Hosting pricing, compared

CloudflareWorkers Paid

Bandwidth (egress) is free on every plan, so bandwidth_gb costs nothing, and static hosting (Pages, or static assets on Workers) is free; Pages Functions are billed as Workers. Requests are Worker invocations — if most of your requests are static assets, the real bill is lower. Workers bill CPU milliseconds rather than memory, so fn_gb_hours is converted as if one GB-hour were one hour of CPU time (3.6M CPU-ms): $0.02 per million CPU-ms = $0.072 per GB-hour, and the 30M included CPU-ms = 8.33 GB-hours. That is an upper bound — I/O-bound code uses far less CPU time than its wall-clock duration. Free's request cap is 100,000 a day ≈ 3M a month, and its 10 ms CPU limit per invocation caps compute at about 8.33 GB-hours at that volume. KV, D1, Durable Objects, R2 and Workers Logs are priced separately and left out. Account members cost nothing.

Hetzner CloudCPX12

Prices are in EUR, shown with 0% VAT; converted at the ECB reference rate of 2026-10-02 ($1.1225 per €1): CPX12 €11.99 → $13.46/month, extra traffic €1.00/TB → about $0.0011/GB. The included 20 TB is per server; the model counts it once, which only matters above 20 TB. CPX12 is listed for EU (Falkenstein, Nuremberg, Helsinki) and Singapore only. When CX23 is in stock it would be the cheapest qualifying server at €5.99 (≈ $6.72). Whether the listed price includes a primary IPv4 address wasn't stated on the page and isn't modelled. No free tier, no seat fees.

Fly.ioPay As You Go

Instance = shared-cpu-1x with 2 GB RAM, always on for 30 days in iad: $0.000000847/vCPU-second plus 1.75 GB of extra RAM at $0.000002316/GB-second ≈ $12.70 (from Fly's per-region price table formula; not a row in the summary table). A shared vCPU is not a dedicated core: the dedicated performance-1x with 2 GB is $33.00/month. Egress is the North America/Europe rate. Volumes ($0.15/GB-month), snapshots beyond 10 GB, dedicated IPv4 ($2/month), Managed Postgres and support plans are left out. Reservation blocks give 40% off compute when prepaid annually; not modelled. Organizations on plans discontinued in October 2024 keep their old allowances.

Amazon Web ServicesPay as you go (Lambda + EC2 t4g.small)

An approximation of one common setup, not AWS's whole price list: Lambda for serverless requests and compute, EC2 for always-on servers, and data transfer out to the internet for bandwidth, all in us-east-1. Lambda x86 duration is $0.0000166667 per GB-second = $0.06 per GB-hour, and the monthly free 400,000 GB-seconds = 111.11 GB-hours (Arm is $0.048 per GB-hour); this holds at any memory size from 128 MB to 10 GB since you pay per GB-second. One instance is an on-demand t4g.small (2 vCPU burstable Arm, 2 GiB) at $0.0168/hour × 730 = $12.26 plus a public IPv4 address at $0.005/hour = $3.65, so $15.91 a month (from the EC2 on-demand and VPC pricing pages). Data transfer out is $0.09/GB up to 10 TB a month, then $0.085, $0.07 and $0.05 in higher tiers. Left out: EBS disk for the instance, API Gateway or a load balancer in front of Lambda/EC2 (Lambda function URLs are free), CloudFront, and the up-to-$200 Free Tier credits for new accounts. There are no per-seat fees.

NetlifyPro 3,000 credits

Netlify bills credits, not dollars per unit: web requests 2 credits per 10,000 (200 per million), bandwidth 20 per GB, functions compute 10 per GB-hour. Extra credits by auto recharge cost $5 per 500 on Personal ($0.01 each) and $10 per 1,500 on Pro ($0.00667 each), which gives the rates. A plan's credits are a single pool, but this model can only give each driver its own allowance, so each plan's credits are split 25% requests / 50% bandwidth / 25% compute — the mix at the calculator's default usage (5M requests, 100 GB, 100 GB-hours = 1,000 + 2,000 + 1,000 credits). Away from that mix the real bill can be somewhat lower, and Free's hard 300-credit limit is reached sooner or later than the per-driver caps suggest. Production deploys (15 credits each), AI inference and database usage are left out. Pro has unlimited free members; Free and Personal allow one member. Accounts created before 4 Sep 2025 may stay on legacy plans, which are not modelled.

RenderHobby

Instance = the 1c-2g compute size (1 CPU, 2 GB RAM), $25/month; the free 512 MB instance has limitations and can't serve as a 1 CPU / 2 GB server. Persistent disks ($0.25/GB-month), Postgres, Key Value, custom domains beyond the included count ($0.25 each) and extra build minutes are left out. Scale ($499/month, 1 TB bandwidth included) is omitted as a governance/compliance plan; Enterprise is contact-sales. A Pro workspace with no services and no activity in a month has its fee waived.

VercelPro + Flat Rate CDN 10M

Rates are the Washington D.C. (iad1) / US rates; other regions cost more. Requests are Vercel's CDN Requests and bandwidth is Fast Data Transfer. Fluid compute bills Active CPU ($0.128 per CPU-hour) plus Provisioned Memory ($0.0106 per GB-hour, billed for the instance's whole lifetime including I/O waits); fn_gb_hours is Provisioned Memory GB-hours on the default 2 GB / 1 vCPU function, with Active CPU assumed to be 25% of instance time — 0.125 CPU-hours per GB-hour, so $0.0106 + $0.016 = $0.0266 per GB-hour. CPU-heavy functions cost up to $0.0746 per GB-hour; I/O-bound ones closer to $0.0106. Under the same assumption Hobby's 4 CPU-hours (not its 360 GB-hours of memory) is the limit, at 32 GB-hours. Pro's $20 monthly credit is modelled as 752 GB-hours of function compute; in reality it covers any usage from the first unit. Left out: function invocations ($0.60 per million beyond Hobby's 1M, only for requests that run a function), Fast Origin Transfer ($0.06/GB), image optimization, ISR, builds, and add-ons. Flat Rate CDN tiers are $20, $100 and $300 a month on top of Pro; the included tier needs no extra fee.

RailwayHobby

One always-on 1 vCPU / 2 GB server at full utilisation = $20 + 2 × $10 = $40 per 30 days ($0.00000772/vCPU-second, $0.00000386/GB-second); an idle or lightly loaded service costs less, so this is an upper bound. The plan credit really applies to all usage (CPU, RAM, egress, volumes); it is modelled as covering part of a server, which is exact whenever you run at least that much compute. Volumes ($0.15/GB-month) and object storage are left out. The Free plan ($1/month credit, max 0.5 GB RAM per service) can't run a 2 GB server and is omitted; Enterprise is contact-sales. Railway may waive the Hobby fee for some active builders (automated, not by request).