Krevo Compute opens in private access today. H100 and H200 GPU clusters and dedicated hosting on the Krevo network, brought up in cohorts as rack capacity lands. Talk with sales to be considered for the next cohort.
Plans auto-select the right blend of CPU, GPU, and NVMe for your workload. No configuration knobs. Tell us what you are running, and we provision the right shape.
The first cohort is small. We are building a partner network for edge and direct provisioning for teams that need bare-metal.
Compute tiers
| Light Compute | Ultra Compute | GPU | |
|---|---|---|---|
| Best for | APIs, web apps, batch jobs | Latency-sensitive services | Training, fine-tuning, inference |
| Compute | Dedicated vCPU, burstable memory | Dedicated vCPU, reserved memory | H100 / H200 pools, NVLink |
| Storage | 100 GB | 1 TB | 4 TB |
| Network | 1 Gbps | 10 Gbps | 25 Gbps |
| Scheduling | Standard | 2.5× priority | 3× priority |
| Onboarding | Same-day | Same-day | 2 weeks |
Effort
Plans default to high effort. Teams can choose max for tighter p99. We recommend max for customer-facing APIs and long-running jobs.
What’s next?
Dedicated bare-metal, the edge partner network, and a custom contract path for regulated workloads are in progress.
We also plan to release a new class of plan with higher throughput than Ultra. A small number of organizations are using Krevo GPU Preview today. Plans at this level need stronger isolation before they can be generally available. GPU-class plans are coming for all customers.
Availability
Krevo Compute is available to a small cohort today. H100s are available today, and H200s are available in early October 2026. Full pricing is sent with the invitation.