Nvidia’s $1 Billion Naver Investment Still Requires $9 Billion in Project Financing

Haejin Lee, Naver’s founder and board chairman, presented the arrangement at a San Francisco AI summit attended by South Korean President Lee Jae Myung. Lee said the partners’ capital, brands and networks could help scale Naver’s operating know-how, as described in an account of his remarks. In a separate report, Lee framed the project as part of his argument for multiple AI systems rather than AI controlled by one or two groups. Those are Naver’s stated aims, not evidence that the service has yet achieved global scale.

Why Naver is the operator in the proposed structure

Naver, founded in 1999, is Korea’s largest internet company. Its businesses include the country’s leading search engine, commerce, fintech, cloud, AI and robotics. It also says it has designed and operated data centers to support internet services developed over roughly three decades. That operating record explains its role as the intended operator, while Nvidia supplies the accelerated-computing platform and Brookfield is the prospective capital and infrastructure partner.

GAK Sejong is Naver’s hyperscale data center. Nvidia says the expanded site is expected to use its Vera Rubin and Blackwell platforms and its DSX stack of chips, systems, software, facilities and partner technologies. The company says the infrastructure is intended to give Korean and U.S. developers production-scale compute for models, agents and AI-powered services. Its claims that DSX lowers token costs and speeds production are product claims, not an independent cost comparison with rival platforms.

Naver is also extending the partnership into software and models. It says it is advancing HyperCLOVA X using Nvidia Nemotron 3 Ultra open models with Naver’s proprietary data and training expertise, plans a Korean AI-agent platform in the second half of the year, and is developing a Seoul World Model from its street-view and spatial-modeling data. These are development and launch plans; the announcement provides no uptake, performance or pricing evidence.

A national program sets a much larger comparison

The government says Korean and global companies are pursuing about 5 gigawatts of AI-data-center capacity and roughly 2 million B200-equivalent GPUs across related projects. The 200MW Naver target is therefore 4% of that announced capacity total; its stated 1GW path would be 20%. Those are comparisons among proposed or pursued capacity, not operating sites, and the 1GW path has no announced completion date.

Naver is not the sole route to that capacity. The government briefing says SK Telecom agreed with Nvidia on support for data-center construction and expansion up to 2GW, including priority allocation of Vera Rubin systems; it is also pursuing gigawatt-scale projects with Anthropic and expanding an existing Ulsan partnership with AWS. The same program includes a Ministry of Science and ICT agreement with AMD on infrastructure linking Korean NPUs with AMD CPUs and GPUs.

That wider context cuts against treating the Naver project as Korea’s AI-infrastructure strategy on its own. Its significance is the proposed combination of a domestic internet and cloud operator, Nvidia’s platform and Brookfield’s infrastructure capital—not a completed monopoly over national capacity.

What would turn the proposal into capacity

The next decisive evidence is concrete: Naver must disclose committed financing that satisfies Nvidia’s condition, Brookfield must move from a nonbinding term sheet to definitive arrangements, and the partners must show that power, equipment and construction can support the 200MW target by 2028.

The eventual test is commercial as well as physical. The companies need to demonstrate that the multi-tenant service can attract customers and that Naver’s model and agent plans can compete alongside other Korean projects and global cloud alternatives. Until financing closes and capacity enters service, the $10 billion figure remains an announced plan with material execution risk.