On March 18, 2025, at the NVIDIA GTC conference, CEO Jensen Huang unveiled the company's roadmap for chip development over the next three years. This roadmap not only outlines significant technological advancements but also highlights NVIDIA's strategic vision for the future of AI and computing.
Future Chip Development Roadmap
2025 H2: NVIDIA will launch the Blackwell Ultra chip, an upgrade to the Blackwell series, offering a 1.5x improvement in AI performance compared to the previous generation. Cloud service providers like Amazon AWS, Google Cloud, Microsoft Azure, and Oracle will be among the first to offer services based on Blackwell Ultra.
2026 H2: The Vera Rubin chip will be introduced, named after the renowned astronomer Vera Rubin. This next-generation GPU will feature 144 independent NVIDIA GPUs.
2027 H2: The Rubin Ultra chip will follow, boasting 576 GPUs and a performance boost of 14x compared to Blackwell Ultra, significantly enhancing AI computing power and efficiency.
2028: NVIDIA plans to introduce a chip named after physicist Richard Feynman, continuing its tradition of naming chips after scientists.
Directions for Chip Development
AI Inference and Efficiency: Huang emphasized that future AI development will focus more on inference capabilities. NVIDIA's chips will support more efficient AI inference, enabling faster deployment of AI models in practical applications. Additionally, NVIDIA launched the open-source software Dynamo to improve AI inference efficiency and reduce costs.
Physical AI and Robotics: The next wave of AI will be "Physical AI," which understands physical concepts such as friction, inertia, and causality. NVIDIA's chips will support the development of robots that can interact with the physical world. For example, NVIDIA introduced the Isaac GR00T N1 open-source foundation model for developing humanoid robots.
Data Center Infrastructure: Huang predicted that by 2028, NVIDIA's data center infrastructure revenue will reach $1 trillion. With the surge in AI demand, data center construction will be a significant direction for future development.
Huang's confidence in future chip development is based on the increasing demand for AI inference and NVIDIA's technological leadership in the AI field.
NVIDIA's announcements at GTC 2025 highlight a future where AI and computing power will continue to advance at an unprecedented pace. As we look ahead, the integration of AI into various industries will drive the need for more powerful and efficient chips. NVIDIA's roadmap not only addresses these needs but also sets the stage for the next generation of technological innovation.