Korea's AI Megaproject: Power Tariffs and a Race Against Time
Lawmaker Lee Hae-min of the Rebuilding Korea Party, a former Google employee, argued the government's three-pronged AI megaproject — memory semiconductors, AI data centers, and physical AI — should be viewed as a single interconnected demand chain, since both data centers and physical AI ultimately translate into chip demand. She read Jensen Huang's recent visit to Korea, in which he met not only chip supply-chain partners but also cloud, robotics, and gaming companies, as evidence Nvidia is building new demand vectors ahead of a potential plateau in raw GPU demand.
She rejected the 'government overreach' criticism, describing the approach instead as the state building basic infrastructure — power and water — so that any company, not just Samsung or SK Hynix, could choose to invest, with the actual investment decision left to the companies themselves. She also voiced concern that turning site selection into a political football distracts from the bigger picture: the whole country needs to capture this demand together.
She singled out the dedicated electricity tariff and PPA (direct power-purchase agreements between generators and consumers) as key issues — both provisions had been stripped from the original bill, yet the government's recent announcement revived the dedicated tariff without a clear legal basis, prompting her to formally request supporting documents from the ministry on July 3. She noted estimates that AI data-center demand awaiting a home in Asia runs to roughly 1,000 trillion won, with Australia, Japan, Korea, and Indonesia named as contenders and China excluded for various reasons.
She pointed to comparable state support abroad — U.S. state-level backing that varies widely, Taiwan's essentially all-in government support for TSMC's expansion, and strong state backing in Japan and France for domestic AI infrastructure and sovereign AI — framing the competition as something close to a war, in which Korea must plant its flag again in fields where it has led before, such as semiconductors, home appliances, and shipbuilding.
She cautioned against overreacting to short-term swings, citing episodes where a single Meta comment or a Google optimization paper briefly rattled markets, and warned that as hardware-driven competition gives way to algorithmic optimization, memory demand could see further volatility. That, she said, makes it all the more important to build out AI data centers and physical AI as independent demand pools alongside the semiconductor push.