Nvidia’s AI Evolution: From Chipmaker to Infrastructure Operator – GTC 2026 Highlights

SAN JOSE, CA – Nvidia CEO Jensen Huang unveiled a sweeping vision of the company’s future at GTC 2026, signaling a shift from hardware vendor to a comprehensive “AI infrastructure operator,” according to remarks delivered Tuesday at the SAP Center. The announcement centered on the “Vera Rubin” AI accelerator architecture, a new CPU-GPU integrated platform, and a suite of software and hardware innovations extending into robotics and even space computing.

Huang detailed the “Vera Rubin” platform, named after the astronomer who confirmed the existence of dark matter, as a culmination of efforts to optimize computing resources. A key element of the new architecture is the “Vera” CPU, designed in-house by Nvidia, and integrated with the Rubin GPU on a single rack scale. This represents a departure from the company’s previous reliance on CPUs from Intel and AMD, and aims to eliminate data transfer bottlenecks through the use of “NVLink 6.0,” a proprietary high-speed interconnect.

The platform also incorporates sixth-generation high bandwidth memory (HBM4), enabling data processing speeds measured in terabytes per second. Nvidia claims this will reduce the training time for large language models (LLMs) by two-thirds while improving power efficiency by more than 50%. Huang projected a $1 trillion market demand for related infrastructure by 2027.

Beyond hardware, Nvidia formally introduced the era of “Agentic AI” with the “NemoClaw” platform. This software is designed to enable AI to move beyond simply responding to queries and instead autonomously set goals and execute tasks. The company envisions AI agents handling complex processes such as supply chain management and financial analysis, potentially maximizing human productivity.

Nvidia’s ambitions extend beyond the digital realm with advancements in “physical AI.” The company unveiled “GR00T N1.7,” a foundation model intended to serve as the brain for humanoid robots, and “Cosmos,” a world model designed to allow robots to learn and interact with the physical world. These developments are expected to accelerate automation in manufacturing and logistics.

Improvements to graphics technology were also highlighted, with the introduction of “DLSS 5,” a neural rendering technology capable of generating real-time lighting effects and textures with minimal power consumption. This technology has applications in gaming and industrial digital twin environments.

Perhaps the most unexpected announcement was “Space-1,” a computing module designed for orbital data centers. This module is engineered to withstand the harsh conditions of space, including extreme radiation and temperature fluctuations, and will enable real-time processing of data generated by satellites and space exploration missions.

The announcements at GTC 2026 position Nvidia as a dominant force in the emerging AI infrastructure landscape, raising challenges for companies like Samsung and SK Hynix. These Korean semiconductor manufacturers are now competing to secure a leading role in the supply chain for high-performance HBM4, a critical component of Nvidia’s “Rubin” architecture. Industry analysts suggest that the continued expansion of Nvidia’s proprietary CUDA ecosystem into robotics and space computing will likely increase the dependence of partner companies.

The implications for the Korean semiconductor industry are significant, with a need to move beyond simple memory supply and towards strategic collaboration in system integration, according to industry observers.