Nvidia's Vera Rubin NVL72 cloud rollout is expanding to Europe, with H2 2026 deployments targeting European cloud providers. The architecture, named after astronomer Vera Rubin, integrates 72 GPUs per rack via NVLink and succeeds the Blackwell platform.
Key facts
- Vera Rubin NVL72 integrates 72 GPUs per rack via NVLink.
- European cloud deployments scheduled for H2 2026.
- Architecture succeeds Blackwell, targets AI training/inference.
- Nvidia claims up to 30x performance gains over prior architectures.
- Nvidia invested $6.5B in photonic interconnect in June 2026.
Nvidia's Vera Rubin NVL72 cloud rollout is expanding to Europe, with deployments scheduled for the second half of 2026, according to Tech Times. The Vera Rubin NVL72 architecture, named after astronomer Vera Rubin, succeeds the Blackwell platform and integrates 72 GPUs per rack via NVLink, targeting cloud providers for AI training and inference workloads.
What Makes Vera Rubin Different
The Vera Rubin NVL72's design is a direct response to the scaling challenges of large language model training, where inter-GPU bandwidth often bottlenecks performance. By packing 72 GPUs into a single rack with NVLink interconnect, Nvidia claims up to 30x performance gains over prior architectures for certain workloads [per the source]. This architecture first appeared in Nvidia's 2024 GTC roadmap, but the European cloud rollout marks its first confirmed deployment outside North America.
Competitive Context
Nvidia's move into Europe comes as cloud rivals race to secure next-generation hardware. AWS launched Blackwell-based EC2 G7 instances in June 2026, as previously reported, while Google Cloud and Microsoft Azure have not yet announced Vera Rubin availability. The European expansion also follows Nvidia's $6.5 billion investment in photonic interconnect supply announced in June 2026, per prior reporting, which likely underpins the data center infrastructure needed for these deployments.
Unanswered Questions
Nvidia has not disclosed which European cloud providers will deploy Vera Rubin NVL72 systems, nor the total number of racks planned for the region. Pricing per rack or per GPU-hour for cloud customers also remains unannounced. The company did not specify whether the European deployments will use Nvidia's own DGX Cloud or third-party providers.
What This Means for AI Workloads
For AI labs training models beyond 100 billion parameters, the Vera Rubin NVL72's rack-scale integration reduces the need for multi-rack communication, cutting latency and power consumption. European AI startups and research institutions, including those partnered with GENCI in France per earlier reporting, stand to benefit from local access to the architecture without shipping data across the Atlantic.
What to watch
Watch for Nvidia's Q3 2026 earnings call, likely in November 2026, which may disclose Vera Rubin NVL72 revenue contribution and named European cloud partners. Also track Google Cloud and Azure Vera Rubin announcements, as their silence signals potential supply constraints or competitive positioning.
Source: news.google.com
