NVIDIA will begin mass production of GB300 artificial intelligence servers this September after supply chain partners adapted to recent design modifications. The company encountered manufacturing difficulties when it initially released Blackwell Ultra systems during the first half of 2025. Design complexity from the Cordelia board architecture and untested SOCAMM memory components created production bottlenecks that limited availability to select partners like Dell and Microsoft.
The chip manufacturer resolved these challenges by adopting the Bianca board design, which mirrors the GB200 architecture that suppliers already understood. This strategic shift reduced manufacturing complexity and allowed production partners to leverage existing knowledge from previous projects. Supply chain companies have begun testing small-scale GB300 shipments and anticipate reaching full production capacity during the fourth quarter.
Strong demand from the Sovereign AI program has generated substantial orders for the new servers. NVIDIA maintains an aggressive six to eight month development timeline, with the next Rubin generation expected by late 2025.
The chip manufacturer resolved these challenges by adopting the Bianca board design, which mirrors the GB200 architecture that suppliers already understood. This strategic shift reduced manufacturing complexity and allowed production partners to leverage existing knowledge from previous projects. Supply chain companies have begun testing small-scale GB300 shipments and anticipate reaching full production capacity during the fourth quarter.
Strong demand from the Sovereign AI program has generated substantial orders for the new servers. NVIDIA maintains an aggressive six to eight month development timeline, with the next Rubin generation expected by late 2025.
