Nvidia’s AI Evolution: From Chipmaker to Infrastructure Operator with Samsung and SK Hynix
Nvidia is redefining its role in the technology landscape, evolving from a semiconductor supplier to a comprehensive “AI infrastructure operator” controlling aspects of both intellectual and physical activities. This shift was prominently displayed at GTC 2026, where CEO Jensen Huang unveiled advancements in hardware, software, robotics, graphics, and space computing, signaling a new era of technological dominance.
Disruptive Hardware Innovation: Vera Rubin and HBM4
The centerpiece of Nvidia’s GTC 2026 presentation was ‘Vera Rubin,’ the next-generation AI accelerator architecture, named after the astronomer who proved the existence of dark matter. A key innovation is the integration of a custom-designed ‘Vera CPU’ with the Rubin GPU on a single rack scale. This move breaks away from reliance on Intel or AMD CPUs, mitigating bottlenecks through the use of ‘NVLink 6.0,’ a high-speed, in-house designed data communication network between chips.
Further enhancing performance, the platform incorporates HBM4, the 6th generation of high bandwidth memory. This technology achieves data processing speeds of terabytes per second, reducing large language model (LLM) learning time by one-third whereas improving power efficiency by over 50%. Huang projected a $1 trillion demand for related infrastructure by 2027.
Autonomizing Intelligence: The Rise of Agentic AI
Nvidia formalized the era of ‘Agentic AI,’ which surpasses generative AI by enabling AI systems to set their own goals and utilize tools independently. The core platform, ‘NemoClaw,’ allows AI to move beyond providing simple answers and develop into an “acting agent” capable of executing complex work processes, including tasks in corporate supply chain management and financial analysis. This advancement is expected to maximize human resource productivity and potentially shift software market leadership towards autonomous performers.
Companies can now readily build ‘dedicated agents’ within Nvidia’s ecosystem, capable of learning specific languages or specialized data sets.
Bridging the Physical and Digital: Physical AI, DLSS 5, and Robotics
Nvidia is extending AI’s reach beyond the digital realm into the physical world. The foundation model ‘GR00T N1.7’ will serve as the brain for humanoid robots, while the world model ‘Cosmos’ will enable robots to learn and understand physical laws, allowing them to perform tasks with human-like precision. This is expected to intensify competition in the unmanned manufacturing and logistics industries.
In graphics, ‘DLSS 5’ integrates neural network rendering technology to create real-time lighting effects and character textures with minimal data, resulting in movie-quality virtual worlds with low power consumption. This technology has applications beyond gaming, including industrial digital twin environments.
Expanding the Computing Territory: Space-1 and Space Computing
Nvidia unveiled ‘Space-1,’ a computing module designed for orbital data centers, capable of stable AI calculations even under extreme radiation and temperature conditions. This embodies a vision of a “space cloud” that processes massive data generated during satellite communication and space exploration locally in space, demonstrating Nvidia’s ambition to extend AI beyond Earth.
Implications for the Korean Semiconductor Industry
GTC 2026 highlighted Nvidia’s evolution into a platform power that controls the production and distribution of “intelligence.” Samsung Electronics and SK Hynix face increasing technological competition to lead the supply chain for high-specification HBM4, crucial for Nvidia’s ‘Rubin’ architecture. Industry analysts suggest that as Nvidia’s CUDA ecosystem expands into robotics and space, its reliance on partners will grow, necessitating stronger strategic cooperation from Korean companies at the system integration (SI) level, beyond simple memory supply.
Source: Korea Herald Source: invenglobal.com Source: investing.com Source: Korea JoongAng Daily