NVIDIA debuted its next-generation Blackwell Ultra architecture, which is intended to drive enterprise AI in the future, at GTC 2025 and Computex Taipei in May. Currently the most powerful GPU NVIDIA has ever released, the flagship GPU, called Blackwell Ultra, is designed for large-scale inference, agentic, and reasoning workloads.
The foundation of the GB300 NVL72 system is this innovative hardware, which consists of a liquid-cooled rack configuration that combines 36 Grace CPUs and 72 Blackwell Ultra GPUs. According to NVIDIA, it outperforms the GB200 system by up to fifty percent. By using Dell-built servers that are currently housed in CoreWeave’s cloud infrastructure, CoreWeave and Dell Technologies have already become the first to implement this platform at scale.
NVIDIA unveiled the RTX PRO 6000 Blackwell Server Edition, a data-center GPU designed for enterprise inference, scientific computing, and round-the-clock operations, concurrently with the Ultra launch. It offers significant speed and flexibility gains thanks to its 96 GB GDDR7 memory, up to 24,064 CUDA cores, and support for MIG (Multi-Instance GPU), which allows it to run multiple isolated workloads. In comparison to previous generation GPUs, real-world benchmarks report up to 5× faster LLM inference, nearly 7× genomic sequencing, 3.3× text-to-video generation, and 2× improvements in recommendation engine and rendering tasks.
In addition to bringing massive memory (96 GB) and AI-first optimisations to the desktop environment for designers, developers, and data scientists, the RTX PRO 6000 Workstation Edition was designed under the new RTX PRO brand, which replaced the previous Quadro and Ada naming schemes.
With the launch, NVIDIA is positioned to spearhead the shift to AI factories, which are highly optimised enterprise infrastructure for both physical and agentic AI. NVIDIA provides full-stack solutions that integrate hardware, networking, and software based on the Blackwell Ultra and AI Enterprise platforms through alliances with Dell, HP, HPE, Cisco, Lenovo, and other companies.
NVIDIA’s Blackwell Ultra is a significant advancement for AI in the enterprise, accelerating previously impractical workloads and redefining the standards for reasoning and large-scale inference. Commercial deployments are already underway, and it offers enterprise-grade performance at scale.
