Key Takeaways
- Nvidia announced that South Korean internet giant Naver will use its technology to build “AI factories” operating at gigawatt‑scale power consumption.
- The initiative aims to satisfy rapidly expanding global demand for both cloud‑based AI services and emerging “physical AI” applications that embed intelligence in hardware.
- Naver’s involvement signals a strategic shift for the Korean conglomerate toward large‑scale AI infrastructure, leveraging Nvidia’s GPUs, software stack, and networking solutions.
- Gigawatt‑scale facilities could deliver the computational throughput needed for training massive foundation models and running real‑time AI at the edge.
- While the project promises to boost South Korea’s position in the AI ecosystem, it also raises challenges related to power supply, cooling, talent acquisition, and geopolitical supply‑chain risks.
Announcement Overview
On Monday, June 8, Nvidia disclosed that Naver, South Korea’s leading internet conglomerate, would deploy Nvidia’s technology to construct AI factories capable of operating at gigawatt‑scale power levels. The statement, issued via a Reuters wire service, emphasized that the project is designed to meet surging global demand for AI services and the nascent field of physical AI. Although the announcement was brief, it underscores a growing trend where cloud providers and hyperscalers partner with chipmakers to build purpose‑built, ultra‑large‑scale data centers dedicated exclusively to artificial‑intelligence workloads.
What Are AI Factories and Gigawatt Scale?
The term “AI factory” refers to a data‑center facility whose primary function is the large‑scale training, fine‑tuning, and inference of AI models, rather than general‑purpose computing. Gigawatt‑scale denotes a power draw of one billion watts (1 GW), comparable to the electricity consumption of a medium‑sized city or a large nuclear reactor. Achieving this level of power usage requires specialized electrical infrastructure, advanced cooling systems (often liquid‑based), and robust power‑distribution units to ensure stable operation under continuous, high‑intensity workloads.
Naver’s Role and Motivation
Naver, best known for its search engine, messaging platform Line, and a growing portfolio of AI‑driven services, has been investing heavily in AI research through its Naver AI Lab and various acquisitions. By committing to gigawatt‑scale AI factories, Naver seeks to secure sovereign control over the computational resources needed to train its own foundation models—such as HyperCLOVA X—and to offer AI‑as‑a‑service to enterprise customers worldwide. The move also diversifies Naver’s revenue beyond advertising and e‑commerce, positioning it as a direct competitor to global hyperscalers like Amazon Web Services, Google Cloud, and Microsoft Azure in the AI infrastructure market.
Nvidia’s Technology Stack
Nvidia’s contribution will likely encompass its flagship GPUs (such as the H100 Tensor Core series), the NVLink interconnect, the AI Enterprise software suite, and its Spectrum‑X networking platform. These components together enable high‑density GPU clusters capable of delivering exaflop‑level performance while maintaining low latency for model training and inference. Nvidia’s AI Enterprise provides container‑optimized frameworks, tools for model orchestration, and security features essential for multi‑tenant, large‑scale deployments. By leveraging this stack, Naver can accelerate model development cycles, reduce time‑to‑market for AI products, and offer customers a performance‑guaranteed service backed by Nvidia’s proven reliability.
Physical AI Explained
Physical AI refers to the integration of artificial intelligence into tangible systems—robots, autonomous vehicles, smart manufacturing equipment, and Internet‑of‑Things (IoT) devices—where AI algorithms directly control hardware in real time. Unlike cloud‑centric AI, which relies on remote data centers for processing, physical AI demands low‑latency inference at the edge, often requiring specialized AI accelerators co‑located with sensors and actuators. Gigawatt‑scale AI factories can support physical AI by generating the massive models needed for perception, planning, and control, then distributing optimized versions to edge devices through model compression, quantization, and federated learning pipelines.
Market Demand for AI Services
Global demand for AI services has accelerated dramatically, driven by generative AI applications, enterprise automation, and the proliferation of AI‑enabled products. According to IDC, worldwide AI software spending is projected to surpass $300 billion by 2027, with a significant portion allocated to infrastructure that enables model training at scale. The rise of foundation models—some exceeding hundreds of billions of parameters—has intensified the need for compute resources that can sustain weeks‑long training runs without interruption. Physical AI applications, such as autonomous logistics and smart factories, further increase the need for both centralized model generation and decentralized inference, creating a dual‑track demand that gigawatt‑scale AI factories are uniquely positioned to address.
Strategic Implications for South Korea and Global Tech
Naver’s partnership with Nvidia could reinforce South Korea’s ambition to become a hub for AI innovation, complementing existing government initiatives like the “AI National Strategy” and investments in semiconductor fabs. By hosting gigawatt‑scale AI factories domestically, South Korea may reduce reliance on foreign cloud providers for high‑performance AI workloads, thereby enhancing data sovereignty and national security. Globally, the move signals a shift toward more vertically integrated AI ecosystems, where internet conglomerates, chipmakers, and energy providers collaborate to build purpose‑built infrastructure. This could spur similar projects in other regions, intensifying competition for scarce resources such as advanced GPUs, skilled AI engineers, and renewable energy supplies.
Challenges and Considerations
Building and operating gigawatt‑scale AI facilities presents substantial hurdles. Power procurement is paramount; sustaining a 1 GW load requires access to reliable, preferably renewable, electricity sources to meet environmental, social, and governance (ESG) goals. Effective cooling—often via direct‑liquid immersion or advanced air‑flow designs—is essential to prevent thermal throttling and hardware fatigue. Talent acquisition poses another challenge, as designing, maintaining, and optimizing such massive AI infrastructures demands expertise in high‑performance computing, data‑center engineering, and AI systems software. Additionally, geopolitical tensions affecting semiconductor supply chains could impact the timely delivery of Nvidia’s GPUs and related components, necessitating strategic stockpiling or diversification of suppliers.
Outlook and Next Steps
While the announcement did not disclose timelines, capital expenditures, or exact locations for the AI factories, industry observers expect Naver to commence pilot phases within the next 12‑18 months, leveraging existing data‑center campuses in South Korea’s metropolitan areas. Subsequent expansion could involve multiple sites to achieve redundancy and geographic load balancing. Success will hinge on Naver’s ability to secure green power contracts, forge partnerships with local utilities and cooling‑technology providers, and attract top‑tier AI talent. If realized, these gigawatt‑scale AI factories could become a cornerstone of the next wave of AI innovation, delivering the computational horsepower needed to train ever‑larger models and to power the AI‑driven machines that will shape factories, transportation, and daily life in the coming decade.