. Environmental groups have called for the government to distinguish between confirmed demand and speculative demand before locking in infrastructure decisions.

There is a clear incentive structure at work. If demand is inflated, the government builds more power plants — likely nuclear, given the political direction — and the costs flow through electricity rates. If demand is understated, factories cannot secure power and investment moves elsewhere. Climate and Energy Minister Kim Sung-hwan acknowledged the trap directly: accurately predicting 2040 demand is “close to impossible.”

The renewable gap that keeps growing

The government’s renewable energy projection, released alongside the demand revision, calls for 220GW of installed capacity by 2040 — up from roughly 37GW today. Solar would account for 155GW; offshore wind would grow from near-zero to 45GW.

But capacity is not the same as dispatchable generation. During a May 2025 panel, analysts calculated that 100GW of renewable capacity would generate only about 58% of the annual electricity consumption created by the mega-projects. The intermittency problem is not theoretical — it is arithmetic.

The industrial response has been direct. Samsung Electronics Vice Chairman Jeon Young-hyun told a national reporting session in June that the company formally requested the government to expand nuclear and combine it with LNG cogeneration to address renewable intermittency. This is the sector’s biggest electricity consumer, speaking plainly about what it needs.

What this means for the global nuclear trade

South Korea’s reversal is not happening in isolation. The country has spent a decade alternating between pro-nuclear and anti-nuclear power plans with each change of administration. The 8th plan under Moon Jae-in canceled six new reactors and halted lifetime extensions on ten aging units. The 10th plan under Yoon Suk-yeol reversed course. The 11th plan, finalized in February 2025, included two new large reactors and one small modular reactor (SMR), with the SMR site selected in Busan’s Gijang district and the large reactor site in Uljin, Gyeongbuk.

The 12th plan — currently under development and the first overseen by the new Climate and Energy Ministry after the presidential transition — will decide whether to add more. The government has publicly stated it is reviewing whether to include new reactors and is considering extending the allowable operation period for existing plants from 10 years to 20 years beyond their original design life.

The timing could work. Professor Yoo pointed out that existing sites at Gorae, Yeongdeok, and Hanbit have spare land, which could save the three to four years typically required for site designation and public consultation. But the financing question remains unresolved. Transmission grid costs are currently borne entirely by Korea Electric Power Corporation (KEPCO) and passed directly to ratepayers. Railway and highway projects receive 30–50% national budget subsidies. The disparity means expanding the grid for semiconductor and datacenter corridors will hit consumers disproportionately unless the funding structure changes.

The uranium angle nobody is discussing

A 27GW increase in baseload demand, if met primarily with nuclear, translates into meaningful uranium demand. South Korea currently operates 19 active reactors. Adding 19 more — even partially — would roughly double the country’s uranium procurement. For a nation that imports virtually all of its nuclear fuel, this is a strategic shift. Japan and South Korea together already account for a significant share of global uranium spot market demand. A Korean rebuild would tighten an already constrained market.

The Global Firepower uranium demand model suggests each new 1GW reactor requires roughly 200–250 tonnes of uranium annually once operational. Nineteen new reactors could add 4,000 to 5,000 tonnes of annual demand — a material shock to a market that produces roughly 60,000 tonnes per year globally. This is not abstract: Korean utility procurement contracts in 2026–2027 will set the tone for the next round of mine development in Kazakhstan, Canada, and Australia.

Who wins and who loses

Samsung and SK Hynix win. They have secured a credible path to the power their expansion requires, and their global competitiveness depends on it. LG Uplus and other datacenter developers win by association — priority grid access in a constrained system is a competitive moat.

South Korean electricity consumers lose. Every new reactor, every transmission line, every demand-side management shortfall gets folded into rate structures. The government’s own projection assumes 19.4GW of demand reduction through efficiency programs — described internally as “highly challenging.” Miss that target and the gap widens further.

Nuclear supplier countries win. France, South Korea’s own reactor export industry, and potentially new SMR developers all stand to benefit from a sudden, state-backed demand surge. The question of whether the industry can deliver at scale within the timeline is separate from the question of whether the political will now exists.

The real story is the timeline

The 12th power plan’s provisional draft will be presented to parliament in October, with final approval expected before year’s end. The 27GW revision is not a final number — it is a signal that the entire post-fukushima energy architecture South Korea built over the past decade is being recalibrated under pressure from forces that outpaced every model.

The semiconductor and datacenter industries are not asking for favors. They are executing globally competitive capital plans, and the grid is the constraint. South Korea’s response — accelerating nuclear, deferring some datacenter demand, stretching existing reactor lifetimes — is a pragmatic, arguably necessary recalibration. But the opacity around demand assumptions and the concentration of grid costs on consumers are liabilities that will surface in the implementation phase.

The 19-reactor gap is a number. What it represents is a country confronting the physical reality of the AI economy — and finding that its previous energy bets did not add up.