Nvidia’s upcoming Blackwell B100 and B200 artificial intelligence chips are expected to range in cost from $30,000 to $70,000 each, with server costs being expressed in millions of dollars. It is anticipated that Nvidia’s Blackwell products, which include GPUs, Superchip platforms, and servers, will generate significantly more revenue than the current Hopper series. According to HSBC analysts, the unit cost of Nvidia’s Blackwell Superchip solutions for AI applications will range from $30,000 to $70,000.
Reportedly, the cost of a single Blackwell GB200 Superchip (CPU+GPU) solution will be around $70,000. However, companies typically do not purchase these hardware components individually, and Nvidia generally does not make single-unit sales. Instead, the AI giant will reach customers through B200 NVL72 (36 CPU and 72 GPU) servers, each of which is expected to cost $3,000,000. It is noted that the GB200 NVL36 variant will cost around $1,800,000.
The average selling price of Nvidia’s new entry-level Blackwell B100 GPU is expected to range from $30,000 to $35,000, which aligns with the current H100. Furthermore, it is reported that the more powerful GB200, which combines a single Grace CPU with two B200 GPUs, will be priced between $60,000 and $70,000.
However, these figures are merely analyst estimates. Nvidia’s pricing may vary depending on the customer and could be higher. Nvidia highlighted the NVL72 server when introducing Blackwell solutions because this server unit aims to encompass the entire setup. It serves a massive GPU function integrated with high-bandwidth connections and provides a total of 13,824 GB of VRAM, a critical factor for AI training.