What new GPU architecture and AI chip did Nvidia unveil in March 2026?

In March 2026, Nvidia put a significant spotlight on its Blackwell GPU architecture, which serves as the foundation (Review) for the company's latest advancements in AI chips and accelerated computing. While the Blackwell architecture itself was initially introduced earlier, March 2026 saw the unveiling of new products and systems built upon this powerful design, cementing its role in the next generation of artificial intelligence. These announcements at events like GTC 2026 showcased how Blackwell-based technologies are pushing the boundaries for AI training and inference.

Blackwell is a successor to the Hopper and Ada Lovelace microarchitectures, engineered specifically to tackle the escalating demands of large-scale AI models. It boasts an impressive 208 billion transistors and features a revolutionary dual-die design, allowing it to circumvent traditional manufacturing limitations and effectively create a unified, massive GPU. This architecture integrates fifth-generation Tensor Cores and advanced NVLink interconnects, which are crucial for accelerating the complex computations involved in training and deploying trillion-parameter AI models and enabling real-time inference.

During the March 2026 events, Nvidia and its partners unveiled new Blackwell-based AI chips and systems, including the Blackwell Ultra GB300 GPU and RTX PRO Blackwell GPUs for workstations. These chips are designed to power everything from individual AI workstations to massive data center infrastructures, offering substantial performance improvements and enhanced energy efficiency for generative AI workloads. The Blackwell platform, with its robust chips and interconnectivity, is positioned as the engine for AI factories, demonstrating Nvidia's commitment to advancing the intelligence age.