ZB Cloud
Four kinds of cloud
Every cloud is best at something. ZB Cloud uses all of them, and Maestro routes every job to the one that runs it best.
| Kind | Best at | Examples |
|---|---|---|
| Hyperscalers | Reach: the big global clouds, with many regions and every kind of service | AWS, Google Cloud, Microsoft Azure |
| Sovereign clouds | Keeping data inside one country, under that country's law | In-country data centers, such as in Bangladesh |
| Neo centers | GPU power by the hour, in clouds built for AI | CoreWeave, Lambda, Nebius |
| Decentralized clouds | Spare machines from many independent owners, at a fraction of the price | Akash, io.net |
Hardware, jobs and data
| Kind | Hardware | Typical jobs | Where data lives |
|---|---|---|---|
| Hyperscalers | CPUs of every size; NVIDIA H100, H200 and B200 | Websites, apps, databases, file storage | In the region you choose, in many countries |
| Sovereign clouds | CPUs; NVIDIA H100 and H200 where installed | Patient records, government work, models trained on national data | Inside the country's borders |
| Neo centers | NVIDIA H100, H200 and B200; AMD MI300X | Model inference, training and fine-tuning, image and video generation | In the provider's regions, chosen per job |
| Decentralized clouds | A wide mix of GPUs and CPUs | Nightly batch jobs, rendering, tests, anything that can run anywhere | Spread across many owners; used for jobs with no location rule |
Choosing the cloud yourself
You can, on any job. Pin it to one cloud or one region and the Cloud Router keeps it there. You will rarely need to. See Job files.
Was this page helpful?Thank you. Tell us more at apply@zbstack.xyz.