Enterprise cloud on the same network as your GPUs.
Virtual machines, storage, private networking, and backup in the same facility as our GPU clusters. For workloads that need a known location and a predictable bill.
- Your applicationsAPIs, data pipelines, and vector stores
- Managed KubernetesUpstream control plane, run by us
- ComputeVMs and dedicated hosts
- StorageBlock, object, and file
- NetworkPrivate networks, load balancing, and VPN
Planned services at launch.
None of these services are available yet. They are planned for our first deployment, for design partners first and then for all customers at general availability.
| Compute | |
|---|---|
| Virtual machines | General-purpose and memory-optimized instances on current-generation x86 servers. |
| Dedicated hosts | Physical servers reserved for one customer, for licensing, isolation, or compliance needs. |
| Managed Kubernetes | Upstream Kubernetes with the control plane run by us, on VMs or dedicated hosts. |
| Storage | |
| Block storage | Replicated NVMe-backed volumes with snapshots, attached to VMs and hosts. |
| Object storage | S3-compatible buckets for datasets, backups, logs, and model artifacts. |
| File storage | Shared NFS file systems for applications that expect a POSIX mount. |
| Network | |
| Private networks | Isolated networks per customer and environment, with your own address space. |
| Load balancing | Layer 4 and layer 7 load balancers with health checks and TLS termination. |
| Private connectivity | Cross-connects and site-to-site VPN into your offices or other clouds. |
| Resilience | |
| Backup | Scheduled backups of volumes and buckets with retention you define. |
| Disaster recovery | A secondary copy of critical systems and data, with a recovery plan you can test. |
| Private cloud | A dedicated environment built for one organization, on hardware no one else uses. |
One facility, one network, one contract.
Next to your GPUs
Data preparation, vector databases, and the applications that call your models sit on the same network as the cluster, without transfer between providers.
A facility you can name
Your data stays in one U.S. facility that your security and compliance teams can identify, assess, and visit.
Costs you can forecast
Committed-use rates with data transfer terms written into the agreement, so the monthly bill matches the budget.
Help moving in
Our engineers plan migrations with your team and stay involved until workloads are running in production.
Enterprise cloud questions
Are these services available today?
No. Cloud services launch with our first deployment and are available to design partners first. The list above is our planned launch catalog.
Can we run cloud services without buying GPU capacity?
Yes. Cloud services are sold on their own. Many workloads will never need a GPU.
Do you support hybrid setups with AWS, Azure, or Google Cloud?
Yes. Private connectivity and VPN let you run part of a system with us and part elsewhere. We will help design the split.