Nvidia used its annual GTC conference in San Jose to announce a sweeping set of new products centered on a single idea: artificial intelligence is shifting from systems that answer questions to systems that act autonomously over long periods of time. Nearly every announcement at the event was designed around “agentic AI”.
The centerpiece was Vera Rubin, a computing platform built from seven chips now in full production. The platform’s flagship configuration, the NVL72 rack, combines 72 Rubin graphics processing units with 36 Vera central processing units. Nvidia claims the system delivers up to ten times more inference output per watt than its previous Blackwell generation. Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure all plan to offer the platform. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman both endorsed the product publicly at the event, according to VentureBeat.
The Vera CPU, which Nvidia describes as purpose-built for agentic workloads, marks a notable strategic turn for the company. As CNBC reports, Nvidia’s graphics chips have driven its rise to become the world’s most valuable publicly traded company, with a market capitalization of 4.4 trillion dollars. But agentic AI requires a different kind of computing: less parallel number-crunching and more sequential task management. Bank of America predicts the CPU market could grow from 27 billion dollars in 2025 to 60 billion dollars by 2030. AMD’s head of data center told CNBC that demand increases over the past six to nine months have been “unprecedented.”
Alongside the large-scale server hardware, Nvidia introduced the DGX Station, a deskside machine designed to run AI models of up to one trillion parameters without cloud infrastructure. It is built around the GB300 Grace Blackwell Ultra Desktop Superchip, delivers 748 gigabytes of coherent memory and 20 petaflops of compute, and supports an air-gapped configuration for regulated industries. Applications developed on the machine are designed to migrate to Nvidia’s data center systems without rewriting code. Early users include Snowflake, Microsoft Research, and Cornell University, according to VentureBeat.
Security was also addressed at launch rather than after the fact. Five vendors announced protections for Nvidia’s agentic stack: CrowdStrike, Palo Alto Networks, Cisco, JFrog, and Worldwide Technology. Nvidia CEO Jensen Huang stated at the event that agentic systems “can access sensitive information, execute code, and communicate externally,” adding that this “can’t possibly be allowed” without controls. VentureBeat notes that cybersecurity professionals rank agentic AI as the top attack vector heading into 2026, and that machine identities now outnumber human employees 82 to 1 in the average enterprise. Analysts caution that agent-to-agent trust, memory integrity, and end-to-end supply chain verification remain unsolved problems.
Additional announcements included Dynamo 1.0, an open-source software system for managing AI inference at scale, already adopted by AWS, Azure, and Google Cloud; the Nemotron Coalition, a group of AI labs that will jointly develop open models on Nvidia infrastructure; and a partnership with Uber to deploy autonomous vehicles across 28 cities by 2028.
Sources: CNBC, VentureBeat, VentureBeat, VentureBeat
