AI and Data Centers: Climate Game-Changer, or Korea's Next Grid Crisis?
In One Sentence
Korea's first national Climate Citizens' Assembly (2026, 200 deliberating citizens — 20 planning + 180 deliberating-only — established under Article 19-2 of the Carbon Neutrality Framework Act) is being asked whether artificial intelligence and the data centers behind it should be governed primarily as a climate solution (grid optimization, materials discovery, fusion control) or as a climate threat (a demand shock the Korean grid is not built to absorb).
Why This Matters — A Distinctly Korean Story
The Double-Edged Sword
AI's climate footprint is genuinely bidirectional, and Korea sits at both edges:
- As an accelerator of mitigation — grid optimization, DeepMind's GNoME (2.2 million candidate materials in a single 2023 release), AI-driven plasma control in fusion experiments, precision agriculture.
- As an accelerator of emissions — global data-center electricity demand is projected to rise from 415 TWh in 2024 to 945 TWh in 2030 (IEA, Energy and AI 2024), more than doubling in six years.
The Assembly is being asked which face of this dual-use technology Korean policy should prioritize — and how to write a rule that does not require choosing in the abstract.
Korea's Structural Mismatch
Korea is unusually exposed to the data-center demand wave, for four specific reasons:
- Seoul Metropolitan concentration — Roughly 70% of Korean data centers cluster in the Greater Seoul Area, the same region whose grid is already capacity-constrained (the direct link to Agenda ③ Seoul-Metro vs Non-Metro).
- Permitting bottleneck — Of 195 new data-center applications recently submitted, only 4 cleared all permits (2.1%), throttled by transmission, environment, and local consent.
- RE100 procurement at just 12% — Korean IT exporters are committed to 100% renewable procurement under global supply-chain pressure, yet Korea's actual renewable supply to data centers sits at one-eighth of that target.
- 11th Basic Plan for Electricity Supply and Demand (2024–2038) — Projects data-center load rising from 8.2 TWh (2025) to 18 TWh (2030) — and that is widely regarded as a conservative number.
The Jevons Paradox
A widely-discussed ACM FAccT 2025 paper warns that efficiency gains in AI compute tend to expand the addressable use cases faster than they reduce per-query energy — total consumption rises even as each model gets more efficient. This breaks the intuitive equation "efficiency = mitigation." For the Assembly, it means that betting on hardware improvements alone is unlikely to keep data-center load inside the 2030 grid envelope.
The Legal and Policy Gap
- Electricity Business Act §16-2 (revised 2024) — introduced a "distributed siting incentive" for new data centers outside the capital region. Implementation has been slow; effectiveness is not yet measurable.
- Data-Center Energy Efficiency Grade System — under inter-ministerial discussion; no statutory rule yet in force.
- No Korean equivalent of Ireland's data-center moratorium debate, Singapore's 2019–2022 moratorium, or Virginia's localized capacity-market constraints.
What Korea Could Learn From — and Adapt
| Reference | Korea's Adaptation Question |
|---|---|
| Ireland — data centers now account for ~21% of national electricity use (CSO 2024); a national policy debate over new connections is open | Should Korea cap data-center share of national electricity at a stated threshold before reaching Irish levels? |
| Singapore moratorium (2019–2022) — paused new data-center approvals, then reopened with a green-DC standard | Could Korea introduce a time-bounded permitting pause to design a Korean green-DC standard? |
| Virginia, USA — the world's densest data-center cluster; localized grid stress now driving cost shifting to residential ratepayers | A direct warning Korea has not yet absorbed: who pays when DC demand collides with household tariffs? |
| IEA Energy and AI (2024) — framed governance, not technology, as the determinative variable | This is the Assembly's central thesis: the outcome is not pre-determined by the technology |
Issues the Assembly Is Weighing
- Siting regulation — A statutory cap on capital-region concentration, with renewable-rich southern provinces (Honam) and east coast (Gangwon) as incentivized destinations.
- Mandatory RE100 for new data centers — Should new DCs be required to operate at 100% renewable procurement from the commissioning date, not on a transition timetable?
- Public-interest treatment of AI outputs — Big Tech monopoly versus open access to climate-relevant outputs (e.g. materials databases like GNoME).
- "AI Net-Positive" certification — A standard for AI applications whose mitigation contribution provably exceeds their electricity footprint.
- Household automation and privacy — Smart appliances and home energy management offer meaningful demand-side flexibility, but only if citizens accept the data-collection trade-off.
En-ROADS Lever Mapping
- L4 Renewables (direct) — DC demand growth is the upstream pressure on Agenda ⑬ (Renewable Allocation).
- L11 Buildings & Industry Efficiency (direct) — DC efficiency standards live here.
- L8 Carbon Pricing (indirect) — a carbon price accelerates DC siting decisions and RE100 procurement.
- L18 Technology-based Removal (indirect) — AI-discovered materials could lower CCUS and DACCS costs over the next decade.
- Moderator tip — Demonstrate L4 + L11 + L8 together. The lesson is not "AI is good" or "AI is bad" — it is that governance, not technology, determines the climate outcome.
Open Questions Before the Assembly
- Is the mandatory RE100 commissioning rule politically achievable, given that it could push Korea's largest tech investors offshore?
- If data centers move to Honam and Gangwon, do the new jobs offset the political fact that Seoul-area citizens have grown accustomed to siting these facilities elsewhere already?
- How should Korea regulate AI inference data centers (high duty cycle, high marginal load) versus training data centers (burstier, more flexible)?
- Does Korea need a single AI Climate Governance Authority, or should the question stay split across the Ministry of Science and ICT, the Ministry of Trade Industry and Energy, and the new Ministry of Climate, Energy and Environment?
This agenda is one of the plenary integration round M2 mega-agendas. Citizens are being explicitly asked not to lean optimist or pessimist; the moderator's stated frame is that the outcome is a governance choice, not a technological inevitability.
Citation
Korea Climate Assembly Wiki. (2026). Agenda #11 — AI and Data Centers. Retrieved from https://climate-assembly.org/en/agenda/ai-datacenter
Disclaimer
This page reflects deliberations of the 2026 Climate Citizens' Assembly, a consultative body established under Article 19-2 of Korea's Carbon Neutrality Framework Act. Recommendations of the Assembly are advisory. This wiki is an independent moderator's archive, not an official publication of any Korean government body.