NVIDIA Unveils Roadmap for Next-Gen AI GPUs: Blackwell Ultra and Vera Rubin

NVIDIA's Next-Generation AI GPU Roadmap

NVIDIA, the leading GPU manufacturer, has unveiled its ambitious roadmap for next-generation AI GPUs, with CEO Jensen Huang confirming the development of Blackwell Ultra and Vera Rubin architectures. These upcoming technologies promise to revolutionize the AI and data center landscape, offering unprecedented performance and capabilities 1

1

2

2

.

Blackwell Ultra: The Midcycle Refresh

Blackwell Ultra, also known as the B300-series, is set to launch in the second half of 2025. This midcycle refresh of the Blackwell architecture is expected to deliver significant improvements over its predecessors 1

1

4

4

:

• Higher compute performance
• Eight stacks of 12-Hi HBM3E memory, providing up to 288GB of onboard memory
• Estimated 50% performance uplift compared to the B200-series (unofficial)
• Integration with Mellanox Spectrum Ultra X800 Ethernet switch, supporting up to 512 ports

Jensen Huang emphasized the annual rhythm of NVIDIA's product releases, stating, "The next train is on an annual rhythm and Blackwell Ultra with new networking, new memories, and of course, new processors, and all of that is coming online" 5

5

.

Vera Rubin: A Leap Towards AGI

Following Blackwell Ultra, NVIDIA is preparing to launch its next-generation architecture, codenamed Vera Rubin, in 2026. This new platform represents a significant step towards achieving artificial general intelligence (AGI) 1

1

2

2

. Key features of the Rubin architecture include:

• Eight stacks of HBM4E memory, offering up to 288GB of memory
• Vera CPU
• NVLink 6 switches operating at 3600 GB/s
• CX9 network cards supporting 1,600 Gb/s
• X1600 switches

Huang described Vera Rubin as "a big, big, huge step up" in terms of performance and capabilities 2

2

4

4

.

Beyond Rubin: Future Innovations

NVIDIA is not stopping at Vera Rubin. The company has hinted at post-Rubin products, with potential discussions scheduled for the upcoming GPU Technology Conference (GTC) in March 2025 1

1

2

2

. One rumored development is the Rubin Ultra, projected for release in 2027, which could push GPU design boundaries even further 1

1

:

• Potential incorporation of 12 stacks of HBM4E memory
• Up to 576GB of total memory
• Utilization of 5.5-reticle-size CoWoS interposers and 100mm × 100mm substrates

Impact on AI and Data Centers

These advancements in GPU technology are expected to have a significant impact on AI and data center operations 4

4

:

• Reduced costs and energy consumption
• Improved training capabilities for large AI models
• Enhanced performance for AI inference tasks

NVIDIA's Blackwell architecture has already demonstrated impressive capabilities, with the ability to train a 1.8 trillion parameter model using only 2,000 GPUs and 4 megawatts of power, compared to 8,000 GPUs and 15 megawatts required by the previous Hopper series 4

4

.

Market Response and Future Outlook

NVIDIA's continued innovation in AI GPU technology has been well-received by the market. The company reported total revenues exceeding $130.5 billion for the financial year, reflecting a 114% year-over-year increase 4

4

. Jensen Huang noted the incredible demand for Blackwell, stating, "Demand for Blackwell is amazing as reasoning AI adds another scaling law -- increasing compute for training makes models smarter and increasing compute for long thinking makes the answer smarter" 4

4

.

As NVIDIA prepares to unveil more details about Blackwell Ultra and Vera Rubin at GTC 2025, the AI and technology industries eagerly anticipate the next wave of innovations that will shape the future of computing and artificial intelligence 2

2

5

5

.