Jensen Huang Declares The Age Of Agents At GTC Taipei

Nvidia is no longer just the company that powers ChatGPT. At GTC Taipei, CEO Jensen Huang declared the dawn of the 'Age of Agents,' announcing that Nvidia's next-generation Vera Rubin system is now in full production and unveiling a trio of new platforms—the Vera CPU, RTX Spark chip, and Cosmos 3—all reframed around autonomous AI agents. The message was clear: the AI industry's next phase is agents that act, not just generate.

Huang took the stage at GTC Taipei to announce that Vera Rubin, Nvidia's successor to the Blackwell architecture, has entered full production. The system is designed for massive-scale AI inference and agentic workflows, marking a strategic pivot from training-focused hardware to systems that can run autonomous AI agents in real time. Alongside Vera Rubin, Nvidia unveiled the Vera CPU, a custom processor optimized for agent orchestration; the RTX Spark, a compact edge chip for deploying AI agents on devices; and Cosmos 3, a platform for building and simulating autonomous agents in physical and virtual environments.

The shift reflects a broader industry trend: AI is moving from passive generation to active agency. Until now, most AI systems have been reactive—answering questions or generating text. Autonomous AI agents, by contrast, can set goals, take actions, and learn from outcomes. Huang argued that the next trillion-dollar opportunity lies in agentic AI, where models operate independently in domains like robotics, customer service, and logistics.

Nvidia's announcement comes at a critical moment. The company has dominated the AI training market with its GPUs, but rivals like AMD and Intel are catching up, and hyperscalers like Amazon and Google are designing their own chips. By betting on autonomous AI agents, Nvidia is creating a new growth vector. The Vera Rubin system is expected to ship to cloud providers later this year, with the RTX Spark available for enterprise edge deployments in early 2027. Cosmos 3 will be released as an open platform for developers.

Industry analysts see this as a savvy move that extends Nvidia's moat into the inference and agent ecosystem. 'Nvidia is not just selling chips—it's selling the infrastructure for the next wave of AI,' said one analyst. However, the agent market remains nascent. Autonomous AI agents still face challenges in reliability, safety, and integration with existing systems.

Looking ahead, Nvidia plans to release a suite of agent development tools and reference architectures at GTC 2027 in San Jose. The company's success will depend on how quickly enterprises adopt agentic AI and whether Nvidia's platforms become the de facto standard. If Huang's bet pays off, the Age of Agents could be as transformative as the Age of AI training that Nvidia helped launch.