NVIDIA’s customers, which include Google, Meta, Microsoft, AWS, Oracle, and others, have already bought every Blackwell GPU that NVIDIA (and TSMC) will produce to the point where Blackwell is effectively “sold out” for the next 12 months. This comes directly from NVIDIA management, including CEO Jensen Huang, who recently sat down with Morgan Stanley analysts.
VIEW GALLERY – 2 IMAGES
Blackwell was recently delayed due to NVIDIA B100 and B200 GPU packaging issues that required a redesign, but production is now well underway. Recently, OpenAI received one of the first engineering builds of the NVIDIA DGX B200 AI server, which includes eight B200 AI GPUs with up to 1.4TB of high-speed HBM3 memory. With Blackwell delivering “massive performance leaps” regarding next-gen AI training and inference performance, it’s at the top of the list for all large-scale AI projects.
And it’s gotten to the point where the high demand for Blackwell means new customers ordering Blackwell GPUs will have to wait until late next year to receive their orders.
According to Morgan Stanley analysts, this demand will see NVIDIA gain market share in 2025, even though it will compete with more AI chips and GPU hardware from AMD, Intel, and other companies.
“Our view continues to be that NVIDIA is likely to actually gain share of AI processors in 2025,” Joseph Moore, an analyst with Morgan Stanley, wrote. “The biggest users of custom silicon are seeing very steep ramps with NVIDIA solutions next year, [and] everything we heard this week reinforced that.”
Blackwell represents NVIDIA’s most cutting-edge AI GPU technology to date, utilizing TSMC’s CoWoS-L packaging and next-gen HBM3E memory. Over the next few quarters, NVIDIA will produce as many Blackwell GPUs as possible while dealing with capacity limitations and potential memory supply issues. The demand for AI GPU hardware is not going anywhere, with Blackwell showing that it’s intensifying.