America Wants Tech to Pay for AI Power. Korea Wants Tax Breaks.

The Divide That Could Define the Next AI Decade

America is making Big Tech pay for the electricity infrastructure powering artificial intelligence. South Korea is doing the exact opposite.

The contrast could not be sharper, and it reveals a fundamental disagreement about who should bear the cost of the AI revolution — a question that will reshape energy policy, technology competition, and the relationship between Seoul and Washington over the next decade.

In March 2026, President Donald Trump hosted Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI executives at the White House for the “Ratepayer Protection Pledge.” The agreement was unambiguous: companies building AI data centers must pay for all new power generation, transmission upgrades, and grid infrastructure — even for electricity they never actually use. Twenty-three state governors and 305 organizations have since signed on.

The backlash from ordinary Americans forced the hand. PJM Interconnection, the grid operator serving Virginia, Pennsylvania, and neighboring states, calculated that data centers had driven up electricity costs for other customers by $23 billion in a single year. Virginia saw residential rates climb 13 percent, Illinois 16 percent, New Jersey 21 percent — far above the national average of 6.1 percent. Local governments fought back. Texas Governor Greg Abbott ordered ERCOT to audit every data center seeking grid access. Virginia became the first state to impose a consumption tax on data center electricity at $0.011 per kilowatt-hour.

Korea’s Mirror Image

Meanwhile, South Korea is racing toward the same problems with its eyes wide open — or perhaps deliberately closed.

In June 2026, President Lee Jae-myung declared AI data centers one of three “mega projects” critical to national advancement, insisting that “only speed tactics can save us.” Within weeks, Climate, Energy and Environment Minister Kim Seong-hwan reported that a dedicated electricity rate system for AI data centers was under review — essentially a subsidized tariff designed to make Korea more competitive in attracting foreign operators.

The policy posture inverted America’s entire framework. Where Washington has moved to internalize the true cost of AI infrastructure, Seoul is moving to externalize it — onto the national grid, onto ratepayers, onto future generations.

At a July 16 parliamentary forum on AI infrastructure, Yoon Eun-seong from SK Telecom’s AI Policy Research Institute made the country’s priorities bluntly clear. SK is pushing for accelerated permitting, government funding for power infrastructure construction, and — most significantly — tax deductions for what it described as AI data center “leasing businesses.”

Here lies the structural twist that most English-language coverage misses: the vast majority of South Korea’s proposed AI data centers are not designed to serve domestic AI development. They are colocation facilities built to house American Big Tech’s infrastructure on Korean soil.

Yoon acknowledged this directly. SK’s planned 5-gigawatt and 15-gigawatt facilities were conceived specifically to attract overseas hyperscalers. Without foreign tenants, the capacity cannot be absorbed domestically. “This is a facility that actively produces tokens to meet their demand,” Yoon said, arguing that current tax law treating data centers as passive real estate was misaligned with economic reality.

Green Transition Research Institute and partner organizations calculated that 88 percent of the phase-one AI data centers announced under the mega project would operate on a colocation basis — essentially foreign data centers running on Korean power, with Korean infrastructure costs socialized.

The Speed Trap

n
The logic driving South Korea’s position is internally coherent, even if the outcome is dangerous. Bank of America projects global cloud AI infrastructure investment could reach $1.5 trillion next year. US communities are actively blocking new data center construction — Brookings Institution data shows 75 projects totaling $130 billion stalled in the first quarter of 2026 alone over local opposition. If America is rejecting AI infrastructure, the argument goes, South Korea should be rushing to fill the vacuum.

Deputy Secretary of the Presidential Policy Office Kim Yong-beom referenced this calculus after the San Francisco AI Summit in late July, noting that SK Telecom is advancing a gigawatt-scale data center partnership with Anthropic while expanding its AWS project in Ulsan.

But speed without scrutiny carries its own costs. The data centers that have sparked community conflict in Korea so far — a Hyosung project in Anyang halted in 2023, a Kimpo facility that faced resident lawsuits before breaking ground in 2024 — have all been under 100 megawatts. The mega projects being pushed through parliament are orders of magnitude larger.

The pattern in US states with concentrated data center development is instructive: when scale hits a certain threshold, the conflicts expand beyond noise and electromagnetic concerns into water usage, grid stability, and regional energy poverty. None of these dynamics have been adequately studied in the Korean context, despite the Ministry being asked to move faster than standard environmental review would allow.

Who Wins, Who Loses

The winners in South Korea’s current trajectory are clear: SK Group and other domestic conglomerates positioned to build and lease colocation facilities, American hyperscalers who get premium infrastructure without bearing full cost, and local governments outside the capital region that see investment promises. Seventy-one out of 77 local election candidates who campaigned on data center attraction were from non-metropolitan districts — a geographic signal of where the political benefits concentrate.

The losers are harder to name precisely because the harm is distributed and deferred. Korean ratepayers who will absorb the gap between subsidized industrial tariffs and the actual cost of grid expansion. Communities near future mega-sites who will face the same water, noise, and reliability conflicts now playing out in Virginia and Texas — only without the regulatory guardrails that conflict has already produced in America. Future taxpayers who will foot the bill for power plants and transmission lines built to serve facilities that generate profits for foreign corporations.

The Geopolitical Fault Line

The policy divergence matters beyond Korea’s borders. If South Korea succeeds in building a regulatory arbitrage zone — where AI infrastructure can be deployed faster and cheaper precisely because the external costs are not priced in — it creates a perverse incentive structure. Other countries will be pressured to follow suit to remain competitive, triggering a race to the bottom on energy and environmental standards.

Conversely, the American model of forcing cost internalization could slow deployment but build more durable infrastructure foundations. Both approaches have trade-offs that will play out over years, not quarters.

What remains striking is how little structured debate exists in Korea about the trade-offs. The government has exempted even the minimal grid impact assessments typically required for large power facilities. The dedicated electricity rate system is still in early review stages, with no published criteria for how subsidized rates would differ from standard industrial tariffs or how the revenue gap would be covered.

Energy Transition Forum, an independent policy group, issued a statement demanding that new grid investment costs and project risks fall on the party causing them — a direct rebuke to the current policy direction. The government has yet to respond substantively.

The question Korea faces is not whether it should pursue AI infrastructure. It is whether the country will force the builders to pay for what they consume, or whether it will export the subsidy burden to its own citizens while the profits flow overseas. America is answering that question by making Big Tech pay. Korea is answering by asking its ratepayers to subsidize someone else’s infrastructure — and calling it strategy.