Usage-based or flat PaaS pricing: what bandwidth, idle time and the US dollar do to your bill.

Every figure below comes from the vendor’s own pricing page or docs, read on 2026-10-29. The four US platforms print prices in US dollars and add sales tax on top. The table sticks to the three things that move a small team’s bill: the base fee, what happens past the bandwidth allowance, and what an idle app costs.

Bandwidth is the charge people underestimate, because it scales with visitors and asset size rather than with anything you deploy. Images, video, a large JavaScript bundle and file downloads all count. So do bots.

Here is the bandwidth part of the bill for an app that sends 500 GB a month, on the cheapest plan of each platform that can carry it. It leaves out compute and requests, which are metered separately everywhere except Canner.

• Vercel looks cheapest here because 500 GB fits inside Pro’s 1 TB. Double the traffic and every extra GB costs $0.15, and Pro’s functions and requests are metered on top.
• Netlify’s credits are shared. Bandwidth, compute, requests and deploys all draw from the same 3,000. A month with heavy builds leaves less for traffic.
• Railway is cheap per GB but meters everything from the first gigabyte, and RAM is billed for every second the service runs.
• Canner Starter includes 250 GB of traffic a month, so an app at 500 GB belongs on Live. Live has no metered bandwidth; heavy, sustained transfer is handled under the Acceptable Use Policy rather than billed per GB.

Usage-based compute sounds like it should cost nothing at 3 a.m. It depends on how the platform defines idle.

Railway bills a running service per second, busy or not. Its opt-in Serverless mode sleeps a service 5 to 10 minutes after the last outbound packet, so a database connection or a health check from your own code can keep it awake. Render’s free web services spin down after 15 minutes without inbound traffic and take about a minute to come back; paid instances run, and bill, continuously. Vercel and Netlify functions bill only while they execute, which is why a quiet site costs almost nothing there.

Canner doesn’t meter compute at all. A Starter project sleeps after 3 hours without a request and wakes on the next one, browser visit, webhook or API call alike; the first visitor waits a few seconds. On Live, two projects stay on permanently, and each extra always-on project is $4 a month. The deploy docs describe the wake-up in detail.

A flat price is its own cap. On metered platforms, the protection is a setting you have to turn on, and it works by taking your app offline.

The Bank of Canada’s rate on 2026-10-29 was 1.3985 Canadian dollars per US dollar. At that rate Vercel Pro’s US$20 is about CA$27.97 a month per developer seat, before tax, and Render’s US$25 Pro workspace is about CA$34.96 before compute. The same invoice costs more in a month the loonie drops, and your card issuer may add a foreign-transaction fee on top.

Canner prices, bills and invoices in Canadian dollars: Live is $9 a month or $99 a year. Canner issues its own numbered invoices with GST and QST itemised by province, which is what an accountant claiming input tax credits needs.

• Usage-based wins for a static site or a few functions with little traffic: Vercel Hobby or Netlify Free can cost nothing for months, as long as the project is non-commercial (Vercel) or fits in 300 credits (Netlify).
• Usage-based wins when traffic is spiky and you can tolerate a pause at the limit, or you’d rather pay for capacity only while it runs.
• Flat wins when traffic is steady or growing, when assets are heavy, or when a surprise invoice is worse than a slightly higher floor.
• Flat CAD wins when you budget, invoice or report in Canadian dollars. See plans and limits.
• Canner is the wrong pick if you need many separate services per app, a worker process with its own memory, or a CDN edge in several countries: Canner runs one process per project in one region, Montréal. Compare the alternatives on the comparison hub.