Guo Mingchi: It is currently expected that the shipment of GB200 NVL72 in 2025 may be about 25,000-35,000 containers.

Famous analyst Ming-Chi Kuo's latest article states that the expected mass shipment of GB200 NVL72 in 2Q25 may still be relatively optimistic. Since September 2024, the production schedule of GB200 NVL72 has been delayed multiple times. Kuo believes that the repeated delays in mass shipments will result in lower than expected shipment volumes. Currently, it is expected that around 25,000-35,000 containers of GB200 NVL72 will be shipped in 2025, significantly lower than the most optimistic market expectations of 50,000-80,000 containers last year. However, what may have the biggest impact on market sentiment right now is not the change in shipment volume, but the negative impact on market confidence caused by the multiple delays of GB200 NVL72. The negative impact on market confidence includes the fact that although the supply chain expects GB200 NVL72 to be mass shipped in 2Q25, investors lack confidence in this commitment until they see evidence of mass shipments; even though some suppliers or customers have posted photos of GB200 NVL72 finished products with small/ sample shipments, the help to Nvidia or supply chain stock prices is limited because investors are more concerned about clear evidence of mass shipments (such as supply chain investigations, revenue, etc.); when stock prices react positively in advance but market confidence is low, related market rumors (which may not entirely be true) are easily magnified or overinterpreted, such as the previous ASIC replacing Nvidia AI chips, and recent CoWoS expansion. Kuo points out that the development time from the start of a general server (x86) to mass shipment, and the upgrade of general specifications, typically takes 1-1.5 years, and if it involves platform upgrades or key technology upgrades, it may take 1.5-2 years. GB200 NVL72 significantly enhances computing power, adopts a variety of high-level technologies that are lacking in production experience, and the development difficulty is much higher than that of a general server. Even if Nvidia integrates top development resources for global server supply chains, a reasonable estimate for development time still requires 1.5-2 years. Comparing the multiple delays of GB200 NVL72, the main reason for the delays is clearly insufficient development time. The development start time for GB200 NVL72 is 4Q23, if a reasonable development time is considered to be 1.5-2 years, 2Q25 is a reasonable and optimistic expectation, and if it is further delayed until 2H25, it will not be surprising (but it will once again undermine market confidence if it happens). Currently, the lower-spec models of GB200 and a few orders of HGX GPUs consume only about 2000-4000 containers of GB200 NVL72, providing limited help in boosting market confidence. In addition, GB300 NVL72 also has several key specification upgrades and requires development time, so it is more likely that mass shipments will not occur until 1H26. Due to low visibility of production capacity, short-term positive factors are limited in boosting market confidence. Since the shipment issue is on the supply side rather than the demand side, and with TSMC cautiously expanding production and advanced processes being scarce resources, Nvidia will not cut orders in the short term unless there are multiple delays in GB200 NVL72 again.