South Korea's New Supercomputer Lets Small Businesses Rent Serious AI Power

South Korea is about to switch on a supercomputer called Hangang. It can perform 614 quadrillion calculations every second — roughly 24 times faster than the country's previous national supercomputer, with 10 times the storage. It packs 8,496 Nvidia AI chips and is expected to rank among the world's ten most powerful machines. It opens for official service in December.

So what? Normally, running serious AI research or huge simulations means buying your own GPUs (the pricey chips that power AI) or renting them from a cloud provider. Both get expensive, and both raise questions about who can see your data. Hangang works more like a shared public utility: the government owns the hardware and hands out slices of it. A small company can start with a single node — four AI accelerators — without spending millions.

The catch: access is limited and competitive. About 55% of the machine is reserved for research and innovation, 30% for national strategy projects, and only 15% for industry. Pilot applications run September 23 to October 8; official applications run September 28 to October 16, with access beginning in December. You don't just log in — you apply, and the science ministry decides who gets how much.

Why it matters beyond Korea: it shows a model where governments buy the expensive AI hardware and lend it out, rather than leaving it to a handful of tech giants. If it works, other countries — and their small businesses — may copy it. For now, it's a reminder that the raw material of the AI boom is computing power, and whoever controls the supply shapes who gets to build.