On Monday, the annual NVIDIA GTC 2026 conference kicked off in San Jose — the largest event in the world of GPU technologies and artificial intelligence. Opening the event, Jensen Huang announced that the company forecasts hardware orders totaling at least one trillion dollars by the end of 2027. This is twice as much as last year and indicates strong industry expectations. According to him, the industry has reached an important milestone: now models can be trained, and the only remaining challenge is to deploy them quickly and cost-effectively.
The main focus is on the Vera Rubin presentation. The new platform features a powerful GPU based on TSMC’s 3nm process, equipped with 336 billion transistors and 288 GB of HBM4 memory. The pinnacle of this system is the NVL72 rack with NVLink bandwidth reaching 260 TB/s, which NVIDIA CEO compared to the total internet bandwidth! The inference performance of the new platform has increased fivefold compared to its predecessor Blackwell, thanks to the NVFP4 format.
Vera Rubin is not just a graphics card but an entire ecosystem. Along with it, NVIDIA announced its own CPU, Vera, designed for memory management and orchestration of large-scale agent systems. Plans are already in place for three subsequent generations: Vera Ultra will be released in the second half of 2027, and the next architecture, codenamed Feynman, in 2028.
NVIDIA is also developing NemoClaw — a new open-source platform for creating and deploying enterprise AI agents. It runs locally on NVIDIA hardware without needing cloud connectivity and allows users to set individual parameters for agents: name, personality, and tools. This is a direct response to OpenClaw and part of NVIDIA’s strategy to establish a presence not only as a hardware provider but also in the software layer.
For PCs, NVIDIA introduced N1X — an ARM processor co-developed with MediaTek. It features 20 custom ARM cores and an integrated GPU comparable to RTX 5070, aimed at laptops and workstations with local inference capabilities.
Special attention is given to DLSS 5 — a new neural network rendering technology that NVIDIA considers a breakthrough following the introduction of ray tracing in 2018. This system uses a new real-time neural rendering model to produce photorealistic lighting and materials in scenes. The release is scheduled for fall this year, with early partners including Ubisoft, Bethesda, Capcom, Tencent, and Warner Bros. Games.
Additionally, a major deal was announced — a strategic partnership with Thinking Machines Lab: the company will create at least one gigawatt of Vera Rubin systems for training AI models. This marks the first large-scale deployment of the platform of this magnitude.
The conference will run until March 19 and promises over a thousand thematic sessions covering robotics, physical AI, and open models.
For those interested in watching the livestream, a recording is already available on YouTube.
Update your knowledge with leading resources in machine learning!
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby