On March 18, Nvidia's CEO Jensen Huang presented the company's latest chips at its annual software developer conference, aiming to reassure investors of its continued leadership in the fast-evolving artificial intelligence industry.
Nvidia unveiled the upcoming graphics processing unit (GPU) named Blackwell Ultra, with enhanced memory to support larger AI models, set for release in the second half of this year.
The new Rubin chips and servers promise faster data transfers between chips, vital for efficient operation in extensive AI systems.
The Vera Rubin computing system, incorporating Nvidia's custom-designed processor Vera, is expected to surpass the current Blackwell architecture. Vera Rubin is scheduled for release in the second half of 2026, followed by Vera Rubin Ultra in 2027 and the Feynman architecture in 2028.
Nvidia announced DGX AI computers, featuring the Blackwell Ultra chips, to aid developers in running large models on desktops, produced by companies like Dell, Lenovo, and HP. It poses a direct challenge to high-end Macs from Apple.
Nvidia's new silicon photonics networking chips aim to connect millions of GPUs across various sites in AI factories, significantly reducing energy consumption. Quantum-X Photonics chips are planned for release this year, with Spectrum-X chips following in 2026.
Nvidia also introduced free software to accelerate reasoning processes in AI models and GR00T N1, a foundational model for humanoid robots, with a dual system for rapid and deliberative thinking. The robot's framework includes Newton, an open-source physics engine created in collaboration with Google DeepMind and Disney Research, tailored for robot development.