Naver, Nvidia to build gigawatt-scale AI factory network spanning Asia, Middle East and Europe

55MW operation to launch in 2027, scaling to gigawatt-class infrastructure

Multi-site strategy targets AI infrastructure markets across Asia, Middle East and Europe

DSX, Cosmos and Nemotron to anchor expanded infrastructure, physical AI and model cooperation

Nvidia CEO Jensen Huang, far right, and Naver Chairman Lee Hae-jin, far left, meet at "Hyungnim Jeoyo," a samgyeopsal restaurant near Hongdae in Mapo-gu, Seoul, on Sunday. [Yonhap]

Naver is teaming up with Nvidia to enter the gigawatt-scale AI infrastructure race dominated by global technology giants.

The two companies announced Monday they will jointly build a global AI factory network more than four times the size of Gak Sejong, South Korea's largest hyperscale data center. The partnership is designed to extend their AI infrastructure alliance from the Asia-Pacific region into the Middle East and Europe.

A 1-gigawatt facility can accommodate hundreds of thousands of the latest Nvidia GPUs. Industry observers say the deal elevates Naver from a simple GPU customer to a core global AI infrastructure partner for Nvidia.

Naver Chairman Lee Hae-jin and Nvidia CEO Jensen Huang met Monday at Naver's second headquarters, 1784, and agreed to jointly develop gigawatt-class AI factory infrastructure.

The partnership goes beyond procuring GPUs for Naver's own services. The two companies will jointly identify global AI demand, build infrastructure and advance technology together — a fully integrated model targeting the global AI infrastructure market. Lee and Huang agreed to expand their foothold in the AI infrastructure ecosystem starting in the Asia-Pacific region and extending to the Middle East and Europe.

Naver Chairman Lee Hae-jin, left, Nvidia CEO Jensen Huang, center, and Naver CEO Choi Su-yeon meet at Nvidia's headquarters in California in June 2024. [Naver]

Under the agreement, Naver will use Gak Sejong as the anchor site for its global AI factory operations. The plan calls for a 55-megawatt launch in the first half of 2027, scaling to 100MW by the end of that year and 200MW by 2028. Naver will then expand overseas sites in stages, with the ultimate goal of reaching gigawatt-class capacity — roughly four times the scale of Gak Sejong, according to the company.

"This will not be a single campus cluster but a distributed multi-site model that eventually reaches gigawatt-class infrastructure," a Naver Cloud official said. "The final scale has not yet been fixed." The official indicated the project could grow to 2GW or 3GW depending on market demand and ongoing discussions with Nvidia.

Industry observers say the deal marks Naver's full entry into the league of hyperscale AI infrastructure builders previously led by global technology majors. Publicly announced gigawatt-scale AI infrastructure projects include the Stargate initiative — backed by OpenAI, SoftBank and Oracle — targeting 10GW, Meta's Hyperion project at 5GW, and an AWS facility in Indiana at 2.2GW. While Naver and Nvidia have yet to set a final capacity target, their formal commitment to gigawatt-class infrastructure places them squarely in that competition.

Technical cooperation between the two companies will also broaden significantly. Naver's accumulated expertise in building and operating large-scale GPU clusters and hyperscale data centers will be combined with Nvidia's next-generation high-performance infrastructure platform, DSX. The aim is to improve data center operational efficiency and build the infrastructure capacity to handle large-scale AI workloads reliably.

A server room inside Naver's Gak Sejong data center. [Naver]

Cooperation in physical AI and spatial intelligence will also move forward. Naver plans to combine Nvidia's world foundation model Cosmos with its own spatial modeling technology and street-view data to build a "Seoul World Model" — a next-generation initiative that would allow AI to understand and simulate real urban environments based on actual city data.

The two companies will extend their collaboration to AI models as well. Naver recently became the first South Korean company to join the Nvidia Nemotron Alliance. It plans to integrate its own data and training expertise with joint Nemotron development work to accelerate performance upgrades for HyperCLOVA X and broaden its global applicability.

"This alliance allows us to offer a concrete alternative for regions and countries around the world to build their own sovereign AI capabilities, which is very encouraging," Lee said. "This partnership is deeply meaningful because it creates an opportunity for Naver's technology infrastructure strengths to take a major leap into the global market."

rim@heraldcorp.com