South Korea's $10 Billion AI Data Center: Who Pays for the Power Infrastructure?

Brookfield Asset Management, Nvidia, and South Korean internet company Naver announced a three-way partnership to expand an AI data center at Naver's GAK Sejong facility to 200 megawatts by 2028, with Brookfield investing up to $9 billion and Nvidia contributing $1 billion.^[1][2] The project is framed as Korea's "sovereign AI factory" and represents one of the largest capital commitments to data-center infrastructure outside the United States. But the disclosure is almost silent on the mechanism that should haunt any energy journalist: who builds the power plant that feeds this load, what it costs, and which ratepayers bear the risk if the capacity sits half-used.

Start with what is hidden. The announcement names capital sources but does not disclose the power supply contract between Naver and Korea's utility (likely KEPCO, the Korea Electric Power Corporation). Is the capacity tied to firm, full-time service, or can it curtail during grid stress? What is the demand ratchet, does Naver pay for reserved generation whether or not it consumes it? Is new transmission or generation assigned to Naver's class, or socialized across all ratepayers? Korea Electric does not publish special contracts the way US state public utility commissions require, so the terms are opaque. Yet the grid cost is real. A 200 MW load represents roughly 1.4 to 1.8 terawatt-hours annually, assuming 70, 80% capacity factor, equivalent to the electricity demand of 300,000 to 400,000 Korean households.

The second move in this deal is generation. The press releases note that Brookfield is investing in "power infrastructure" alongside the data center itself, which typically means Brookfield will own or finance dedicated generation, solar, wind, or gas, to serve the facility.^[5] This is the protective model: the load owner builds its own capacity, sparing ratepayers. But if Brookfield is capturing the investment tax credits and accelerated depreciation available under Korea's renewable energy incentives, the public is already subsidizing the deal. And if new grid upgrades are needed to interconnect Brookfield's generation to the data center or to firm up supply during low-wind or low-solar hours, those costs will flow to KEPCO's balance sheet and eventually to Korea's ratepayers. The precedent is instructive: in the US, when Amazon or Microsoft build data centers with behind-the-meter solar or battery storage, they often negotiate a firm interconnection to the utility grid as a backup, and the cost of reinforcement flows to the rate base. The difference is that US dockets are public; Korea's are not.

Third, the load forecast itself becomes a rate-case weapon. Naver has announced ambitions to reach 1 gigawatt by 2029, five times the 200 MW now being funded.^[5] If KEPCO incorporates that gigawatt target into load forecasts, it will justify new generation and transmission capex. But has Naver signed contracts guaranteeing that gigawatt load? Or is the 1 GW target aspirational, announced to attract investor and government support, with actual buildout dependent on customer demand for AI compute? If KEPCO commits to generation and network upgrades on the basis of the announced target but Naver builds to only 500 MW, the stranded capacity becomes a charge on all Korean electricity consumers. This is the exact pattern documented in the US: utilities inflate data-center load forecasts in integrated resource plans, win approval for new plants, and when the load materializes at 60, 70% of the forecast, ratepayers carry the shortfall. Grid Strategies has shown that national US data-center load forecasts ballooned in recent years, and audits routinely find that a large share of the pipeline is announced but not contracted, and subject to cancellation or delay.

The fourth layer is subsidy arbitrage. Korea has national AI ambitions and has already committed government funding to data-center expansion. Brookfield's $9 billion is framed as private capital, but it will likely flow to assets with government tax incentives, export credit guarantees, or land and power availability commitments from Seoul. If KEPCO or the Korea Energy Agency commits preferential rates or reserved capacity to Naver's facility in exchange for the economic-development optic, those rents come from other ratepayers. The deal does not disclose any such commitments, which suggests they either do not exist or are confidential.

What Korea's regulator should demand now: (1) Publication of the Naver, KEPCO power supply contract, including the demand charge structure, ratchet, term length, and collateral. (2) A separate tariff schedule isolating Naver's cost responsibility for all new generation and transmission needed to serve the 200 MW initial deployment and any committed follow-on phase. (3) A demonstration that the load forecast used to justify new capex is based on signed load commitments, not press-release aspirations. (4) Disclosure of any government subsidies, tax incentives, or reserved-capacity commitments. The US precedent, documented by the Harvard Electricity Law Initiative, shows that without these protections, utilities routinely extract ratepayer subsidies by locking in special contracts that cover only a fraction of the facility's true system cost.