Roughly half of what developers asked for did not survive the spreadsheet. Companies submitted data centre requests totalling 20.8GW; the working group counted the 10.8GW in the first phase where the details were firm and set the other 10GW aside as too uncertain [14][3].
That filter is the most useful line in the exercise, because it draws a distinction no capacity announcement ever draws. What developers do is book headroom early, at every site they might eventually use. What a demand forecast tends to assume is that a request is a load.
The fab line got the opposite treatment. It was multiplied rather than filtered, to six or seven times its April value in four months [2], as the Yongin cluster, four new fabs in the Honam region and expansions at Pyeongtaek and Cheongju were folded in [12], all of it tied to a chip and AI investment programme Reuters sizes at 800tn won [6]. Poured concrete is more persuasive than a submitted form.
Put the revisions together and the 2040 peak forecast moves from 131.8 to 138.2GW in April to 158.4 to 165GW now [7][8], about 26.7GW of new load appearing inside a planning document over four months [1]. Current peak demand is around 100GW [10], so the revision by itself is about a quarter of today's peak [4], and the top of the new range sits some 65% above it [5].
The published figures also do not agree with each other. Kim Sung-hwan's 25 to 30GW for AI and fabs [1] is smaller than the roadmap's own components, which add to 36.2 to 36.5GW [6], and the 11.9GW data centre line [13] is 1.1GW larger than the amount the planners called firm, a difference the account does not explain [7]. None of these numbers is an allocation to anyone. If you are siting a load in Korea, the thing to establish is which of them your utility is working from.
The calendar is why this becomes a nuclear argument almost immediately. Offshore wind plus the transmission to move it takes roughly a decade, the same as a reactor and longer than the fabs now going up [19]. Renewables are 11% of the mix, one of the lowest shares in the OECD, with coal at 29% and gas at 27% [18]. Park Woo-young of the Korea Energy Economics Institute told the Korea JoongAng Daily that among carbon-free sources nuclear is effectively the only one that can follow the load patterns a grid actually experiences [16], which is what a cluster of fabs running flat out produces. Nationally the politics are available: 79% of 2,000 respondents to a poll by the state-run Korea Energy Information Culture Agency backed including new plants in the roadmap [20], while in May a civic group in Ulsan, one of the candidate sites, collected more than 10,000 signatures against additional reactors on safety grounds [21].
For anyone with a load to place, the frame has two axes: whether a planner will count your demand as firm, and whether your project can wait ten years for supply. If you are counted and patient, the 12th plan is written for you. Counted but urgent, and you are queueing behind Yongin for the same early-2030s megawatts. Fall into uncounted and urgent, and you are in the 10GW that was set aside, negotiating privately for capacity no plan has promised. Land in uncounted and patient, and you are the part of the forecast that gets generation built for it and then does not turn up.
A government proposal is due in October, with the plan finalised by the end of the year [24]. The firmness test is the number worth watching there, because that is the line deciding whose megawatts are treated as real.
