Cloud providers
Which clouds emma connects to, in which regions, and what emma can do on each.
emma is a control plane over the clouds you connect, and one of the providers itself. It manages resources in accounts you own — add a cloud under Settings → Cloud Providers → Add Provider, and emma provisions into it and adopts what you already run there — and it can also give you GPU capacity from its own fleet. This page lists the clouds emma supports and what it can do on each.
Connect once, then adopt. Import existing infrastructure covers connecting an account and bringing resources emma didn't create under management — evaluation never requires a migration.
Two ways to get capacity
You can get GPU capacity two ways:
- Rent emma's fleet. emma is itself one of the providers on the platform. Rent GPU capacity emma operates, billed by emma, with no cloud account of your own required.
- Provision into your own accounts. Run GPUs — and everything else — inside an AWS, GCP, Azure, or other account you already hold, billed by that provider at your own rates. Add a cloud under Settings → Cloud Providers → Add Provider.
Which GPU cards you can get depends on the provider and current availability — emma exposes the cards each provider supports and adds more as they do. emma's own fleet currently includes L40, A100, H100, H200, B200, and B300. You always see the exact set available for your provider and region when you pick a configuration, so this page names capability, not a frozen card list.
The providers emma connects to
Available today:
- emma — rent GPU capacity emma operates directly (see Two ways to get capacity above).
- Hyperscalers — Amazon Web Services, Google Cloud, Microsoft Azure.
- Regional & specialist clouds — DigitalOcean, Gcore, OVHcloud, IONOS, Nebius, Hetzner.
- Private cloud — VMware — bring emma to infrastructure you run yourself.
Coming soon: Alibaba Cloud, Exoscale, Leaseweb, Oracle Cloud (OCI), Scaleway, Vultr.
emma fetches the provider catalogue at runtime, so the set you can choose from when adding a cloud always reflects what the platform supports for your account right now.
What emma can do on each
emma provisions and manages the same resource types across the clouds it connects to. What's available on a given provider, and which GPU cards, depends on that provider — emma exposes what each supports and adds more over time, and you always see the live set for your provider and region when you pick a configuration.
| Capability | What it is | Where |
|---|---|---|
| Compute (VMs) | instance type, vCPU/RAM, GPU count, HPC shapes, bare metal, driver-ready OS images | connected clouds, per provider |
| GPU | GPU-backed instances for training and inference; exact cards depend on the provider and availability | connected clouds + emma's fleet |
| Spot | the same machine, interruptible, at a lower rate | AWS, Google Cloud, Azure |
| Managed Kubernetes | managed clusters and node groups, created and scaled like any other resource | AWS, Google Cloud, Azure (DigitalOcean and Gcore on request) |
| Block storage | persistent volumes, snapshots, and backups | per provider |
| Managed databases | provisioned managed database instances | AWS (RDS Postgres) today |
| Networking | project networks, subnets, security groups, public IPs | per provider |
| Multi-cloud network | private connectivity between resources on different providers, over emma's backbone | AWS, Azure, Google Cloud instances |
| Direct Connect | dedicated cross-cloud links, per macro-region (EMEA / AMER / APAC) | AWS, Google Cloud, Azure |
Regions
emma provisions into the regions each connected provider exposes; the exact set is read live per provider. Direct Connect is enabled per macro-region — EMEA, AMER, and APAC — each with its own real monthly cost you can see before enabling.
Where a provider's region sits legally — operating entity, governing jurisdiction, EU-controlled — is not covered here yet.
What's next
- Import existing infrastructure — connect a cloud and adopt what's already there
- Self-hosted inference — run models on a GPU you rent or own
- Observability — push metrics to your own stack
Last updated on 25 Aug 2026