Porter

What Porter actually is

Porter is a Platform-as-a-Service layer that runs inside your own cloud account. Connect an AWS, Azure, or Google Cloud account, and Porter provisions a production-ready Kubernetes cluster there, then gives you a Heroku-style workflow on top: point at a Git repository, every push builds and ships with zero-downtime deploys, and every pull request gets its own preview environment. Underneath it is Kubernetes — with Karpenter bin-packing workloads, automatic CVE patches, and cluster version upgrades — but the pitch is that nobody on your team needs to know that.

The company was founded in 2020 by Justin Rhee and Trevor Shim, went through Y Combinator's Summer 2020 batch, and is headquartered in New York City. It started life as a remote development environment product, pivoted, briefly ran as an open-source project (the old porter-dev/porter repository drew a Hacker News crowd), and relaunched self-serve in 2021 with an internal control system for managing clusters distributed across customer cloud accounts. A $20M Series A followed in January 2026, and the current marketing leans heavily on AI startups: GPU instances for training and inference, one-click SOC 2 and HIPAA, and versioned rollbacks for teams that ship model-backed products.

The funding history is modest by sector standards, which cuts both ways. A $1.5M seed in July 2021 — Y Combinator, Venrock, Translink Investment, Soma Capital, and angels — kept the company lean through its self-serve relaunch, and a Heroku Migration Program announced in April 2022 gave it a clear wedge: applications leaving Heroku needed somewhere to go that did not require a platform team. The January 2026 Series A improves the odds of Porter being a long-term vendor, but this is still a company that has survived by picking its niches carefully.

The architecture decision that shapes everything is the BYO-cloud model. Your workloads, data, and network live in your AWS/Azure/GCP account, on your bill with your cloud credits where applicable. Porter's control plane — what it calls the Cluster Control Plane — monitors and manages the cluster, watched by its own SREs around the clock. In our view this is the middle path between shared-node PaaS products (where noisy neighbors and platform outages are your problem but not your fault) and running Kubernetes yourself (where everything is your problem). Reliability complaints about shared infrastructure simply do not apply here, because there is no shared infrastructure.

Pricing: what you really pay

Porter charges for two things separately, and keeping them separate is the key to reading the bill. First, Porter's own fee: on the Standard plan it is $6 per month per GB of RAM and $13 per month per vCPU, metered on the resources your applications request — not on cloud instance capacity — and prorated to the minute. Second, the underlying cloud bill, paid directly to AWS, Azure, or GCP; Porter does not mark it up, and cloud credits can cover it. Their own worked example: a workload of 2 vCPU and 4GB RAM runs about $112/month on AWS App Runner, versus roughly $80/month total through Porter including the EC2 instances. The pricing calculator on the pricing page shows how fine-grained this gets — a 0.1 vCPU, 0.1GB service computes to $1.90 per month in Porter fees.

Enterprise is a volume-discount track starting at 40 vCPU and 80GB RAM, adding premium support, advanced RBAC, SAML-based SSO, custom alerts, and an on-premises installation option. There is also a startup deal — six months free — aimed at early companies; note that this is a promotion rather than a permanent free tier, and no standing free plan exists on the current pricing page.

Two things we like here. The metering basis (requested app resources) means bin-packing saves you real money on the cloud side without Porter skimming the difference. And the exit terms are unusually clean: if you stop paying, the Kubernetes cluster stays up and intact — you simply operate it yourself from then on. We have seen plenty of PaaS products where the bill is a mystery and the exit is a migration project; Porter is neither. The honest caveat is that your total cost is two invoices, and predicting it means understanding both.

Performance & infrastructure

There are no Porter datacenters to benchmark — the infrastructure is whatever regions you choose in your own cloud account. EU, US, and APAC deployment is therefore entirely available, but as a property of AWS/Azure/GCP rather than of Porter. Performance for application workloads is native cloud performance: EC2/GCE/Azure VMs with Kubernetes on top, no extra virtualization layer, and Karpenter-style bin-packing to keep node utilization high. The one number Porter publishes of its own — the App Runner comparison above — is a cost comparison, not a speed claim, and we would treat it accordingly.

What Porter actually contributes is operational infrastructure: VPC and networking setup, load balancing and DNS, certificate management, monitoring and logging out of the box, alerting, autoscaling on usage or custom metrics, jobs and cron jobs, and an upgrade pipeline that keeps cluster versions and CVE patches moving without a maintenance window from your team. SOC 2 Type II is in hand, and SOC 2 and HIPAA configurations are pitched as one-click. For a startup without a platform engineer, that bundle is the product; the Kubernetes cluster is just where it lands.

GPU instances for training and inference are available through the same flow, which is why the AI-startup positioning is not just marketing: a team that needs to ship an inference endpoint next to their web app can do it in the same account and the same workflow.

Where it falls short

Setup friction is the first real cost. Unlike Render or Railway, where you sign up and deploy, Porter needs a cloud account, credentials, and roughly half an hour while infrastructure provisions — plus the ongoing reality that your bill is two invoices from two vendors. Teams looking for the shortest path from repository to URL will find faster options.

Second, you are still, at some level, running Kubernetes. Porter abstracts it well for the day-to-day, but when a networking issue or a node-level problem surfaces, the debugging surface is Kubernetes-shaped. The no-lock-in exit (you keep the cluster) is also, read the other way, an admission that the exit is: you now operate a Kubernetes cluster. Budget for that possibility.

Third, this is a venture-funded company that has pivoted more than once — remote dev environments to PaaS, open source to hosted, self-serve to upmarket and back. The $20M Series A of January 2026 is a sign of health, but buyers should weigh platform risk accordingly. Support on the Standard tier is not the premium package; enterprise-grade support lives behind the Enterprise track. And small teams should watch metered costs creep: every requested vCPU and gigabyte bills, idle staging environments included.

Finally, the provider list has narrowed with time — the current site is AWS, Azure, and Google Cloud, where earlier materials mentioned DigitalOcean as well. If you are on DigitalOcean today, confirm support before planning a move.

Who should pick this

Our pick for growth-stage engineering teams — AI startups squarely included — that want Kubernetes-grade scalability and GPU workloads without hiring DevOps, and that want their data and network to stay inside their own cloud account for compliance or cost reasons. It is also a strong candidate for teams migrating off Heroku or a shared-node PaaS because reliability has become a problem, and for organizations with AWS credits to spend.

We would look at Render or Railway instead when the team is small, the app is simple, and speed of deployment matters more than infrastructure control. We would look at managed Kubernetes from the cloud providers directly when you already have the platform skills and want to own the stack. For the wider field, see managed cloud and managed Kubernetes.

Key takeaways

• PaaS experience on top of Kubernetes clusters provisioned in your own AWS, Azure, or GCP account — your data and bill stay with your cloud provider.
• Metered pricing: $6/GB RAM and $13/vCPU per month on requested resources, prorated by the minute, on top of cloud costs; Enterprise adds volume discounts from 40 vCPU.
• No permanent free tier; a six-month startup promotion exists. Exit is clean — the cluster survives if you stop paying.
• Operational toil it removes: CI/CD, preview environments, autoscaling, monitoring, CVE patches, cluster upgrades, and one-click SOC 2/HIPAA.
• Trade-offs: ~30-minute provisioning and two invoices versus PaaS peers, Kubernetes-shaped debugging when things go deep, and a pivot-prone startup history behind the product.

Sources

• https://www.porter.run/
• https://porter.run/pricing
• https://ycombinator.com/companies/porter
• https://porter.run/blog/our-journey-building-porter
• https://fundraisingfox.com/companies/porter