What is Kestrel GPU Fleet?
A production-grade machine learning offering from Kestrel, Kestrel GPU Fleet balances cost and performance with automatic scaling and built-in redundancy. Pricing is published on a single page, quotas are documented per region, and support covers everything from capacity planning to incident response. Kestrel operates it as a fully managed service: patching, capacity, and failover are handled upstream, while access controls and spend alerts stay in your hands.
Key features
- Role-based access control with audit logging
- Drop-in Terraform provider and CloudFormation templates
- Live migration for planned maintenance, no downtime window
- Transparent egress rates published per region
- Granular metrics exported to your own monitoring stack
Typical use cases
- Data pipelines that need predictable throughput at a fixed cost
- Staging environments that mirror production configurations
- Regulated workloads that require audit trails and data residency
- Internal platforms serving dozens of product teams
Why Kestrel?
Kestrel publishes list prices for every region and never gates core capabilities behind enterprise-only tiers. Capacity is pre-provisioned, so the performance you benchmark in a sandbox is the performance you get in production. And because egress rates are flat, the bill at the end of the month matches the estimate you made at the start.
Getting started
Create an account, pick a region, and the default configuration is live in under a minute. The quickstart walks through your first deployment in about ten minutes, and the managed Terraform module turns the same setup into code you can review.
