Naver's $10B AI factory: where your inference runs
Nvidia is putting $1B into Naver for a 200MW Korean AI factory, with $9B of the envelope still nonbinding. Sovereign AI capacity is real — and it's a 2028 delivery.
Naver confirmed today that it's issuing about 7.24 million new shares to Nvidia — a 4.5% stake, its first third-party share allotment since it listed on the KOSPI in 2008. The money funds a national AI factory in Sejong. The reason this matters to a business in Phoenix and not just to Korean equity analysts: sovereign AI capacity is the thing that decides which countries you can legally and physically run inference in, and it's now being financed like a power plant.
What actually happened
Per Nvidia's announcement, Nvidia plans to invest $1 billion into Naver Corp and Brookfield has signed a nonbinding term sheet to fund up to $9 billion. Naver covers the rest. The buildout expands Naver's Nvidia DSX AI factory at the GAK Sejong hyperscale data center from the 55 MW deployment announced last month to 200 MW by 2028, running Vera Rubin and Blackwell hardware, with a stated long-term ambition of 1 gigawatt.
Read the conditions, because they're printed in the release. Nvidia's $1 billion is contingent on Naver "finalizing at least $9 billion of committed financing" — and the only candidate for that $9 billion is a term sheet that isn't binding. The $10 billion headline is one signed equity check sitting on top of a letter of interest.
Korean press put more detail on the ground: Korea JoongAng Daily reports the site is sized for roughly 100,000 GPUs and that Nvidia becomes Naver's third-largest shareholder, behind the National Pension Service and BlackRock. Naver shares jumped about 10% on the news.
Why sovereign AI capacity matters for your business
Most operators think of "where the model runs" as a latency question. It's turning into a jurisdiction question. Korea is building national inference capacity for the same reason the EU and the Gulf states are: so that regulated data — health records, payroll, government contracts — doesn't have to leave the country to touch a frontier model. If you sell into those markets, or you hold customer data that does, the region your model endpoint resolves to becomes a compliance fact, not a config detail.
The trap is planning around capacity that doesn't exist yet. 200 MW in 2028 does nothing for a workload you're shipping this quarter. Neither does 1 GW "eventually." What you can do now is make the swap cheap when it arrives: keep the base URL, the model name, and the region in environment variables behind one adapter, and know for each dataset whether it's allowed to cross a border. That's a short afternoon of work that turns a future regulatory demand into a deploy.
And when the next $XX-billion AI infrastructure headline lands, check which sentence has a signature under it. In this one, 90% of the number is a term sheet.
Key takeaways
- Nvidia is taking a 4.5% stake in Naver via a ~$1B share issuance — Naver's first third-party allotment since its 2008 listing
- The $10B envelope is $1B of Nvidia equity plus a nonbinding Brookfield term sheet for up to $9B; Nvidia's check is conditioned on Naver locking down that $9B
- GAK Sejong goes from 55 MW to 200 MW by 2028 on Vera Rubin and Blackwell, with a 1 GW long-term target
- Sovereign capacity changes where regulated data can be processed — but 2028 capacity doesn't help a 2026 workload, so build for region portability instead of waiting
If you can't name the region your AI calls terminate in, you can't answer a data-residency question from a customer. We build automation with the model and the region behind an adapter, so moving jurisdictions is a config change and not a rewrite. See how we structure it, or send us your stack and we'll map where your data actually goes.
Sources: NVIDIA Newsroom, Korea JoongAng Daily.
- #ai-infrastructure
- #sovereign-ai
- #data-residency
- #nvidia
- #capacity-planning
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Pilot Protocol's $4.5M: a directory isn't a standard
A seed-stage AI agent network wants to be where agents discover each other. Useful, but know the difference between an open spec and someone else's front door.
Read itNvidia backs Safe Superintelligence, discloses no terms
Nvidia's Safe Superintelligence partnership names no dollar figure and no term length. Reported at $5B. Here's how to read AI deals that omit the numbers.
Read it