GTC 2026: NVIDIA reveals next generation AI chips and platform for robotics

NVIDIA started the GTC 2026 conference on Monday in San Jose, California, with CEO Jensen Huang's keynote full of news. Among the highlights are the new generation of chips for training language models and a unified platform for robotics with physical artificial intelligence. Ads reinforce the company's position as the backbone of the global AI ecosystem.

NVIDIA announcements on GTC 2026 keynote

Jensen Huang presented the chip NVIDIA Vera, successor of Blackwell architecture, with performance gains of up to 3x in inference and 2.5x in training large models. The novelty is accompanied by DGX Vera, a complete system that integrates 72 GPUs into a single blade, designed for hyperscale data centers. Huang also demonstrated the NVIDIA Cosmos 4, an open source physical AI model that leads robotic simulation and autonomous driving rankings — an evolution of NVIDIA Cosmos 3.

On the software side, the company launched the NVIDIA AI Enterprise 5.0, with new agents ready for use in sectors such as health, manufacturing and finance. Integration with humanoid robots was highlighted through the Isaac GR00T platform, which now supports real-time reinforcement learning directly in factories.

Impact of NVIDIA’s announcements on the AI market

For Brazilian developers, the main impact is on democratizing access to cutting-edge hardware. DGX Vera will be available via cloud from the second quarter, allowing startups to train competitive models without millionaire investment. The Cosmos 4 platform, being open source, also opens doors for academic research in robotics in the country, an area where Brazil seeks advances in precision agriculture and industrial automation.

In addition, the partnership with Foxconn, announced earlier, is already leading agentic and robotic AI to hospitals in Taiwan, signaling a path for medical applications that may reach Brazil soon.

Expectations for NVIDIA after GTC 2026

The roadmap presented indicates that NVIDIA is doubling the bet on A physical — i.e. systems that interact with the real world through robots and autonomous vehicles. The forecast is that by 2028 the AI robotics market exceeds $100 billion, and the company wants to be the dominant supplier of chips and development platforms. In addition, Jensen confirmed that the post-Vera architecture, codenamed Rubin, is already in development and should arrive in 2028.

For those who followed the event live, the message was clear: the AI race is just beginning, and NVIDIA maintains the pace of innovation that consolidated it as a reference. More details about all ads can be found in the official NVIDIA blog.

Frequently Asked Questions about GTC 2026

When the NVIDIA chip Is Vera available?

The first systems with DGX Vera begins to be sent to selected customers in the third quarter of 2026, with general cloud availability in the fourth quarter.

Is NVIDIA Cosmos 4 free?

Yes, the Cosmos 4 model is distributed under open source license, allowing commercial and academic use without licensing costs.

What differentiates GTC 2026 from previous editions?

The focus on physical and robotic AI was the great differential, with practical demonstrations of humanoid robots operating in real time using the Isaac GR00T platform.