South Korea boosts AI budget to 9.4T won and 10k GPUs

South Korea is effectively declaring AI a strategic industry on par with semiconductors by putting 9.4 trillion won of the 2027 science budget into AI and related megaprojects. The headline numbers are striking: an 84 percent year‑on‑year increase for AI, 5.6 trillion won for frontier‑class models, and enough capital to secure roughly 10,000 next‑generation Nvidia Vera Rubin GPUs plus a trillion‑token national dataset. That kind of spend gives Korea a credible shot at running its own top‑tier models rather than renting US‑controlled frontier systems.

Equally important is the structure of the budget. Money is not only going into model training and GPUs, but also into an “AI for All” program that will give citizens free access to domestic chatbots and agents, and into physical AI, AI data centers and semiconductor projects. That suggests a whole‑of‑stack strategy: compute, models, applications and hardware. It positions Korean telcos, chipmakers and cloud providers as key intermediaries between state compute and consumer or enterprise use.

For the race to AGI, this accelerates the multi‑polar trend. The more countries that can train and deploy frontier‑class systems on their own infrastructure, the less concentrated technical and governance power becomes, but the harder it is to coordinate global safety norms.