SKIP_TO_MAIN_CONTENT
06.0 // Comparison Apples to apples Same YAML

BuildJet is gone.
Here's where to go next.

machine.dev is a drop-in alternative. Same managed SaaS model, one-line runs-on change — and you get GPU runners too.

01 Opening
[REF_001]

The situation.

BuildJet shut down on March 31, 2026. If you're reading this, you probably need to move your CI somewhere today. We're not going to make this a sales pitch — BuildJet was a decent service and their users are in an annoying spot.

The short version: if you were using BuildJet for CPU runners, the setup here is the same — change runs-on, push, done. It takes about five minutes. The pricing model is different (machine.dev is per-minute in USD with spot and on-demand; BuildJet was flat per-minute only). Use the free trial to compare actual costs before you commit.

02 Why machine dev
[REF_002]

Five reasons.
One migration.

[REF_010] MODEL // SAAS

Same model, more runners

Fully managed, no self-hosting, no infrastructure to babysit. x64 CPU runners from 2 to 64 vCPU. ARM64 Graviton too.

[REF_011] GPU // PRE_INSTALLED

GPU runners with CUDA ready

BuildJet had no GPU runners. machine.dev has T4G, T4, L4, A10G, L40S, and RTX 6000 — with CUDA 12.1, cuDNN 9.2.1, and NVIDIA drivers pre-installed.

[REF_012] STATE // EPHEMERAL

Clean machine, every job

Every run gets a fresh VM. No shared state, no runner recycling, no "works on my runner" bugs.

[REF_013] TENANCY // SPOT

Spot pricing available

Add tenancy=spot to your runner label for jobs that tolerate a retry. Significantly cheaper than on-demand.

[REF_014] EGRESS // BUNDLED

Data transfer is bundled

No ingress/egress fees. Pulling large dependencies or pushing build artifacts adds nothing to your invoice.

03 Migration steps
[REF_003]

Four steps.
Five minutes.

01 SIGN_UP

Connect_Github_Org

Go to machine.dev, sign up, and authorise machine.dev on your GitHub org. Takes about two minutes. No credit card required.

02 MAP_LABELS

Translate_Runner_Names

BuildJet used flat string labels like buildjet-8vcpu-ubuntu-2204. machine.dev uses an array syntax in runs-on:

BUILDJET_LABEL MACHINE_DEV_EQUIVALENT
buildjet-2vcpu-ubuntu-2204 [machine, cpu=2]
buildjet-4vcpu-ubuntu-2204 [machine, cpu=4]
buildjet-8vcpu-ubuntu-2204 [machine, cpu=8]
buildjet-16vcpu-ubuntu-2204 [machine, cpu=16]
buildjet-32vcpu-ubuntu-2204 [machine, cpu=32]
buildjet-2vcpu-ubuntu-2204-arm [machine, cpu=2, architecture=arm64]
buildjet-4vcpu-ubuntu-2204-arm [machine, cpu=4, architecture=arm64]
buildjet-8vcpu-ubuntu-2204-arm [machine, cpu=8, architecture=arm64]
buildjet-16vcpu-ubuntu-2204-arm [machine, cpu=16, architecture=arm64]
buildjet-32vcpu-ubuntu-2204-arm [machine, cpu=32, architecture=arm64]

Add tenancy=spot for spot pricing — e.g. [machine, cpu=8, tenancy=spot].

03 UPDATE_WORKFLOWS

Find_And_Replace

Find/replace your runs-on labels. If you have a lot of workflows, a simple grep does it:

$ grep -r "buildjet-" .github/workflows/

Then swap each one for the equivalent machine.dev label.

04 EXECUTE

Push_And_Test

Commit, push, watch the first job run. The machine.dev dashboard shows you job status and credit usage in real time.

04 Pricing
[REF_004] // LINUX_X64

Same or cheaper.
Plus spot.

Per-minute USD. BuildJet historical prices frozen at the March 31, 2026 shutdown. machine.dev prices pulled from the live API.

[NOTE_SPOT] BuildJet had no spot tier — just flat per-minute. machine.dev's spot is 8.6–12.5× cheaper than BuildJet's pay-as-you-go at every size, on x64 and ARM both.

[NOTE_OD] On-demand is 26% cheaper at 2–8 vCPU and roughly tied at 16–32 vCPU. If your workload was small/medium BuildJet runners, machine.dev on-demand is a free win. If you used the largest BuildJet sizes, on-demand is a wash — and spot is where the savings live.

04b Features
[REF_005]

Beyond price.

SPEC BUILDJET MACHINE_DEV
Status Shut down March 31, 2026 Active
Pricing model Flat per-minute, PAYG only Per-minute, USD, spot + on-demand
Spot tier No Yes (8.6-12.5× cheaper than BuildJet PAYG)
GPU runners No Yes (T4G, T4, L4, A10G, L40S, RTX 6000)
AWS Trainium / Inferentia2 No Yes
CUDA pre-installed No (no GPU) Yes (12.1 + cuDNN 9.2.1, drivers 555.58)
macOS / Windows No No (Linux only — same coverage)
Data transfer fees Unknown No (bundled)
Free tier $5 one-time credit $10 free compute, never expires, no card
Setup Find/replace runs-on label Find/replace runs-on label (same effort)
05 Gpu upsell
[REF_005]

BuildJet had
no GPUs.
We do.

If any of your jobs involve ML inference, model evaluation, or GPU-accelerated compute, that's something you simply couldn't do on BuildJet. machine.dev gives you six GPU tiers from T4G (16GB) to RTX 6000 (96GB).

T4G starts at $0.00372/min on spot. CUDA, cuDNN, and the NVIDIA Container Toolkit are pre-installed — no setup step required. torch.cuda.is_available() returns True on the first run.

VIEW_GPU_SPECS
06 Live pricing
[REF_PRICING] // LIVE_DATA

Live CPU Pricing

Live per-minute rates for CPU runners. Compare with BuildJet's flat $0.004/min.

07 Initialize
READY
> MIGRATION_READY // ETA_5_MIN

Your CI needs
to run somewhere.

Might as well be somewhere with GPUs. $10 free compute. Migrate in about five minutes.