Roboflow + Amazon EC2
Run Roboflow models and Workflows on an EC2 instance in your own AWS account, with one Inference CLI command to provision a GPU or CPU server.
How to connect Amazon EC2 to Roboflow
5 stepsInstall the CLI and SkyPilot
On your own machine, run `pip install inference "skypilot[aws]"`.
Authenticate with AWS
Follow SkyPilot's cloud account setup so it can create instances in your AWS account, using your configured AWS CLI credentials.
Deploy the server
Run `inference cloud deploy --provider aws --compute-type gpu`, or `--compute-type cpu` for a CPU instance. The command prints the deployment name and the server address, for example http://<ip>:9001.
Point your client at the server
Set `api_url` in the Python SDK (or your HTTP requests) to the instance address on port 9001, then run a model or Workflow with your Roboflow API key.
Stop the instance when idle
Use `inference cloud stop <name>` to pause it (you pay only for disk while stopped), `inference cloud start` to resume and `inference cloud undeploy` to delete it.
What teams build with Amazon EC2
Keep images inside your AWS account
Process images on a VM you control when data residency or security rules rule out sending frames to a hosted API.
Run a fixed-cost GPU server
Pay for the instance instead of per-inference credits when you send a steady, high volume of images.
Serve Workflows next to your data
Run the same Workflows you built in Roboflow on a server in the AWS region where your cameras or storage already live.
Questions
Not with `inference cloud deploy`. The CLI provisions the VM through SkyPilot and installs the Inference server for you. If you launch the instance yourself, follow the Linux install guide to set up Docker and the CPU or GPU image.
No. Self-hosted inference does not use credits; you pay AWS for the machine.
The deploy command opens port 9001 automatically. Check that this is acceptable for your account, and review Roboflow's guide to securing a self-hosted server before exposing it beyond a private network.
Object detection, segmentation, classification and keypoint models from your Roboflow workspace, plus foundation models such as CLIP, PaliGemma and SAM2.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
