NVIDIA CEO Jensen Huang took center stage at Computex 2025 in Taipei to unveil the company's next evolutionary leap in AI computing: the GB300 'Blackwell Ultra' platform, scheduled for release in the third quarter of this year.
Built on the groundbreaking Blackwell architecture introduced last year, the GB300 NVL72 rack-scale solution delivers 1.5 times more AI performance than its predecessor, the GB200 NVL72. The system features 72 Blackwell Ultra GPUs and 36 Arm-based NVIDIA Grace CPUs in a fully liquid-cooled design, creating a unified computing platform specifically engineered for AI reasoning workloads.
The technical specifications reveal significant improvements, including 288GB of HBM3E memory per GPU (up from 192GB), representing a 50% increase in memory capacity. This expanded memory, combined with architectural enhancements, enables the system to handle more complex AI workloads, particularly for reasoning, agentic AI, and physical AI applications. The GB300's GPUs will consume 1,400 watts of power, 200 watts more than the GB200, while delivering substantially higher performance.
"AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance," Huang explained during his keynote. "We designed Blackwell Ultra for this moment — it's a single versatile platform that can easily and efficiently do pretraining, post-training and reasoning AI inference."
The GB300 platform represents a critical component of NVIDIA's vision for AI infrastructure. Huang has repeatedly emphasized that AI is becoming fundamental infrastructure similar to electricity or the internet, requiring specialized "factories" to produce valuable computational output. The GB300 is positioned as the engine for these AI factories, with NVIDIA claiming it increases Blackwell's revenue opportunity by 50x compared to systems built with the previous Hopper architecture.
The announcement comes as NVIDIA continues to dominate the AI chip market, with data center revenue accounting for over 90% of its total revenue. Major cloud providers including Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure will be among the first to offer GB300-powered instances when the platform becomes available in the second half of 2025.