GTC 2025 Keynote Announcements

NVIDIA sets a new standard for AI Ecosystem at GTC 2025

At GTC 2025, NVIDIA delivered a strong message: AI is entering a new era, and accelerated computing will drive the next generation of progress across every industry. From revolutionary GPUs to data centre-scale systems, NVIDIA outlined how its technologies are shaping the future of infrastructure.

Boston Limited welcomed the keynote, highlighting its commitment to supporting customers as they adopt these innovations. The company shared a summary of the most exciting and relevant announcements from the event, focusing on the hardware, systems and platforms designed to help organisations move faster, scale smarter and achieve more with AI.

The Dawn of a New era in AI Infrastructure

The keynote opened with a look back at NVIDIA’s journey — from early GPU innovation to becoming a global leader in AI computing. The rise of agentic AI, models that can reason, plan and act independently, is driving new demands for advanced infrastructure.

With AI continuing to scale rapidly, NVIDIA predicts global data centre investments will soon reach $1 trillion. GPUs, CPUs and high-speed networking have now become critical building blocks of enterprise innovation.

Blackwell architecture sees significant expansion

The headline announcement was the expansion of the Blackwell architecture, designed to support trillion-parameter models, large-scale inference and energy-efficient AI at every level.

• Blackwell B200 GPU: A dual-die design featuring 208 billion transistors and 20 petaflops of AI performance.
• Grace Blackwell GB200 Superchip: Combines two B200 GPUs with a Grace CPU, delivering up to 30 times faster inference than previous generations.
• NVLink Switch System and NVL72 platform: Enables rack-scale deployment with 720 petaflops of AI performance, 1.4TB of memory, and high-speed NVLink interconnects.

Looking to the future: The Future of AI as we know it

Introducing Blackwell Ultra: higher performance, smarter inference

Later this year, NVIDIA will introduce Blackwell Ultra — the next evolution of the Grace Blackwell platform. This next-generation system is designed to deliver a significant leap in inference performance, with improved energy efficiency and scalability.

A demonstration of the NVIDIA Omniverse Blueprint showcased how a 1-gigawatt AI Factory can be designed, tested and optimised using digital twin technology, offering a glimpse into the future of intelligent infrastructure design.

Blackwell Ultra systems, expected in the second half of this year, will enhance training throughput and test-time scaling inference — applying additional compute power at the point of inference to improve model accuracy and responsiveness. These systems are intended to support a new class of AI applications, including AI reasoning, agentic AI and physical AI.

Vera Rubin architecture: the future of data centre platforms

Looking further ahead, NVIDIA presented the next stage of its technology roadmap, with new details on the Rubin Ultra GPU and Vera CPU architectures. These platforms are designed to deliver significant performance improvements and cost efficiencies across enterprise infrastructure.

Systems built on Rubin Ultra, including the Vera Rubin NVL144, are expected in the second half of next year. These platforms aim to enhance scalability while reducing total cost of ownership.

Further systems, fully based on Rubin Ultra, are planned for release in the second half of 2027. Boston Limited will continue to support customers in aligning their infrastructure strategies with this roadmap.

The next big leap will be new platform called Feynman, expected in 2028.

Powering the next industrial revolution: AI Factories

NVIDIA’s infrastructure announcements form the foundation of its AI Factory vision — purpose-built facilities designed to generate intelligence at scale. These environments are positioned as the modern equivalent of manufacturing plants, focused on producing AI tokens, decisions and insights.

By combining Grace Blackwell Superchips, NVLink Switch Systems and platforms such as NVL72, a new class of data centre is now possible, capable of handling foundation model training and real-time inferencing at scale.

Boston Limited is ready to help customers plan, build and scale AI Factories — delivering integrated systems that move AI from experimentation to production-ready deployment.

CUDA X, NIMs and enterprise AI platforms

NVIDIA highlighted CUDA X — a unified software stack bringing together key tools, APIs and libraries across AI, data science, simulation and robotics.

It integrates with:

• NVIDIA NIMs (Inference Microservices) — containerised services for rapid enterprise AI deployment.
• NVIDIA AI Enterprise — a scalable, production-ready software suite available across cloud and on-premise infrastructure.

Boston Limited’s infrastructure solutions are fully optimised for these platforms, enabling high-performance deployment in real-world enterprise environments.

DGX Spark and DGX Station: AI supercomputing for the desktop

NVIDIA also introduced DGX Spark — previously known as Project DIGITS — and DGX Station, a high-performance desktop supercomputer powered by the Grace Blackwell architecture and built on the Blackwell Ultra platform.

These systems bring enterprise-grade AI capabilities to the desktop, enabling users to train, fine-tune and run models locally, with the ability to scale seamlessly into DGX Cloud or other accelerated infrastructure environments.

Boston Limited supports customers in integrating DGX Station systems into broader AI Factory workflows — enabling local development to transition smoothly into production-scale environments, including NVL72 and DGX SuperPOD clusters.

Accelerated networking and NVIDIA Photonics

To support the scale of AI infrastructure, NVIDIA introduced Spectrum-X and Quantum-X — next-generation networking switches built on silicon photonics.

These technologies deliver:

• 3.5 times greater power efficiency
• 63 times better signal integrity
• 10 times more resiliency at scale
• Faster deployment across multi-site AI infrastructure

These advances will enhance Boston Limited’s high-performance infrastructure offerings, supporting large-scale AI clusters and distributed compute environments.

AI agents and robotics: Project GR00T, Cosmos and Newton

NVIDIA also unveiled key innovations in physical AI and robotics:

• Project GR00T: A general-purpose AI model for humanoid robots 
• Isaac GR00T N1: A customisable, open foundation model for robot control 
• Cosmos world models: AI models designed to simulate complex environments 
• Newton physics engine: A simulation platform developed in collaboration with Google DeepMind and Disney Research 

These tools enable developers to create, test and deploy robotics systems with new levels of precision — a space where Boston Limited continues to support customers through its HPC infrastructure expertise.

Boston Limited: ready to support next-generation AI

GTC 2025 has set the direction for the next phase of AI infrastructure development. From Grace Blackwell Superchips to photonic networking and DGX systems, Boston Limited stands ready to support organisations in deploying the technologies that will drive tomorrow’s breakthroughs.

Organisations looking to accelerate their AI strategy are encouraged to contact Boston Limited to explore next-generation infrastructure solutions.

Visit us at GTC 2025

Want to dive deeper into the latest AI infrastructure technologies?

Come and meet the Boston Limited team at SourceCode’s booth (1605) on the show floor at GTC 2025. Our experts will be on hand to discuss how these innovations can be tailored to your organisation’s needs — and how we can help you build, scale and optimise your AI infrastructure.

We look forward to seeing you there.

Tags: gtc, gtc25, gtc keynote, nvidia, nvidia gtc, gtc2025, jensen huang, keynote announcements