Renowned analyst Ming-Chi Kuo of TF International Securities recently highlighted ongoing delays in the production and assembly timeline of NVIDIA's GB200 NVL72, which are expected to result in shipments falling below earlier projections. Current estimates indicate that shipments for 2025 will range between 25,000 and 35,000 units, a significant drop from the optimistic market forecast of 50,000 to 80,000 units.
The assembly and mass production of the GB200 NVL72 have faced multiple delays, originally scheduled for September 2024, then pushed to December, and later postponed to the first and second quarters of 2025. These repeated delays have hindered large-scale production, causing a significant shortfall in shipment expectations.
Typically, server development, particularly x86 servers, requires 1 to 1.5 years for general specification upgrades and 1.5 to 2 years for platform or key technology advancements. However, the GB200 NVL72 incorporates cutting-edge technologies aimed at significantly enhancing computational performance, many of which are either scarce or lack mass production experience. This has resulted in a development timeline that exceeds standard expectations. According to Kuo, even with NVIDIA leveraging the world's top server development resources, the 1.5 to 2-year development period remains reasonable. The ongoing delays highlight insufficient development time as the primary cause.
The GB200 NVL72 project began in the fourth quarter of 2023. A reasonable timeline for mass production would be mid-2025. While optimistic projections suggest the second quarter of 2025 as a potential launch window, any further delays into late 2025 could undermine market confidence.
Furthermore, the GPUs consumed by lower-spec GB200 models and several HGX orders only equate to approximately 2,000 to 4,000 GB200 NVL72 units. This limited volume provides little support for market sentiment.
Looking ahead, the next-generation GB300 NVL72, which incorporates several key specification upgrades, is expected to require additional development time, with mass production likely delayed until the first half of 2026. Given the low visibility surrounding production readiness, its short-term impact on market confidence remains minimal.
Kuo also emphasized that the shipment challenges stem from supply-side constraints rather than demand issues. In light of TSMC's cautious capacity expansion and the scarcity of advanced manufacturing resources, it is unlikely that NVIDIA will reduce orders unless the GB200 NVL72 faces further significant delays.