Nvidia Begins Shipping Vera Rubin AI Platform Samples, Promising 5× Inference Performance Over Blackwell
Chips & Infrastructure March 8, 2026 📍 Santa Clara, United States News

Nvidia Begins Shipping Vera Rubin AI Platform Samples, Promising 5× Inference Performance Over Blackwell

Nvidia delivers early samples of its next-generation Vera Rubin platform featuring Rubin GPUs and Vera CPUs to select customers, with full production shipments scheduled for H2 2026.

Key Takeaways

Nvidia Vera Rubin platform Rubin GPU Vera CPU Blackwell successor 5x inference performance 3.5x training agentic AI six chips CES 2026 Jensen Huang production H2 2026


Nvidia has begun delivering early samples of its Vera Rubin AI computing platform to select customers, marking the first physical transition from the current Blackwell architecture to its successor. The platform, unveiled by CEO Jensen Huang at CES 2026, features an integrated system of six new chips — including the Rubin GPU and Vera CPU — designed specifically for the era of agentic AI.

Performance Leap

Nvidia projects that Vera Rubin will deliver five times the performance of Blackwell for AI inference workloads and 3.5 times better performance for model training. These improvements come from a combination of architectural innovations, new memory technologies, and the tight integration between the Vera CPU and Rubin GPU components of the platform.

The platform is explicitly designed for agentic AI applications — autonomous AI systems that plan, execute, and adapt without continuous human oversight. This design focus reflects Nvidia's bet that the next major wave of AI compute demand will come from enterprise agent deployments rather than simply larger model training runs.

The Annual Cadence

Nvidia AI Platform Architecture Roadmap
graph LR
    A[Hopper 2022] --> B[Blackwell 2024]
    B --> C[Vera Rubin 2026]
    C --> D[Next Gen 2028]
    style C fill:#4CAF50,color:#fff
Source: Nvidia CES 2026 and investor presentations

Vera Rubin continues Nvidia's aggressive two-year architecture cadence: Hopper in 2022, Blackwell in 2024, Vera Rubin in 2026. This pace ensures that even recently purchased Blackwell hardware faces competitive pressure from within Nvidia's own product line — a dynamic that some investors have flagged as a risk (the 'GPU depreciation question') but that Nvidia frames as essential to maintaining its 85% share of the AI processor market.

Production Timeline

Initial customer samples shipped in February 2026, with full production shipments scheduled for the second half of the year. Nvidia has indicated that its earlier guidance of $300 billion in combined Blackwell and Rubin product revenue for calendar year 2026 may prove conservative — a signal of strong pre-order demand from cloud providers and enterprise customers eager to deploy the next generation of AI infrastructure.

Share X Reddit LinkedIn Telegram Facebook HN