What are the key specs of the Vera Rubin NVL72?

72 Rubin GPUs + 36 Vera CPUs in a liquid-cooled rack; 3.6 TB/s per-GPU NVLink 6 bandwidth (260 TB/s rack total); 75 TB total fast memory (20.7 TB HBM4 + 54 TB LPDDR5X); 1.6 PB/s HBM bandwidth; 3.6 EFLOPS NVFP4 inference (10x performance/watt and 1/10 cost per token vs Blackwell); 2.5 EFLOPS training with 1/4 the Blackwell GPU count for MoE models.

What is the Vera CPU and how does it differ from Intel/AMD server CPUs?

The Vera CPU is purpose-built for agentic AI and reinforcement learning—not a general-purpose server CPU. Its 88 Olympus cores at 1.2 TB/s bandwidth and 1.8 TB/s NVLink-C2C with Rubin GPUs make it 50% faster and 2x more efficient than traditional rack-scale CPUs for AI workloads. It cannot simply replace Intel Xeon or AMD EPYC for general compute.

When is Vera Rubin available and what does it cost?

NVIDIA confirmed Vera Rubin is in full production and will be delivered to customers in H2 2026. First recipients are major cloud providers (AWS, Azure, Google Cloud). Rack pricing is expected in the multi-million dollar range (the Blackwell NVL72 was priced around $2.7M). Individual access will be through cloud services.

Vera Rubin Platform Deep Dive: NVIDIA's Supercompute Foundation for Agentic AI

NVIDIA launched the Vera Rubin platform at GTC 2026, positioned as supercompute infrastructure for the agentic AI era. Integrating new Vera CPUs and Rubin GPUs into NVL72 and HGX NVL8 configurations, it's designed for the full AI lifecycle from large-scale pre-training to real-time agentic inference. The Vera Rubin Space Module extends AI compute to orbital data centers.

Vera Rubin Platform Deep Dive: NVIDIA's Supercompute Foundation for the Agentic AI Era

NVIDIA's Vera Rubin platform, fully detailed at GTC 2026, is purpose-built for Agentic AI—not just another generational GPU upgrade.

Complete Technical Specifications

Vera CPU:

88 custom "Olympus" cores with NVIDIA Spatial Multithreading
227 billion transistors
LPDDR5X memory: up to **1.5 TB capacity**, **1.2 TB/s bandwidth**
50% faster + 2x more efficient than traditional rack-scale CPUs for agentic AI
NVLink-C2C bandwidth with Rubin GPUs: **1.8 TB/s** (7x PCIe Gen 6)
Full Confidential Computing support

Vera Rubin NVL72:

72 Rubin GPUs + 36 Vera CPUs in fully liquid-cooled rack
NVLink 6: **3.6 TB/s bidirectional per GPU**, **260 TB/s rack total**
Memory: 20.7 TB HBM4 + 54 TB LPDDR5X = **75 TB total fast memory**
HBM bandwidth: **1.6 PB/s**
Inference: **3.6 EFLOPS** NVFP4 — 10x performance/watt vs Blackwell, 1/10 cost per token
Training: **2.5 EFLOPS** NVFP4 — trains MoE models with 1/4 the Blackwell GPU count
Integrated: NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet, **NVIDIA Groq 3 LPU**

The Groq 3 LPU Integration: A Surprise Partnership

The integration of Groq Inc.'s LPU (Language Processing Unit) enables ultra-low latency inference alongside high-throughput GPU inference—a critical combination for AI agents that need fast decision-making with high-quality outputs simultaneously.

Why Vera Rubin is Different from Blackwell

Blackwell was optimized for large-scale parallel training. Vera Rubin is optimized for **continuous agentic AI inference**: always-on operation, low-latency decision-making, massive concurrent agent instances, and reinforcement learning continuous updates.

The Vera CPU's 88-core design, Groq 3 LPU integration, and NVLink 6 high-bandwidth interconnect are all precisely engineered for this agentic workload profile.

Availability: H2 2026

Vera Rubin is in full production; customer deliveries expected H2 2026. First recipients will be major cloud providers and hyperscale AI labs.

In-Depth Analysis and Industry Outlook

From a broader perspective, this development reflects the accelerating trend of AI technology transitioning from laboratories to industrial applications. Industry analysts widely agree that 2026 will be a pivotal year for AI commercialization. On the technical front, large model inference efficiency continues to improve while deployment costs decline, enabling more SMEs to access advanced AI capabilities. On the market front, enterprise expectations for AI investment returns are shifting from long-term strategic value to short-term quantifiable gains.