CNBC analysts have assessed the Blackwell GPU's pricing strategy and concluded that it aligns with market expectations, especially given the growing demand for AI training and deployment solutions. The price range of the Blackwell GPU remains competitive, particularly when compared to its predecessor, the H100, which was priced between $25,000 and $40,000 upon its release. The H100's launch price marked a significant increase from its own predecessor, the A100, highlighting Nvidia's ongoing investment in advancing GPU technology. The B200, built on the advanced Blackwell design and packed with 192 GB of high-speed HBM3E memory, stands out because it has two chips. Together, these chips have over 204 billion tiny parts (transistors), making it more complex than the simpler GH100 model, which only has 80GB of memory. The B200, with its two chips, is pricier to make than the GH100, with estimated costs of $3,100 for the GH100 and $6,000 for the B200.
Nvidia has spent more than $10 billion on creating the GB200 and its other new tech. Before, Nvidia's partners sold the H100 part for $30,000 to $40,000. This high price was due to a lot of people wanting it and not many being made, partly because of production limits at TSMC.
Nvidia seems to prefer selling big, complete servers with the B200 chips, like the DGX B200 server with eight B200 GPUs or even larger setups called SuperPODs with 576 B200 GPUs. These products are meant for very powerful computing tasks and show Nvidia's focus on selling full systems rather than just parts. As a result, you won't find prices for just the B200 parts on Nvidia's website, only for the big server setups. So, when looking at prices for Nvidia's B200, it's important to keep in mind this strategy and be careful with the information, even though it comes from a reliable source.
Source: ithome