Nvidia's Next-Gen AI Chips Poised to Revolutionize Robotics and Digital Assistance

In a major announcement at the GTC 2025 conference held in San Jose, Nvidia CEO Jensen Huang introduced two groundbreaking AI chips: Rubin Ultra, set for a 2027 release, and Feynman, planned for 2028. These innovations are poised to transform the fields of robotics and digital assistance, leveraging cutting-edge technology to push the boundaries of AI capability.

Breaking New Ground in AI Performance

The Rubin Ultra chip marks a substantial leap forward in AI capabilities, following the highly anticipated Vera Rubin chip expected in late 2026. Named in honor of influential scientists, these chips reflect Nvidia’s dedication to advancing AI technology. The Vera Rubin chip features dual GPUs on a single die, achieving a remarkable 50 petaflops of FP4 inference performance per chip. In a full NVL144 rack configuration, it offers 3.6 exaflops, surpassing the capabilities of its predecessor, the Grace Blackwell.

Rubin Ultra takes this further in 2027, doubling performance potential with its NVL576 rack setup, achieving 100 petaflops per chip. This scale allows for racks providing an extraordinary 15 exaflops of FP4 inference compute and an impressive 5 exaflops of FP8 training performance. These enhancements highlight Nvidia’s commitment to creating highly efficient and powerful technologies, meeting the increasing demands of AI applications.

The Feynman Horizon

Anticipated for 2028, the Feynman chip promises to introduce yet another level of innovation. While specific details are not yet disclosed, it is noted that Feynman will feature a novel architecture centered around the advanced “Vera” CPU instead of the anticipated “Richard.” This strategic decision suggests a bold innovation in computational design, although further specifics remain under wraps.

Nvidia’s Vision: A Future Fueled by AI

Beyond just performance improvements, CEO Jensen Huang envisions a future where data centers operate as “AI factories.” This vision includes developing “physical AI” within humanoid robots capable of executing human-like tasks, all supported by Nvidia’s powerful platforms designed for AI training in simulated environments. Moreover, Nvidia’s chips are expected to power “10 billion digital agents,” revolutionizing the way intelligent systems assist in human activities.

Key Takeaways

Nvidia’s announcement establishes a new benchmark in AI technology with the upcoming Rubin Ultra and Feynman chips. These advancements set the stage for a future rich with potential, advancing both the hardware and software boundaries of AI. As Nvidia refines its strategic focus, it prioritizes scalable and efficient designs, crucial for developing autonomous agents and robots capable of sophisticated operations. The journey to 2028 looks promising, filled with opportunities for groundbreaking developments that are poised to redefine the landscape and capabilities of artificial intelligence.