South Korea’s AI Dilemma: Plan for the Data Centres You Hope For, or the Ones You Can Count?
Every government that wants an AI industry eventually runs into the same unglamorous question: where will the electricity come from? South Korea is now facing a sharper version of it. According to Korean business press reports this week, the Ministry of Climate, Energy and Environment is weighing a cut of roughly 3 to 4 gigawatts to the AI data centre demand it pencilled into the country’s next long-term power plan. That would remove close to half of the 7.9 GW provisionally allotted for 2040.
How the numbers grew, and why they may shrink
The story of the forecast is a story of ambition. In spring, planners expected AI data centres to add around 4 GW of peak demand by 2040. After the government unveiled a set of national “mega-projects” covering semiconductors, AI data centres and physical AI, that figure jumped to 11.9 GW in an August revision. Only the first phase of the data centre plan was included; a second phase was left out because its demand looked too uncertain.
Now even part of phase one looks shaky. A ministry official was quoted as saying that some projects companies proposed are moving ahead while others are not, and that it would be more reasonable to put unclear demand into the next plan. Lawmakers had already challenged the numbers in October. One pointed out that of 96 data centre projects that passed a grid impact review, only four were actually receiving power, and more than half had not even applied for a connection. The minister conceded the forecast had been somewhat inflated. The final 12th Basic Plan for Electricity Supply is due early next year.
The case for caution
The argument for trimming is about who pays if hopes do not materialise. Power plants and transmission lines take years to build and decades to pay off. If state utility KEPCO builds a grid for data centres that never arrive, the cost lands on ordinary consumers. An independent climate think tank estimates that Korea’s announced data centre plans add up to about 20 GW by 2035, but only around 2.5 GW is backed by firm demand such as customer contracts. There is also a climate angle: analysts warn that if the extra load is met mainly by gas, Korea’s emissions targets get harder to reach.
The case for speed
Industry sees it very differently. In the AI build-out, companies talk about “time to power”: the months or years it takes to secure a grid connection now decide where global tech firms put their servers. Korea’s data centre sector is around 1.9 GW today, a fraction of the United States. Executives argue that a latecomer cannot afford to plan conservatively, because cloud giants will not commit until power is guaranteed. Some propose a different fix: let data centre operators sign direct power purchase agreements with gas or nuclear generators, rather than relying solely on KEPCO, which today allows such deals only for renewables.
A regional test case
Korea is not alone. Japan, Malaysia, Singapore and India are all juggling data centre booms against grid limits, local opposition and climate pledges. What makes Seoul’s debate useful is that it is happening in the open, with hard numbers on the table. The likely compromise is not a single forecast but a tiered one: firm demand built now, conditional demand unlocked as contracts are signed, speculative demand parked for later.
That may disappoint both sides a little. But in an industry where today’s certainty can become tomorrow’s stranded asset, an electricity plan that can bend may be worth more than one that simply aims high.