Ten platforms, rated 0 to 3 on six criteria and weighted as set out in how we rank. A high score means a platform does well on what these criteria value, which puts weight on running in your own cloud. If that matters less to you, reweigh the criteria and watch the order change.
A PaaS that deploys into your own AWS, GCP or Azure account.
Verdict: The strongest pick when your apps must run in your own cloud account and you don't want to hire DevOps to get there; skip it if you want nothing to do with a cloud account.
Best fitStartups and growth-stage teams without a dedicated DevOps engineer who want push-to-deploy, but need their apps, data and cloud bill in their own AWS, GCP or Azure account: for compliance, cloud credits, or a customer's security review.
Not forTeams that want a fully vendor-hosted platform with no cloud account to own, side projects that would fit a $5 tier, and anyone who needs to deploy outside AWS, GCP and Azure.
Strengths
Deploys into your own AWS, GCP or Azure account, so data stays in your cloud and cloud credits pay for the compute.
Its docs describe one-click SOC 2 and HIPAA-compliant infrastructure for what it manages.
GPU instances for training and inference, from your own cloud's capacity.
Its pricing page states that if you stop paying, your servers keep running: the Kubernetes cluster stays in your account.
Limits
Two bills: Porter's fee plus your cloud provider's, which you have to estimate yourself.
You own a cloud account, with its quotas, permissions and support plan.
A Kubernetes cluster sits underneath, and a cluster has baseline cloud costs, so it is rarely the cheapest home for a single tiny app.
Its one-click compliance doc is written about the AWS resources Porter manages; check coverage on GCP and Azure for your case.
What it costs
$13 per vCPU and $6 per GB of RAM per month for the resources your apps request, plus your cloud provider's bill. Volume discounts above 40 vCPU / 80 GB RAM, 50% off for non-profits, and a startup deal.
On the criteria
Where apps run (25)
Strong
Runs in your own AWS, GCP or Azure account (its docs: "a platform as a service that runs in your own AWS, GCP, or Azure account").
Compliance (20)
Strong
Holds a SOC 2 Type II report; its docs describe one-click SOC 2 and HIPAA-compliant infrastructure for the AWS resources it manages (EKS, RDS, S3 and more).
GPU support (15)
Strong
GPU instances listed on its home page, with a docs guide to running GPU workloads.
Cost model (15)
Strong
Published rates per requested vCPU and GB of RAM, prorated to the minute; your cloud bill is separate and credits can cover it.
Ease of exit (15)
Strong
Its pricing FAQ: if you stop paying, servers keep running and the Kubernetes cluster stays intact in your account.
Operations handled (10)
Good
Handles cluster setup, CI/CD, autoscaling, patching and upgrades, but you still own and pay for the cloud account underneath.
Containers, jobs and databases on Northflank's cloud or in your own.
Verdict: The most flexible option here, strong on both hosting models and on GPUs; compliance paperwork comes through its enterprise tier.
Best fitTeams that want to start vendor-hosted and move into their own VPC later without switching platforms, and AI teams that want GPUs across several clouds from one place.
Not forTeams that need SOC 2 and HIPAA paperwork on a self-serve plan, or that want the fewest possible concepts to learn.
Strengths
Both hosting models: Northflank Cloud, or bring your own cloud (AWS, Google Cloud, Azure, Oracle, bare metal).
Published per-second rates, with no seat fees.
GPUs on its own cloud (NVIDIA L4 up to B200 listed) and any GPU your cloud offers in BYOC.
A free Sandbox tier that includes one BYOC cluster.
Limits
SOC 2 Type II, HIPAA and ISO 27001 are presented under its enterprise offer.
Broad feature set (services, jobs, add-ons, pipelines) means more to learn than the simplest platforms.
In BYOC mode you pay Northflank per node vCPU and your cloud provider for the nodes.
What it costs
Free Sandbox; pay as you go from $0.01667 per vCPU-hour and $0.00833 per GB-hour on Northflank Cloud; BYOC at $0.02083 per node vCPU-hour plus your cloud bill. GPUs on Northflank Cloud from $0.80 an hour (L4).
On the criteria
Where apps run (25)
Strong
Northflank Cloud, or fully managed clusters in your own VPC on AWS, Google Cloud, Azure, Oracle or bare metal.
Compliance (20)
Good
SOC 2 Type II, HIPAA and ISO 27001 listed in its enterprise section; no public price for that tier.
GPU support (15)
Strong
GPU rates published for its cloud; in BYOC, any GPU your cloud offers at $1.00 per GB of VRAM per month.
Cost model (15)
Strong
Per-second usage billing with published rates and no seat fees, on its cloud and in BYOC.
Ease of exit (15)
Good
Workloads are standard containers; in BYOC the cluster is in your account. It does not publish what keeps running if you leave.
Operations handled (10)
Good
Builds, deploys, previews, logs and autoscaling are handled; more configuration surface than push-and-forget platforms.
A self-service deployment layer on your own Kubernetes cluster.
Verdict: A strong own-cloud choice for teams that already think in Kubernetes; the flat plan pricing suits mid-size teams better than a tiny startup.
Best fitTeams that already run Kubernetes (EKS, GKE, AKS, k3s or on-premise), platform teams that want a self-service layer for developers, and European teams that need Scaleway or DORA coverage.
Not forTwo-person startups that want usage-based pricing, and teams with nobody comfortable around Kubernetes when something breaks.
Strengths
Runs on a Kubernetes cluster you own, or one it creates on AWS, GCP, Azure or Scaleway.
A free plan on your own cluster with no time limit.
SOC 2 Type II report on its Scale plan, a HIPAA add-on, and HIPAA, HDS and ISO 27001 on Enterprise.
Your workloads stay in your account, so leaving means removing a control layer, not moving apps.
Limits
Flat plan pricing ($299 to $2,999 a month, billed yearly) on top of your cloud bill.
Plans cap users and deployment minutes.
GPU support follows your cloud: its docs list GPU instance types for AWS clusters.
What it costs
Free on your own cluster (3 users, 1,000 deployment minutes a month); Starter $299, Growth $899 and Scale $2,999 a month, billed yearly; plus your cloud provider's bill.
On the criteria
Where apps run (25)
Strong
Installs on your cluster (EKS, GKE, AKS, k3s, on-premise) or creates one on AWS, GCP, Azure or Scaleway.
Compliance (20)
Good
SOC 2 Type II report and DPA on Scale ($2,999 a month); HIPAA as an add-on without a public price; HIPAA, HDS and ISO 27001 on Enterprise.
GPU support (15)
Good
Its cluster docs list GPU instance types among those available on AWS.
Cost model (15)
Good
Published flat plans billed yearly, with user and deployment-minute limits; cloud costs are separate.
Ease of exit (15)
Strong
Workloads run on your own cluster in your own account.
Operations handled (10)
Good
Handles deploys, previews, cluster upgrades and patches; you still own the cloud account and the cluster.
Google's serverless container platform, inside your own Google Cloud project.
Verdict: The best value for teams already on Google Cloud, especially for GPU inference that scales to zero, but more a building block than a complete platform.
Best fitTeams already on Google Cloud, containerised APIs with bursty traffic, and GPU inference that should scale to zero when idle.
Not forTeams that want multi-cloud, or a complete app platform with databases, previews and pipelines in one dashboard.
Strengths
Runs in your own Google Cloud project, billed on your Google Cloud account.
Per-second billing with a monthly free tier, and scale to zero.
GPUs (NVIDIA L4 and RTX PRO 6000) that also scale to zero.
Covered by Google Cloud's HIPAA BAA, which spans its whole infrastructure.
Limits
Google Cloud only.
A building block: CI/CD, databases, secrets and networking come from other Google Cloud products you set up.
Google Cloud's IAM and project model take time to learn.
What it costs
Billed per second of CPU, memory and, if attached, GPU, after a monthly free tier; rates vary by region and billing mode. See Google's Cloud Run pricing page and calculator.
On the criteria
Where apps run (25)
Good
Runs in your own Google Cloud project; one cloud only.
Compliance (20)
Strong
Google Cloud's BAA covers its entire infrastructure; Google publishes SOC 2 reports for Google Cloud.
GPU support (15)
Strong
NVIDIA L4 and RTX PRO 6000 GPUs, one per instance, with scale to zero.
Cost model (15)
Strong
Published per-second rates for CPU, memory and GPU, with a free tier.
Ease of exit (15)
Good
Deploys standard container images; the surrounding IAM, networking and services are Google Cloud's.
Operations handled (10)
Limited
Runs containers and scales them; builds, databases and environments are yours to assemble.
Usage-billed app and database hosting with a very fast start.
Verdict: The quickest way for a small team to get an app and its database running; look elsewhere once compliance or GPUs are on the list.
Best fitSolo developers and small teams who want the shortest path from a repo to a running app with its database.
Not forRegulated workloads without an enterprise contract, GPU workloads, and teams that need to run in their own cloud on a standard plan.
Strengths
Per-second usage billing with small plan minimums.
Databases and services side by side in one project canvas.
High per-service limits on Pro (up to 1,000 vCPU / 1 TB RAM per service).
Limits
HIPAA BAAs, SSO and bring-your-own-cloud are Enterprise features.
No GPU instances on its pricing page.
Vendor-hosted on standard plans.
What it costs
Free trial with $5 of credit for 30 days; Hobby $5 minimum usage and Pro $20 minimum usage a month, then per-second usage (CPU $0.00000772 per vCPU-second, memory $0.00000386 per GB-second); egress $0.05 per GB.
On the criteria
Where apps run (25)
Good
Vendor-hosted with global regions; bring your own cloud is listed under Enterprise.
Compliance (20)
Good
SOC 2 and HIPAA BAAs appear in its plan comparison, with BAAs on Enterprise (custom pricing).
GPU support (15)
None
No GPU instances listed on its pricing page.
Cost model (15)
Strong
Published per-second rates with $5 and $20 monthly minimums that count as usage credit.
Ease of exit (15)
Good
Builds from your repo or a Dockerfile; databases are standard engines you can dump and restore.
Operations handled (10)
Strong
Push code, and builds, deploys, TLS and logs are handled.
Fast-booting VMs on Fly's own hardware, close to your users.
Verdict: Excellent for globally distributed apps and cheap small machines; expect to do more of the operating yourself than on Render or Heroku.
Best fitApps that benefit from running close to users in many regions, and teams comfortable with a CLI and a config file.
Not forTeams that want a hands-off dashboard, and roadmaps built around GPUs.
Strengths
Fly Machines in 18 regions, billed by the second.
Small machines are cheap: the smallest preset is listed at $2.19 a month.
HIPAA package at a published $99 a month with a pre-signed BAA, on any paid plan; SOC 2 Type II.
Limits
Its own blog post "We Were Wrong About GPUs" scaled back its GPU bet; check its docs for current GPU availability.
More of the operating is yours (machines, volumes, regions) than on Render or Heroku.
Runs only on Fly's hardware.
What it costs
Per-second Machines billing, from $2.19 a month for shared-cpu-1x with 256 MB (prices vary by region); volumes $0.15 per GB-month; support plans from $29 a month; HIPAA package $99 a month.
On the criteria
Where apps run (25)
Limited
Vendor-hosted on hardware Fly owns, with a choice of 18 regions.
Compliance (20)
Strong
SOC 2 Type II; HIPAA package at $99 a month with a pre-signed BAA, available on any paid plan.
GPU support (15)
Limited
Its post "We Were Wrong About GPUs" describes scaling its GPU bet back, and its pricing page lists no GPU presets; check its docs for current availability.
Cost model (15)
Strong
Published per-second rates by machine preset and region.
Ease of exit (15)
Good
Deploys Docker images; fly.toml and Machines APIs are Fly-specific.
Operations handled (10)
Good
CLI-first: deploys, TLS and routing are handled, but you choose and manage machines, volumes and regions.
Web services, workers, cron jobs and Postgres in one vendor-hosted platform.
Verdict: The most Heroku-like experience on a more modern price structure; a safe default if vendor hosting is fine and you don't need GPUs.
Best fitTeams leaving Heroku who want a similar experience, and teams who want web services, workers, cron jobs and Postgres in one place.
Not forGPU workloads, and teams that need to run in their own cloud account.
Strengths
Heroku-like simplicity, with preview environments and infrastructure as code.
SOC 2 Type II and ISO 27001; HIPAA-compliant workspaces on its Scale plan.
Native runtimes or Docker, so apps stay portable.
Limits
No GPU instances on its pricing page.
HIPAA-enabled workspaces carry a 20% compute premium.
Vendor-hosted only.
What it costs
Workspace plans: Hobby $0, Pro $25 a month, Scale $499 a month, Enterprise custom, each plus compute billed by instance size; HIPAA workspaces add 20% to compute.
On the criteria
Where apps run (25)
Limited
Vendor-hosted, with a choice of regions.
Compliance (20)
Strong
SOC 2 Type II and ISO 27001; HIPAA BAA on Scale ($499 a month) and above, with a 20% compute premium.
GPU support (15)
None
No GPU instances on its pricing page.
Cost model (15)
Good
Published workspace fees plus compute by fixed instance size.
Ease of exit (15)
Good
Native runtimes or Docker images; Postgres can be dumped and restored elsewhere.
Operations handled (10)
Strong
Push code; builds, deploys, TLS, autoscaling and previews are handled.
The original git-push PaaS, with a large add-on marketplace.
Verdict: Still the reference for simple deploys and add-ons, but you pay for it at scale, and compliance means Shield dynos and Private Spaces.
Best fitTeams already running on it, apps that integrate closely with Salesforce, and teams that lean on its add-on marketplace.
Not forGPU workloads, cost-sensitive scaling, and teams that need their own cloud account.
Strengths
The reference experience for git-push deploys, with buildpacks for many languages.
A large marketplace of add-ons.
Shield dynos sign BAAs for HIPAA and support PCI compliance.
Limits
Compliance needs Shield dynos (from $150 a month each) inside a Private Space.
No GPU dynos on its pricing page.
Fixed dyno sizes, priced per month, so the bill climbs in steps as you scale.
What it costs
Dynos from Eco $5 a month (sleeps after 30 minutes of inactivity), Basic $7, Standard-1X $25; Shield-S $150 and Shield-M $300, which require a Private Space.
On the criteria
Where apps run (25)
Limited
Vendor-hosted, with a choice of regions through Private Spaces.
Compliance (20)
Strong
Shield dynos: the ability to sign BAAs for HIPAA, plus PCI compliance; a Private Space is required.
GPU support (15)
None
No GPU dynos on its pricing page.
Cost model (15)
Limited
Published dyno prices, but the Private Space that compliance tiers require is not priced on the dyno table.
Ease of exit (15)
Good
Buildpacks and Docker; add-ons have to be replaced one by one.