NVIDIA to Launch GB300 AI Systems in Q3 2025

At Computex 2025, NVIDIA CEO Jensen Huang announced that the company’s next-generation GB300 AI systems will begin rolling out in Q3 2025.

These advanced systems will replace the current Grace Blackwell lineup, now used by cloud leaders like Amazon and Microsoft.

The GB300 continues NVIDIA’s hybrid architecture strategy by tightly integrating CPUs and GPUs for faster AI training and inference.

Alongside GB300, Huang also unveiled the DGX Spark AI workstation and NVLink Fusion — a high-performance interconnect technology that will now be available to other chipmakers.

In a major expansion move, NVIDIA is partnering with Humain, an AI arm of Saudi Arabia’s Public Investment Fund, to build up to 500MW of data centers in the region.

NVIDIA will deliver 18,000 GB300 chips for a supercomputer there, with hundreds of thousands more planned. The collaboration supports “sovereign AI” ambitions and includes NVIDIA’s Omniverse platform for industrial use.

At Computex, Huang also introduced Lepton, a new AI platform designed to link developers with a global marketplace of GPU compute resources.

Lepton connects tens of thousands of GPUs from NVIDIA’s cloud partners—including CoreWeave, Foxconn GMI, Lambda Labs, SoftBank, and others—enabling developers to access regional compute capacity on demand or for long-term needs.