SAN JOSE, California – Nvidia CEO Jensen Huang's keynote at this week’s GTC 2025 event promised a future where artificial intelligence (AI) fundamentally reshapes business operations, moving beyond incremental improvements to a completely new production model. In support of that future, Nvidia is building the infrastructure for the age of AI, enabling enterprises to transition from traditional data centers to something Huang called “AI factories.”

Nvidia’s Blackwell GPU architecture is the foundation of this push and has been infused with what the vendor claims is not just an upgrade, but a major advancement in AI computing – something up to 100-times more powerful than its previous generation of Hopper processors.

Nvidia's Blackwell GPU is specifically designed to speed up data-heavy AI workloads, particularly those related to generative artificial intelligence (genAI) and large language model (LLM) (LLMs). The additional power and speed result in substantially faster training and inference times.With Blackwell Ultra, Nvidia is pushing the boundaries of AI reasoning, allowing AI models to handle more complex tasks and provide higher-quality responses. This is particularly valuable for applications that require advanced problem-solving and decision-making.

IT companies often extol their new products as the latest and greatest, but Huang might not be far off from the truth in this statement: “This represents the largest scale-up of AI processing power ever attempted, designed to meet the exploding demands of large-scale AI inference,” Huang said.

The Grace Blackwell NVLink72 rack, which boasts more than an exaflop of computing power, signals a move toward liquid-cooled, high-density infrastructure – a necessity to handle the power and efficiency challenges of these new workloads.

Nvidia Dynamo: An AI operating system for enterprise-scale inference To fully leverage Blackwell's potential, Nvidia unveiled Dynamo, an AI-optimized operating system (OS) designed for large-scale inference. This software unlocks a 40-times performance improvement, Huang said, addressing the critical need for speed and scalability in AI applications.

Dynamo streamlines key processes – pre-filling information, optimizing memory access, and accelerating response times – essentially providing the software foundation to run AI factories, which will mass-produce new AI models and eventually replace old-school data centers.

“AI factories reflect the accelerating adoption of enterprise AI, driven by the need for full-stack AI solutions and an AI-powered workforce,” Huang said. “Nvidia is providing the tools and infrastructure to enable this transition with our own full-stack solutions and Dynamo AI OS.”

Huang added that Blackwell enables businesses to tackle complex AI tasks previously deemed impossible, while Dynamo optimizes AI workflows, leading to faster insights and improved resource utilization. This can support AI-driven robotics and digital twins to revolutionize industries and drive the next wave of automation and operational efficiency.

It's all about enterprises making strategic infrastructure investment, Huang said.

“Last year our business-pitch phrase was: 'The more you buy, the more you save,’” Huang said. “This year it's been revised: 'The more you buy, the more revenue you get.' Sounds funny but it's true.”