NVIDIA is facing hurdles in selling their latest GB300 AI server lineup, despite the initial optimism surrounding its unveiling at the GTC 2025 conference. The feedback from the supply chain, however, presents a stark contrast. Reports suggest that major cloud service providers (CSPs), including tech giants like Microsoft, are reconsidering their orders for the GB300 series. The companies seem more inclined towards established and “mature” solutions, like NVIDIA’s HGX systems, rather than diving into the newer Blackwell Ultra technologies.
The previous GB200 lineup inflicted quite a dent in NVIDIA’s reputation, primarily due to initial massive yield rate issues with TSMC’s advanced packaging. Though those problems have been resolved, the setup time for GB200 server racks remains a significant pain point. Furthermore, the dependency on NVIDIA to troubleshoot any cluster issues is causing unease among companies.
Such challenges have understandably made CSPs wary about embracing the GB300 series, with whispers of potential delays in mass production until next year due to an exhausted supply chain. In light of these complications, the projected annual shipments of the GB200 series are about 15,000 units—markedly fewer than the previous Hopper generation. This decline clearly reflects dwindling interest with each new generation of hardware.
Despite these setbacks, the demand for robust computing solutions persists. Companies seem eager to invest in the tried-and-tested HGX 100 servers, which have shown themselves as leaders in the Hopper generation. This trend suggests that while the appetite for advanced computing remains, NVIDIA needs a strategic overhaul in its approach to supply and support if it hopes to regain and retain customer trust and interest in its product lineup.
