Nebius achieves NVIDIA Exemplar Cloud on NVIDIA GB300 for training: Validated performance for hyperscale AI
Nebius achieves NVIDIA Exemplar Cloud on NVIDIA GB300 for training: Validated performance for hyperscale AI
In the past year, Nebius has hit a series of milestones with NVIDIA that define what a full-stack AI cloud looks like in practice: Exemplar Cloud on NVIDIA H200 GPUs for training workloads, Europe’s first NVIDIA GB300NVL72 Blackwell Ultra deployment, a $2 billion strategic investment with deep engineering collaboration on AI factories, and a purpose-built cloud for physical AI and robotics. The trajectory has been consistent: build the hyperscale infrastructure that training, agentic AI, and production inference demand while proving performance at every layer.
Now, we’re announcing that Nebius has achieved NVIDIA Exemplar Cloud on NVIDIA GB300 NVL72 for training, powered by NVIDIA Blackwell Ultra GPUs. This validates performance against NVIDIA’s reference architecture, meeting the 95% threshold across all benchmarking recipes.
What Exemplar Cloud means
The NVIDIA Exemplar Clouds initiative evaluates cloud providers against real-world AI training workloads using NVIDIA’s Performance Benchmarking recipes. It validates performance, resiliency, and TCO against NVIDIA’s standards — not synthetic benchmarks. For teams deciding where to run large-scale training, it’s a transparent, independently verified signal that infrastructure will perform as expected.
Nebius is among the first cloud providers globally to achieve Exemplar on NVIDIA GB300 NVL72 for training, and one of a small group to hold it across multiple GPU generations.
More than a Benchmark
Exemplar Cloud is one outcome of a collaboration that now spans AI factory design, fleet health monitoring, inference optimizations, and the Physical AI Data Factory Blueprint — where Nebius was named alongside Azure as a cloud provider in the NVIDIA announcement. The engineering collaboration that drives Exemplar validation is the same collaboration shaping how Nebius builds and operates infrastructure across the full stack — from cluster acceptance and networking tuning to production inference through Token Factory.
“The NVIDIA Exemplar Cloud achievement is a testament to Nebius’s commitment to delivering world-class AI infrastructure,” said Warren Barkley, Vice President of Product Management at NVIDIA. “Achieving Exemplar across multiple GPU generations, from NVIDIA HGX H200 to NVIDIA GB300 NVL72, reflects the depth of engineering collaboration between our teams and gives developers and enterprises confidence that their most demanding workloads will perform at scale.”
“As we scale into gigawatt-class AI factories and next-generation platforms with NVIDIA, Exemplar Cloud is how we prove the infrastructure performs — not just at peak specs, but against real-world training workloads at scale,” said Laurelle Roseman, VP of Global Partnerships. “That confidence comes from our ability to rapidly bring up and validate real-world training and inference workloads, and deliver results customers can trust.”
With NVIDIA Vera Rubin NVL72 arriving in H2 2026 and gigawatt-scale AI factories under construction across the U.S. and Europe, Exemplar Cloud on GB300 NVL72 establishes the performance baseline for what comes next.
