South Korea has committed more than one trillion dollars — combining government funding and corporate commitments — to a "triple axis" of memory chips, AI data centers, and humanoid robots. President Lee Jae Myung framed the three pillars as the country's next industrial leap. The money spans chip fabrication, large-scale data centers in outlying provinces, and a humanoid robot factory targeting 30,000 units per year by 2028.
Key facts
- What: The government and its biggest companies committed more than $1 trillion to memory fabs, AI data centers, and a goal of building tens of thousands of humanoid robots a year by 2028.
- When: 2026-06-29
- Primary source: read the source
South Korea already dominates memory-chip production through Samsung and SK Hynix — the fast storage every AI accelerator depends on. AI's explosive demand made that lead strategically priceless but also exposed how much sits downstream: the models need vast data centers to run, and the country sees physical robots as the next arena where AI leaves the screen and enters factories and warehouses. The plan fortifies the existing strength and builds out the two layers stacked on top of it.
Samsung and SK Hynix are putting roughly $585 billion into new chip fabrication plants, with the stated goal of doubling the nation's DRAM output — the workhorse memory in servers and devices — within five years. A separate roughly $357 billion, from SK Group, GS Group, and the web company Naver, goes into large-scale AI data centers built in outlying provinces. On the robot side, Hyundai is committing $5.8 billion to a robot manufacturing plant and data center, aiming to produce 30,000 Atlas humanoid robots per year by 2028 in partnership with Boston Dynamics, the robotics company it owns. The government has labeled physical AI a "national strategic industry" and wants a homegrown, general-purpose foundation model — built around a world model, the kind of AI that learns how physical environments behave — within three years.
The strategy is to own an entire supply chain rather than one link: the chips, the compute, and the robots that all of them ultimately serve. A country that already makes the world's best engines deciding to also build the car factories and the highways — so that when demand surges, it controls the parts, the assembly, and the roads they run on.
This is sovereign-scale industrial policy pointed squarely at the physical foundation of AI. Model announcements grab headlines, but someone has to build the fabs that make the memory and power the data centers that run the models. A commitment this size shapes global memory pricing, robot supply, and "physical AI" capacity for the rest of the decade — and it's a bet that the next phase of AI is as much about hardware and atoms as it is about software and tokens.
The caveats are substantial. New chip fabs can take up to nine years to come fully online, so the near-term effect on memory prices is genuinely uncertain — this is a decade-long bet, not a quick fix. The resource demands are staggering: the chip plants alone are projected to need around 6.3 gigawatts of electricity and hundreds of thousands of tons of water, with the data centers needing several gigawatts more, raising hard questions about where that power and water come from and how vulnerable the plan is to energy shortages. The human cost is already surfacing: Hyundai's labor union has approved a potential strike over fears that humanoid robots will displace jobs, pushing for profit-sharing and job protections. The politics of replacing workers with the robots you're building is no longer hypothetical — it's a live negotiation.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)