devtake.dev

South Korea's $576 billion bet on AI chips runs through Samsung and SK Hynix

South Korea announced a roughly $576 billion (W900 trillion) AI-chip push led by Samsung and SK Hynix, targeting HBM, DRAM fabs, and AI data centers.

Hiro Tanaka · · 4 min read · 4 sources
Close-up of integrated-circuit packages, the kind of memory and logic chips at the center of South Korea's investment plan
Mister rf / CC BY-SA 4.0 via Wikimedia Commons · Source

President Lee Jae Myung put a number on South Korea’s AI ambition this week: roughly $576 billion, about W900 trillion, in chip and AI investment. He announced it on television flanked by the leaders of Samsung Electronics and SK Hynix, the two companies that will write most of the check.

This is industrial policy at national scale, not a single corporate capex line. The plan bundles new memory fabs, expanded HBM and DRAM output, and a buildout of AI data centers into one state-backed program, with the government accelerating permits and the two memory giants supplying the capital. Korea already makes most of the world’s high-bandwidth memory that AI accelerators depend on, and the bet here is to turn that lead into a durable industrial position rather than a one-cycle boom.

What we know

  • The headline figure is roughly $576 billion (about W900 trillion) across chips, AI data centers, and what officials call “physical AI.” Samsung and SK Hynix, with their suppliers, account for about W800 trillion ($517.87 billion) of it, Reuters reported.
  • Samsung Group separately laid out a 10-year program of about W1,000 trillion ($646 billion), with roughly W300 trillion for new fabs in the southwest, W360 trillion for its Yongin cluster, and more than W350 trillion earmarked for AI data centers, per CNBC.
  • The money builds new fabrication sites in the country’s southwest, with SK Hynix tied to Gwangju and Samsung to Asan, plus back-end packaging plants and data centers, the Korea JoongAng Daily reported. A single leading-edge fab runs at least W60 trillion ($40 billion).
  • Industry Minister Kim Jung-kwan said the country aims to double DRAM output within five years, pulling forward fab construction in the Seoul metropolitan area to the mid-2030s.

President Lee framed the stakes bluntly. “We must secure the core elements of AI faster than any other country,” he said. “Semiconductors, physical AI, and AI data centres are the triple axis for our great leap forward.”

What we don’t know

The W900 trillion is a multi-year envelope, and big national capex numbers rarely land in full or on schedule. How much is genuinely new spending versus repackaged existing roadmaps isn’t yet clear from the announcement. The split between government incentives and private money is also fuzzy: the state is leaning on faster permitting and infrastructure more than direct cash, but the exact subsidy mix hasn’t been published.

Two harder questions sit underneath. Power and water are real constraints. Korea Electric Power and Korea Water Resources joined the rollout for a reason, because a fab cluster of this size needs grid and cooling capacity that doesn’t exist yet. And the schedule for the second southern cluster is still open. Presidential policy adviser Kim Yong-beom said demand could force the country to pull facility construction forward “by more than 10 years to 2034-2035,” and added: “we are faced with the challenge of finding a massive new site for a second cluster.”

Who said what

This was an official, on-the-record government announcement, not a leak, so the sourcing is unusually clean. President Lee Jae Myung delivered the framing on television. Industry Minister Kim Jung-kwan supplied the DRAM-doubling target. Presidential policy adviser Kim Yong-beom set out the timeline pressure and the unresolved siting problem in remarks first reported by Reuters. The corporate figures, including Samsung Group’s W1,000 trillion decade-long program, came through company briefings around the event and were carried by CNBC and Reuters.

What this means for you

If you build, buy, or budget for AI hardware, the memory bottleneck just got a long-term answer that is concentrated in one country. Samsung and SK Hynix already supply roughly 80% of global HBM, the memory that sits next to every AI accelerator, and this plan deepens that concentration rather than spreading it. More capacity over the next five to ten years should ease the supply crunch that has been pushing memory prices up across servers, laptops, and phones. We covered how that crunch is already forcing consumer price hikes; a doubling of DRAM output is the supply-side counterweight. The catch is timing and geography. Fabs take years, the power and water are not built yet, and so much of the world’s AI memory riding on two firms in one place is a supply chain with a single point of stress. Watch the first fab construction starts in the southwest for whether the W900 trillion turns into concrete or stays a headline.

Share this article

Quick reference

HBM
High-bandwidth memory, stacks of DRAM bonded together for huge throughput. It's the memory AI accelerators like Nvidia's GPUs depend on, and the supply now competes with phones and PCs.
DRAM
Dynamic random-access memory, the fast working memory in phones, laptops, and servers. It's volatile, so it loses its contents when power cuts, and it's made by Samsung, SK Hynix, and Micron.

Sources

Mentioned in this article