Developer-first bare metal with a proper REST API, a maintained Terraform provider, and hourly billing on stock configurations. BareRoot keeps a deliberately small catalogue of five machine types across four regions, provisions most of them in under three minutes, and exposes IPMI, rescue mode and custom iPXE boot through the same API. Support is ticket-only by design and the docs do most of the talking.
Share your experience to help others choose the right provider.
We spin up twelve machines for four hours before every major release and tear them down after. On any other dedicated host that is a monthly commitment we would eat for no reason. Billing has been accurate to the cent every month, no surprise line items, no minimum term hidden in the ToS. Network to our EU users is solid, US west a bit less impressive but fine.
For eight months this was the best hosting decision we made. Then a NIC started flapping and I learned what ticket-only support really means. First response took nineteen hours, and the replacement window they offered was two days out. I moved the workload myself in the meantime. The automation is genuinely excellent - just do not build anything on the assumption that a human will pick up quickly.
I have wanted this from a bare-metal vendor for about a decade. The Terraform provider is real, it is versioned properly, and destroying and recreating a node is genuinely a normal operation rather than an act of desperation. We now rebuild our whole CI fleet from scratch every Sunday night and it costs us almost nothing because of the hourly billing. Support is slow but honestly I have opened two tickets in eleven months.
Five machine types sounds limiting until you realise most of what you actually run is one of five machine types. Ours is the mid AMD box and it has been rock solid. Custom iPXE boot through the API let us stop maintaining our own PXE infrastructure entirely, which removed a whole class of 3am problems. Docs are the best I have read in this space.
Ordered two nodes, one of them came up with out-of-band access that just refused my credentials. The API said the machine was healthy, the KVM console said otherwise. Ticket sat unanswered for about 22 hours, then got closed with a one-line "reprovisioned, please retry". It did work after that but no explanation and no apology. I still use them for disposable CI runners, I would not put anything that matters here.
We spin up twelve machines for four hours before every major release and tear them down after. On any other dedicated host that is a monthly commitment we would eat for no reason. Billing has been accurate to the cent every month, no surprise line items, no minimum term hidden in the ToS. Network to our EU users is solid, US west a bit less impressive but fine.
For eight months this was the best hosting decision we made. Then a NIC started flapping and I learned what ticket-only support really means. First response took nineteen hours, and the replacement window they offered was two days out. I moved the workload myself in the meantime. The automation is genuinely excellent - just do not build anything on the assumption that a human will pick up quickly.
I have wanted this from a bare-metal vendor for about a decade. The Terraform provider is real, it is versioned properly, and destroying and recreating a node is genuinely a normal operation rather than an act of desperation. We now rebuild our whole CI fleet from scratch every Sunday night and it costs us almost nothing because of the hourly billing. Support is slow but honestly I have opened two tickets in eleven months.
Five machine types sounds limiting until you realise most of what you actually run is one of five machine types. Ours is the mid AMD box and it has been rock solid. Custom iPXE boot through the API let us stop maintaining our own PXE infrastructure entirely, which removed a whole class of 3am problems. Docs are the best I have read in this space.
Ordered two nodes, one of them came up with out-of-band access that just refused my credentials. The API said the machine was healthy, the KVM console said otherwise. Ticket sat unanswered for about 22 hours, then got closed with a one-line "reprovisioned, please retry". It did work after that but no explanation and no apology. I still use them for disposable CI runners, I would not put anything that matters here.