NVIDIA's AI server ecosystem is now channeling development resources into the next-generation GB300 system, aiming to capture the momentum of the upcoming wave in AI performance growth. Industry insiders indicate that system integrators are on track to begin shipments by the end of Q3 2025, while key component manufacturers suggest that cloud service providers (CSPs) will require additional time to customize the GB300 architecture to meet their specific infrastructure needs—pointing to a larger volume ramp expected in Q1 2026.
During his keynote at COMPUTEX Taipei, NVIDIA CEO Jensen Huang announced that the GB200 systems, built on the Grace Blackwell architecture, have officially entered mass production. He also confirmed that numerous CSPs are actively deploying these systems. The upgraded GB300 platform is expected to complete its update cycle by Q3 2025, further enhancing AI inference capabilities and enabling broader deployment across generative AI, agent-based AI, and embodied AI applications.
Foxconn Chairman Young Liu previously noted that while the company does not comment on individual products, experience gained from GB200 manufacturing—particularly in system design, thermal management, and production automation—has significantly accelerated the mass production readiness of the GB300 systems. He added that learning curves for future AI server cabinet production are improving rapidly.
According to Mike Yang, Executive Vice President of Quanta Computer and President of its cloud computing division, QCT, GB300 development is proceeding according to schedule. The system is currently undergoing testing and customer validation, with shipments anticipated in September 2025.
Auras Technology Chairman Lin Yu-Shen confirmed that GB200 systems have entered large-scale delivery, while the GB300 has experienced slight delays. Nevertheless, it is expected to begin rolling out between Q3 and Q4 2025.
Component-level sources reveal that while the GB300 shares its core architecture with the GB200, it includes several design modifications. Each CSP is expected to implement its own tailored configuration, further supporting forecasts that significant shipment volumes will commence in January 2026.