Nvidia Blackwell chips will ship 450,000 units in Q4, bringing in over $10 billion in revenue

Latest update time：2024-09-26

Reads：

????If you hope to meet more often, please mark the star ?????? and add it to your collection~

Source: Content compiled from tomshardware, thank you.

Morgan Stanley analysts believe that Nvidia will produce about 450,000 AI GPUs based on the Blackwell architecture, despite low production volumes due to significant but easily fixable design issues. If the information is accurate and the company is able to successfully sell these products this year, it could represent a revenue opportunity of more than $10 billion.

According to The_AI_Investor, a blogger who tends to get their hands on such reports, analysts at investment bank Morgan Stanley wrote in a note to clients: "It is expected that 450,000 Blackwell chips will be produced in the fourth quarter of 2024, which will bring Nvidia a potential revenue opportunity of more than $10 billion."

While the $10 billion and 450,000 numbers seem impressive, Nvidia has been selling very limited quantities of its in-demand Blackwell GPUs for around $22,000 each, well below the rumored $70,000 per Blackwell GPU module. While actual pricing for data center hardware depends on factors like volume and demand, it’s a little odd that the first batch of ultra-high-end GPUs are being sold for less than Nvidia’s current generation H100.

Of course, Nvidia would rather sell “reference” AI server cabinets with Blackwell GPUs: the NVL36 with 36 B200 GPUs is expected to sell for $1.8 million to $2 million, while the NVL72 with 72 B200 GPUs is expected to start at $3 million. Selling cabinets is much more profitable than GPUs, GPU modules, or even DGX and HGX servers, and it is not surprising to make $10 billion from selling such machines. However, in this case, Nvidia does not need to provide 450,000 GPUs to make $10 billion in revenue.

At the end of August, Nvidia said it had to change the photomask design of its GPUs, codenamed B100/B200, in order to increase production. The company also noted that these GPUs will enter mass production in the fourth quarter and continue until fiscal 2026. In the fourth quarter of fiscal 2025 (which the company starts in late October), Nvidia expects to achieve "multibillion-dollar Blackwell revenues," which is much lower than the figure mentioned by Morgan Stanley analysts.

SemiAnalysis reports that a mismatch in the coefficient of thermal expansion (CTE) of the Blackwell die and packaging material causes failures. Nvidia’s B100 and B200 GPUs are the first to feature TSMC’s CoWoS-L packaging, which uses passive local silicon interconnect (LSI) bridge chips integrated into a redistribution layer (RDL) interposer. The precise placement of these bridges is critical, but CTE mismatches between the die, bridges, organic interposer, and substrate can lead to warping and system failures. Nvidia reportedly had to redesign the top metal layer of the GPU to improve yields, but no details were revealed, only mentioning new masks. The company clarified that no functional changes were made to the Blackwell silicon and that it was only focused on improving yields.

Due to the required design changes for the B100 and B200 GPUs, as well as insufficient TSMC CoWoS-L (which is significantly different from the mature CoWoS-S technology) packaging capacity, analysts do not expect Nvidia to ship significant volumes of Blackwell-based GPUs in the fourth quarter of calendar 2024 or the fourth quarter of the company's fiscal 2025.

Reference Links

https://www.tomshardware.com/pc-components/gpus/nvidia-expected-to-produce-450000-blackwell-ai-gpus-in-q4-potential-dollar10b-in-revenue-for-the-chipmaker

END

????Semiconductor boutique public account recommendation????

▲Click on the business card above to follow

Focus on more original content in the semiconductor field

▲Click on the business card above to follow

Focus on the trends and developments of the global semiconductor industry

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 3897th content shared by "Semiconductor Industry Observer" for you, welcome to follow.