Naver (KRX: 035420) has signed a partnership with NVIDIA to expand its AI infrastructure to the gigawatt scale, the companies announced. The deal taps NVIDIA's DSX software stack to beef up the computing backbone behind Naver's HyperCLOVA X model and its broader sovereign AI push. It's a step that underscores how far cloud and search firms are willing to go to lock down the hardware and software needed for large-scale artificial intelligence.
A gigawatt-scale ambition
The plan targets a level of power consumption rarely seen outside the biggest hyperscale data centers. A gigawatt is roughly the equivalent of two large nuclear power plants, and the figure signals that Naver intends to run thousands of high-end GPUs around the clock. The company didn't say how much it will invest, but building out that kind of capacity typically runs into the billions of dollars.
Naver already operates its own data centers in South Korea and has been expanding its cloud business. The NVIDIA tie-up gives it access to a tightly integrated hardware-software ecosystem that many AI labs rely on. For NVIDIA, the deal locks in a major customer and extends its reach into Asia's fast-growing AI market.
The role of NVIDIA DSX
NVIDIA DSX is a suite of infrastructure software designed to manage GPU clusters, orchestrate workloads, and optimize performance for deep learning. Naver will use it to coordinate training and inference across whatever hardware it deploys. That's important because running AI at scale means dealing with things like job scheduling, networking, and failure recovery — exactly what DSX handles.
By adopting DSX, Naver avoids building those tools from scratch. The company can instead focus its engineering talent on improving HyperCLOVA X and the sovereign AI applications it wants to sell to government and enterprise clients. Sovereign AI — the idea that nations should control their own AI models and data — has become a theme in several countries wary of relying on foreign platforms.
Advancing HyperCLOVA X and sovereign AI
HyperCLOVA X is Naver's large language model, first unveiled last year. The partnership aims to make it more powerful by training it on bigger datasets and with more compute. Naver also plans to offer the infrastructure as a service to other organizations that want to run their own sovereign AI workloads — a product it calls 'Naver Cloud AI.'
The move comes as regulators in South Korea and elsewhere scrutinize how AI models are built and who controls the underlying data. By owning the entire stack — from hardware up to the model — Naver can offer guarantees about data residency and compliance that resold cloud services can't always provide.
Still, building a gigawatt-level data center takes years. Naver hasn't said where it will build the new facilities or when the first phase will go live. It's likely that the rollout will be gradual, with power and hardware coming online in stages as demand for HyperCLOVA X and sovereign AI services grows.
The company has not disclosed a timeline for reaching gigawatt capacity, but the partnership marks a clear scaling commitment.
