Lenovo First Vera Rubin Partner at GTC

🔑 Enhanced Key Takeaways

• •Lenovo's AI Cloud Gigafactory integrates Vera Rubin NVL72 with Neptune liquid cooling and lifecycle services to reduce deployment time to first token to weeks.[1][4][9]
• •Vera Rubin NVL72 achieves up to 10x higher inference throughput per watt and one-tenth cost per token versus Blackwell, enabling training of large MoE models with one-fourth the GPUs.[3][7]
• •NVIDIA announced Vera Rubin platform at CES 2026, with production underway and partner systems targeted for second half of 2026 availability.[3][8]

🛠️ Technical Deep Dive

• •Vera Rubin NVL72 unifies 72 Rubin GPUs, 36 Vera CPUs (each with 88 custom Arm Olympus cores supporting 176 threads), NVLink 6 switches (260TB/s rack bandwidth, 3.6TB/s per GPU), ConnectX-9 SuperNICs, BlueField-4 DPUs, and Spectrum-6 Ethernet.[2][3][5][6]
• •Vera CPU connects to Rubin GPUs via NVLink C2C at 1.8TB/s bandwidth, supports up to 1.5TB SOCAMM LPDDR5X memory at 1.2TB/s.[3]
• •Features cable-free modular tray for quick component swaps, zero-downtime NVLink maintenance, rack-wide trusted execution environment, and advanced RAS for reliability.[3]

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline