NVIDIA has announced at the annual GTC the next evolution of its NVIDIA Blackwell AI factory platform, the NVIDIA Blackwell Ultra. Accordingly, this will pave the way for “the age of AI reasoning”.

Moreover, Jensen Huang, founder and CEO of NVIDIA, also unveiled the timeline of the company’s next-generation GPUs until 2028 that will complement the demand for AI and will lead the company to pave the way to gigawatt AI factories.

“AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance,” said Huang in his jampacked keynote address at GTC 2025 at SAP Center in California, which he called the “Superbowl of AI”.

“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.”

NVIDIA’s GTC, which is short for GPU Technology Conference, got its start in 2009 and has been an annual global event for artificial intelligence (AI) conference for developers, engineers, researchers, inventors, and IT professionals. Every year, the topics focus mainly on advancements and innovations in AI, computer graphics, data science, machine learning, and other autonomous applications.

Better AI Performance

Built on the groundbreaking Blackwell architecture introduced a year ago, Blackwell Ultra includes the NVIDIA GB300 NVL72 rack-scale solution and the NVIDIA HGX™ B300 NVL16 system.

The Blackwell Ultra GB300 NVL72, coming in second half of 2025, delivers 1.5x more AI performance than the NVIDIA GB200 NVL72, as well as increases Blackwell’s revenue opportunity by 50x for AI factories, compared with those built with NVIDIA Hopper.

Moreover, Huang said the NVIDIA GB300 NVL72 connects 72 Blackwell Ultra GPUs and 36 Arm Neoverse-based NVIDIA Grace CPUs in a rack-scale design, acting as a single massive GPU built for test-time scaling. With the NVIDIA GB300 NVL72, AI models can access the platform’s increased compute capacity to explore different solutions to problems and break down complex requests into multiple steps, resulting in higher-quality responses.

Next in the timeline which Huang showed the audience is the Vera Rubin NVL 144 and will be coming in the second half of 2026. Moreover, Huang also showed the coming of Rubin Ultra NVL576 in the second half of 2027. Huang described it as an “extreme scale up” as it features 2.5 million parts and is connected to 576 GPUs.

“This gives you an idea of the pace at which we are moving,” Huang told the audience.

Software Innovations, Too

The full-stack NVIDIA AI platform supports the entire NVIDIA Blackwell Ultra product portfolio. The NVIDIA Dynamo open-source inference framework, which Huang also announced at GTC 2025, scales up reasoning AI services, delivering leaps in throughput while reducing response times and model serving costs by providing the most efficient solution for scaling test-time compute.

NVIDIA Dynamo is new AI inference-serving software designed to maximize token revenue generation for AI factories deploying reasoning AI models.
Blackwell Ultra-based products are expected to be available from partners starting from the second half of 2025.

Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro are expected to deliver a wide range of servers based on Blackwell Ultra products, in addition to Aivres, ASRock Rack, ASUS, Eviden, Foxconn, GIGABYTE, Inventec, Pegatron, Quanta Cloud Technology (QCT), Wistron and Wiwynn.

Cloud service providers Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure and GPU cloud providers CoreWeave, Crusoe, Lambda, Nebius, Nscale, Yotta and YTL will be among the first to offer Blackwell Ultra-powered instances.

19 March 2025