NVIDIA Unleashes Blackwell: Next-Gen AI GPU Samples Ship Worldwide

NVIDIA CEO Jensen Huang announced that they will begin distributing engineering samples of the Blackwell GPU, designed for AI applications, worldwide this week. NVIDIA plans to bring products based on the Blackwell architecture to market in the fourth quarter of this year.

Huang made this announcement at Siggraph, noting that NVIDIA’s hardware partners, including Foxconn, Quanta, Wistron, Pegatron, and ASUS, had already showcased servers based on the Blackwell architecture at this year’s Computex in Taipei. This suggests that these partners have had access to the Blackwell GPU for some time. However, not all software developers currently have access to NVIDIA’s latest Blackwell GPU.

NVIDIA and its partners will charge a premium for each AI server rack equipped with Blackwell GPUs. NVIDIA has introduced two reference server racks: the NVL36, equipped with 36 B200 GPUs, priced at approximately $2 million, and the NVL72, equipped with 72 B200 GPUs, starting at $3 million.

These server racks will be provided by NVIDIA and its traditional partners, such as Foxconn, Quanta, and Wistron, as well as new entrants like ASUS. NVIDIA is expected to ship 60,000 to 70,000 B200 server racks next year, generating at least $210 billion in revenue. Major companies like AWS, Dell, Google, Meta, and Microsoft are anticipated to adopt NVIDIA’s latest Blackwell GPU.