AWS and Railway are both places to run software, and the pricing models reveal what each one is. AWS bills for provisioned infrastructure. Its public on-demand pricing for EC2 in us-east-1 lists a t3.medium with 2 vCPU and 4 GiB of memory at $0.0416 an hour, and the cheaper t3a.medium at $0.0376 an hour, both Linux and EBS-only. Data transfer in is free, and AWS includes 100 GB of data transfer out to the internet each month, aggregated across services and regions. Everything above that, and every additional service from load balancers to databases to monitoring, is billed separately.
TL;DR
AWS sells cloud infrastructure by the instance: an EC2 t3.medium with 2 vCPU and 4 GiB is $0.0416 an hour on-demand in us-east-1, and a t3a.medium is $0.0376, with 100 GB of monthly data transfer out included across services. Railway sells application hosting by the second: Free runs $0 then $1 a month, Hobby is a $5 monthly minimum and Pro a $20 minimum, with CPU at $0.00000772 per vCPU-second, memory at $0.00000386 per GB-second and egress at $0.05 per GB. AWS gives you the whole cloud and the responsibility that comes with it; Railway gives you a deployment target that scales down to zero cost when idle.
Railway bills for what an application actually consumes. The free tier runs $0 and then $1 a month, capped at 1 vCPU and 0.5 GB of RAM per service with 0.5 GB of volume storage. Hobby is a $5 monthly minimum that includes $5 of usage credit, and Pro is a $20 minimum with a $20 credit. Past the credit, usage is metered by the second: CPU at $0.00000772 per vCPU-second, memory at $0.00000386 per GB-second, volumes at $0.00000006 per GB-second and egress at $0.05 per GB, with object storage at $0.015 per GB-month.
The practical difference is the surface area you operate. AWS exposes hundreds of services and hands you the configuration: instances, security groups, subnets, scaling policies and the monitoring to know whether any of it is right. Railway exposes a deployment target: push code, get a running service, pay for the seconds it runs. That convenience is the product, and it costs more per unit of compute than raw EC2 while saving the operational work that raw EC2 requires.
| Feature | AWS | Railway |
|---|---|---|
| Free tier | Free tier services with time or usage limits | Free plan at $0 then $1 a month within caps |
| Entry paid plan | EC2 t3.medium at $0.0416 an hour on-demand | Hobby at $5 monthly minimum with $5 usage credit |
| Upper tier | Instance family and size of your choosing | Pro at $20 monthly minimum with $20 usage credit |
| Billing unit | Provisioned hours and service usage | Per-second CPU, memory, volumes and egress |
| Compute rate | t3.medium $0.0416, t3a.medium $0.0376 per hour | CPU $0.00000772 per vCPU-second |
| Memory rate | Included in the instance price | Memory $0.00000386 per GB-second |
| Data transfer out | 100 GB a month included, then metered | Egress $0.05 per GB |
| Storage | EBS volumes billed per GB-month | Volumes $0.00000006 per GB-second; object storage $0.015 per GB-month |
| Idle cost | Continues while instances are provisioned | Drops to the plan minimum when idle |
| Operating model | You configure the infrastructure | Deployment target with configuration handled |
Pick AWS if
AWS is the pick when the application needs managed databases, queues, load balancers, edge network and regional control that a single hosting platform cannot supply. The instance rates are lower, t3.medium at $0.0416 an hour, and the included 100 GB of monthly data transfer out is generous. It is the cheaper per unit of compute and the more expensive per hour of engineering time, so the trade is explicit.
Pick Railway if
Railway is the pick for a team that wants to deploy an app without operating a cloud. The $5 and $20 minimums include matching usage credit, and billing by the second means cost follows consumption rather than provisioned capacity. It costs more per unit of compute than EC2, and the free and hobby tiers cap resources, but the operational work it removes is the reason to pay.
Railway runs workloads on cloud infrastructure and layers a platform on top. You are paying for that platform's deployment and metering, which is why its per-second rates sit above raw EC2 prices.
For a small, always-on app, Railway's Hobby tier at $5 a month with $5 of included credit can be cheaper than a provisioned instance. For a busy steady workload, EC2's $0.0416 an hour for a t3.medium usually works out lower.
Storage, load balancing, data transfer above the included 100 GB a month, and any additional services such as managed databases or monitoring. The instance hour is one line among several.
By the second: CPU at $0.00000772 per vCPU-second, memory at $0.00000386 per GB-second, volumes at $0.00000006 per GB-second and egress at $0.05 per GB, applied after the monthly usage credit.
Within its model, yes, with replicas and larger resource allocations. It does not expose the breadth of AWS services or the same low-level networking control, which is the trade for the simpler operating model.
Yes, if instances are provisioned, they bill by the hour regardless of traffic. Railway's model bills for consumption, so an idle service drops to the plan minimum.
Spot an outdated price or feature? Let us know and we'll verify it.