Nvidia's GB300 Might Set for Q4 Release? Morgan Stanley Sees AI Supply Chain Revolution
Nvidia's upcoming GB300 may introduce key changes in hardware design, potentially setting new standards for AI hardware.
According to a research report released by Morgan Stanley analysts Sharon Shih and Howard Kao on January 6th, Nvidia's next-generation AI GPU, the GB300, is expected to feature significant hardware specification changes and is set to begin shipping in 2025 Q4.
Blackwell Supply Chain —— Redesigned for GB300
With the introduction of the GB300, there will be significant changes in the supply chain and components provided by Nvidia. For the GB200, $NVIDIA(NVDA)$ provided the entire Bianca board (including the Blackwell GPU, Grace CPU, 512GB LPDDR5X, VRM components, all integrated on a single PCB), as well as the switch tray and copper backplate.
For the GB300, Nvidia will no longer provide the complete Bianca board. Instead, it will offer the B300 SXM Puck module, the BGA package for the Grace CPU, and HMC from the US startup Axiado. Ultimate customers will now directly purchase the remaining components on the compute board, with the second-level memory being the LPCAMM module rather than the soldered LPDDR5X. $Micron Technology(MU)$ will become the primary supplier of these modules.
In addition to hardware updates, the report also anticipates an increased value for ODMs (Original Design Manufacturers) in the production process of the GB300 series.
Compared to the GB200, Nvidia will reduce its involvement in the mechanical component design of the GB300, allowing ODMs more design space. ODMs can provide customized design services for hyperscalers, potentially increasing their profit margins.
Impact on Nvidia's Profit Margin
Overall, the ASP (Average Selling Price) for the GB300 is approximately $4,000 higher than the GB200. As mentioned earlier, the biggest change in the BOM is the addition of HBM, which costs about $2,500 more. This means that the incremental profit margin appears to be below 40%, while the overall profit margin of GB200 is in the middle to low range of 70%.
However, the profit margin for the GB300 is compensated for by the reduced overall content provided by Nvidia. Nvidia no longer provides 512GB of LPDDR5X for each Grace CPU, helping Nvidia recoup most of the additional HBM costs. The next most significant part is the savings on the PCB. Combining all these factors, Nvidia's BOM cost has increased by over $1,000, while the ASP has increased by $4,000. The incremental gross profit margin for GB300 compared to GB200 is 73%, meaning that the product's profit margin is neutral if the yield remains unchanged.
While this may seem unremarkable, note that the refresh cycle of HBM is typically a profit margin dilution (as seen in the cases of H200, MT325X), marking a break from this trend. Furthermore, as various engineering issues are resolved, yields will improve, and the profit margin will actually increase throughout the year, although the challenging ramp-up of Blackwell will temporarily suppress profit margins.
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
- sunshineboy·01-09This could really reshape the AI hardware landscapeLikeReport