Naver Cloud: "Even 60,000 GPUs Are Not Enough... AI Bubble Concerns, Value Must Outweigh Costs"

Kim Yuwon, CEO of Naver Cloud, recently commented on the company's agreement to procure 60,000 Blackwell GPUs from Nvidia, stating, "Even 60,000 units are not enough to realize our own large language model (LLM), on-service artificial intelligence (AI), and vertical AI services."

After delivering a presentation at Naver's annual conference 'DAN25' at COEX in Gangnam-gu, Seoul on the 6th, CEO Kim held a press conference and said, "Nvidia GPUs will be used for Naver's AI initiatives," sharing these remarks.

Kim Yuwon, CEO of Naver Cloud, is holding a press conference after delivering a presentation at Naver's annual conference 'DAN25' on the 6th at COEX in Gangnam-gu, Seoul. Photo by Lee Myunghwan

He added, "The exact timeline for the introduction of Nvidia GPUs has not been determined yet," and explained, "Even if 60,000 GPUs were to arrive overnight, we would still need to secure data centers to accommodate them."

Regarding the physical AI collaboration with Nvidia, he said that Naver intends to serve as a kind of 'bridge.' Naver has signed a memorandum of understanding (MOU) with Nvidia to expand the application of physical AI in industrial settings. CEO Kim responded, "We will work with Nvidia in the fields of vertical (specialized) AI and manufacturing AI to build a platform."

On the issue of dependence on Nvidia GPUs, he stated, "Efforts to address the over-reliance on Nvidia chips are a global trend. Many neural processing unit (NPU) developers are challenging themselves to create their own chips, and for this, testing and use must take place in real-world environments, not just in laboratories." He continued, "We are currently testing chips developed by companies like FuriosaAI, and if their performance is good, we plan to implement them. The same applies to chips from Intel or AMD."

Regarding recent claims of an 'AI bubble,' he said, "Ultimately, the value generated must outweigh the costs invested, and what matters is the value created by the technology developed. For such a structure, lightweight versions of AI models are needed. We have been preparing businesses that can deliver value relative to cost through collaboration with enterprises and government agencies."

At DAN25, CEO Kim delivered a keynote speech unveiling Naver's industrial AI strategy, including industry-specific AI and global market expansion. He emphasized, "AI is a core infrastructure that determines industrial competitiveness. Naver, as a company that understands the language, data, and industrial structure of its home country, is building vertical AI models for each industry based on Sovereign AI 2.0."

On the status of Naver's physical AI business, he explained, "Physical AI is technology that allows machines to recognize and understand spaces, make decisions, and act autonomously. Naver's physical AI is evolving through a virtuous cycle involving three elements: accumulated real-world data, a cloud platform, and onboard AI (AI embedded in devices)."