South Korea just announced it will build more AI data center capacity in the next decade than most countries have plugged in since the internet began.
The Summary
- SK Telecom plans to deploy 15 gigawatts of AI data center capacity by 2035, partnering with Nvidia on gigawatt-scale infrastructure across Asia
- Naver secured $1B from Nvidia and up to $9B from Brookfield for separate gigawatt-scale facilities, with full deployment targeted by decade's end
- The broader SK Group and Nvidia partnership totals $500B, featuring next-gen memory chips and infrastructure that could reshape global compute access
- South Korea is building sovereign AI infrastructure at national scale, betting that whoever controls the chips controls the agent economy
The Signal
Fifteen gigawatts is not a typo. For context, the entire US data center industry currently runs on roughly 17 GW. SK Telecom's buildout plan would nearly match that in a single country, for a single use case, by 2035. This is not expansion. This is the foundation for a parallel compute economy.
The timing matters. Naver's separate $10B funding round from Nvidia and Brookfield Infrastructure signals that multiple South Korean players are moving simultaneously. Naver is the search giant. SK Telecom is the telco. Different companies, same thesis: AI inference at the edge requires density that nobody has built yet.
"South Korea is treating AI infrastructure like it treated semiconductor manufacturing in the 1980s: as a national imperative, not a private bet."
The SK Group and Nvidia partnership goes beyond just racks and cooling towers. Next-gen memory chips are the tell. Training models is compute-intensive but batch-oriented. Inference, running millions of agent requests per second, is memory-bandwidth constrained. If SK is co-developing memory with Nvidia, they are building for a world where agents outnumber API calls 100 to 1.
Three implications stand out:
- Geographic arbitrage ends: If Seoul has the same inference capacity as Virginia, latency becomes the only moat. Agents will run where users sleep.
- Crypto mining pivots: The crypto angle here is not coincidental. Gigawatt-scale power infrastructure built for AI can absorb Bitcoin mining during off-peak demand, turning data centers into two-sided energy markets.
- Sovereign compute matters: China restricts chip exports. The US restricts chip access. South Korea is building a third pole, and Brookfield's $9B commitment to Naver suggests Western capital sees the geopolitical hedge.
The broader $500B figure likely includes chip manufacturing, energy infrastructure, and real estate. That is not just data centers. That is vertical integration from silicon to container. If you are building 15 GW of AI capacity, you cannot rely on someone else's supply chain. You become the supply chain.
The Implication
Watch who else announces gigawatt-scale plans in the next six months. SK and Naver are not lone actors. They are early movers in a race that most countries have not realized they are in yet. The agent economy will not run on cloud providers alone. It will run on national infrastructure plays disguised as corporate capex.
For builders: if you are deploying agents that need sub-50ms inference globally, your infrastructure assumptions just changed. The compute will be there. Plan accordingly.