The use case: how Recraft deploys Nebius AI Cloud
Recraft trained its 20-billion-parameter text-to-image foundation model from scratch on Nebius AI, using Managed Kubernetes with Kubeflow and PyTorch for distributed GPU training. Nebius engineers helped fix network and hardware issues during training.
More reported results
Figures published in Nebius AI Cloud’s case study. They are not independently measured by usedby.
These adjustments made the network 4x faster, leading to a ~6x increase in training speed.
In the customer’s words
Nebius stands out when compared to other clouds, especially if you’re looking for flexibility and quick support. On the scalability front, they always have GPUs available, so no frustrating delays waiting for quotas. The storage speed on Nebius is also higher in comparison to some other clouds. Nebius offers more control, better support and reliable scalability compared to some other clouds.If you have simple code, you can switch to the new GPUs without doing anything special,Nebius AI Cloud users in Artificial Intelligence
Companies in the same sector, with public evidence.
DubformerConfirmedHelicalConfirmed
HiggsfieldConfirmed
KrispConfirmedPositronic RoboticsConfirmed
Prime IntellectConfirmedGo straight to the source
Based on Nebius AI Cloud’s published case study. We check the public evidence; the results are reported by the vendor.
nebius.com- Last checked
- Nov 20, 2026
Open source capture What the story claims, and what we checked
What we compared with the page.
- The figure: ~6xChecked
Printed word for word on the page, near the name of Recraft.
- The passage quoted aboveChecked
Copied word for word from the page, near the name of Recraft.
- The numbers in our summaryChecked
Each one is printed on the page.
- Recraft uses Nebius AI CloudChecked
Confirmed line.
- The result itselfNot checked
We quote it; we did not measure it.
After the initial problems were solved, Recraft noticed the training process was very slow, operating at an 8x slower rate than in an ideal scenario where the network is not a bottleneck.
