Nvidia CEO Jensen Huang on AI, Korea, and Open Models

Nvidia CEO Jensen Huang recently sat down with Bloomberg to discuss the company's significant investments and partnerships in South Korea, a nation he described as being in a "golden age" for its booming semiconductor and industrial sectors. The conversation, which touched upon everything from High Bandwidth Memory (HBM) development to the philosophical debate around open versus closed AI models, underscored South Korea's pivotal role in the global AI revolution.

Nvidia's South Korean AI Push

Huang revealed several key announcements stemming from his visit, including a substantial business partnership with SK Group, valued at over $500 billion. This collaboration will involve Nvidia purchasing memory from SK Hynix and selling AI supercomputers to SK Group as they scale up their AI factories, which are projected to require 2 gigawatts of power.

Furthermore, Nvidia is investing $1 billion in Naver, South Korea's leading AI cloud provider. Naver plans to scale its AI cloud infrastructure significantly in Korea and expand globally, a move that Nvidia aims to support with its technology.

The full discussion can be found on Bloomberg Podcast's YouTube channel.

The HBM Road Map and Supply Chain Challenges

The conversation also delved into the critical importance of High Bandwidth Memory (HBM) for AI systems. Huang confirmed Nvidia's direct involvement in the road map for future generations of HBM, building on past collaborations with memory makers like SK Hynix. He recalled earlier predictions about the supply chain's ability to keep pace with demand, a challenge that persists.

Huang pointed out that the entire supply chain, not just memory, is facing constraints. "We're constrained in in HBM memories, LPDDR memories. We're constrained in just about every part of the supply chain. We're even constrained now with land and power and construction workers to set up the data centers," he stated. He elaborated that while the industry can likely double capacity annually, scaling much faster is difficult due to the physical infrastructure requirements, which are harder to expand than electronic devices.

The Shifting Landscape of Computing

Huang offered a forward-looking perspective on the computer industry, noting a fundamental shift in how computers are built and used. He explained that while Nvidia continues to build computers for human collaboration with AI, the future will also involve "AI agents and robots. And they're going to be using computers."

This expansion from a billion people using computers to potentially hundreds of billions of agents and robots all requiring computing power means the current semiconductor industry may need to grow significantly. "My guess is that that the semiconductor industry is probably going to have to be 10 times larger than it is today over the next decade or so," Huang predicted. He stressed the importance of working with global partners to scale the supply chain to meet this anticipated demand.

Open vs. Closed AI Models: A Necessary Dichotomy

In a recent public statement on X, Huang joined other tech leaders in advocating for the importance of open models in AI development. He clarified his stance on the open versus closed debate, emphasizing that both have their place.

"I see a future where the world uses tons of closed models. And I I encourage everybody, including my company, to use OpenAI and Claude and Cursor and and Cognition and Perplexity. Use everything that you can because it out of the cloud because it's just easier," Huang said. He added, "And so, in order to build what you must, you need to have open models to do that with."

He outlined reasons why companies might need to build their own AIs using open models, citing proprietary intelligence, regulatory requirements, and national sovereignty. Huang believes that closed models are often more convenient and cost-effective for general use, as they don't require the extensive expertise for fine-tuning, maintenance, and security. However, for specialized use cases requiring control and deep customization, open models are essential.

Huang also addressed concerns about the safety of open models, citing recent incidents where closed models were breached. He argued that neither closed nor open models are inherently safer, and that a diversity of approaches, including open models for self-defense and vulnerability identification, is crucial for the industry's overall security.