NVIDIA Unveils Vision for Next-Gen 'AI Factories' with Vera Rubin and Kyber Architectures

NVIDIA's Vision for Next-Generation AI Infrastructure

NVIDIA has unveiled its ambitious plans for the future of AI computing at the 2025 OCP Global Summit, showcasing new architectures and collaborations aimed at creating gigawatt-scale 'AI factories' 1

1

2

2

. The company's vision centers around two key innovations: the Vera Rubin NVL144 open architecture rack server and the Kyber rack-scale generation.

Vera Rubin NVL144: Powering the AI Factories of Tomorrow

The Vera Rubin NVL144 is designed as a 100% liquid-cooled, modular system that promises to revolutionize data center efficiency and performance 1

1

2

2

. Key features include:

1. A central printed circuit board midplane replacing traditional cable connections
2. Modular expansion bays for NVIDIA ConnectX-9 800GB/s networking
3. Support for NVIDIA Rubin CPX for massive-context inference
4. 45°C liquid cooling and a new liquid-cooled busbar for higher performance
5. 20x more energy storage for steady power delivery

NVIDIA is contributing the Vera Rubin NVL144 design to the Open Compute Project (OCP) as an open standard, allowing widespread adoption across the industry 2

2

.

Kyber: Scaling to New Heights

The Kyber rack-scale generation represents NVIDIA's vision for the future of AI infrastructure, designed to support up to 576 NVIDIA Rubin Ultra GPUs by 2027 1

1

4

4

. Kyber introduces several innovations:

1. Vertical blade design, allowing up to 18 compute blades per chassis
2. Purpose-built NVIDIA NVLink switch blades integrated at the back
3. Cable-free midplane for seamless scale-up networking
4. Support for 800-volt direct current (VDC) power delivery

Revolutionizing Power Delivery and Efficiency

A significant focus of NVIDIA's new designs is the transition to 800 VDC power delivery 1

1

2

2

3

3

. This shift from traditional 415 or 480 VAC three-phase systems offers several benefits:

1. 150% more power transmitted through the same copper
2. Reduced material usage and costs
3. Improved energy efficiency
4. Higher capacity for performance in data centers

Industry Collaboration and Adoption

NVIDIA's vision is supported by a growing ecosystem of partners and early adopters:

1. Over 50 NVIDIA MGX partners are preparing for the Vera Rubin NVL144 1

   1

2. 20-plus companies are joining NVIDIA's ecosystem supporting 800 VDC data centers 1

   1

3. Meta and Oracle have announced plans to standardize their data centers on NVIDIA's Spectrum-X Ethernet networking switches 2

   2

4. Foxconn is building a 40-megawatt data center in Taiwan designed for 800 VDC 1

   1

Performance and Efficiency Gains

NVIDIA claims significant performance improvements with its new architectures:

1. The Blackwell GB200 GPUs show a 15-fold gain in inference throughput compared to the previous Hopper generation 3

   3

2. The company suggests that a $5 million investment in Blackwell can generate $75 million in token revenue 3

   3

Looking Ahead

As NVIDIA continues to push the boundaries of AI infrastructure, the company is positioning itself at the forefront of the AI revolution. With the Vera Rubin and Kyber architectures, NVIDIA is laying the groundwork for the next generation of AI applications and services, promising unprecedented computational power and efficiency for the AI-driven future 1

1

2

2

3

3

4

4

.