Dora NodaRender changed its prices twice in one week — price-point changes logged on August 22 and again on August 28, 2026. Neither change, on its own, was dramatic. Together they make a point that no pricing page will ever print at the top: the headline starting price of a platform is a moving target, and the number that decides whether you should stay or migrate is never on the landing page. It is the full monthly bill for your actual workload, recomputed line by line.
This post does that recomputation. A typical always-on web service — one app server plus a managed Postgres — priced line by line on Render's current published meters at three traffic levels, set against the same workload on a flat-rate Hetzner node. Then the wider entry-plan market, from roughly $5 to $29, mapped by billing model so you can see which kind of meter you are signing up for.
The cost sheet: one app, one database, three traffic levels
The workload is deliberately boring, because boring is what most migrations are about: a single always-on web service and a small managed Postgres, running 24/7. No autoscaling fleet, no multi-region mesh, no edge functions. Just the thing a side project becomes when it starts earning money.
On Render's current published pricing, that workload prices as follows on a Hobby workspace ($0 workspace fee):
- Web service, Starter tier: $7/month. This is the cheapest always-on tier — 512 MB of RAM, and the free tier is not an alternative for production because free instances sleep after 15 minutes of inactivity.
- Managed Postgres, Basic-256mb: $6/month for the compute, plus storage at $0.30/GB/month. A small database on 5 GB of storage adds $1.50, for $7.50 total. Note the floor here: free Postgres expires 30 days after creation, with a 14-day grace period before deletion, so production pays for the database from day one.
- Bandwidth: 5 GB included on Hobby, then $0.15/GB. Render cut this overage from $30 to $15 per 100 GB on August 1 — a genuine price cut that still leaves bandwidth as the bill's swing variable, as the table shows.
| Monthly line item | 5 GB traffic | 25 GB traffic | 100 GB traffic |
|---|---|---|---|
| Web service (Starter) | $7.00 | $7.00 | $7.00 |
| Postgres ($6 + 5 GB storage) | $7.50 | $7.50 | $7.50 |
| Bandwidth overage ($0.15/GB past 5 GB) | $0.00 | $3.00 | $14.25 |
| Render total | $14.50 | $17.50 | $28.75 |
| Hetzner CX22 (2 vCPU, 4 GB, 20 TB included) | ~$4.10 | ~$4.10 | ~$4.10 |
| Delta | ~$10.40 | ~$13.40 | ~$24.65 |
The Hetzner side is one line because that is the entire point of flat pricing: a CX22 at €3.79/month (about $4.10) includes 20 TB of traffic, so all three columns cost the same number. Postgres runs on the same box or a second small one — a cost in operational attention, not in a second meter, which Section 5 prices honestly.
Two things to read off this table before moving on. First, at low traffic the workload costs about $14.50 on Render — matching Render's own July 2026 guidance that an always-on Starter service plus Basic Postgres runs about $13/month before bandwidth and storage growth. The sheet reconciles with the vendor's own math, which is how you know the meters were read correctly. Second, the bandwidth row is the only row that moves, and it roughly doubles the bill between the first and third column. The compute did not change. The database did not change. Traffic did — and traffic is the thing a growing app is supposed to have more of.
The 29 entry market: three billing models wearing similar price tags
A daily-verified comparison tracking 18 platforms puts paid entry plans between roughly $4.99 and $29. That spread looks like a choice between cheap and expensive. It is really a choice between three different billing models that happen to overlap at the bottom:
- Pure usage (Railway, Fly.io). Railway's Hobby plan has a ~$5 minimum against metered usage; Fly.io bills pay-as-you-go from roughly $2 for a small machine. There is no per-service plan to pick — the meter runs on what you consume. This is the fairest model for small, bursty workloads and the most surprising one when traffic grows, because every dimension (compute, egress, volumes) scales independently.
- Tiered per-service plans (Render compute, Heroku dynos). Heroku's Eco dyno starts at $5, Render's Starter service at $7. Each service gets a named tier with fixed RAM and CPU, and each service, database, and worker is billed separately. Predictable per unit, multiplicative across units — the second service doubles the compute line.
- Hybrid: workspace fee plus usage (Render since April 2026). On top of the per-service tiers, Render added a flat workspace subscription on April 23, 2026 — Hobby free, Pro $25/month flat, Scale $499/month flat — replacing per-seat pricing ($19/user Professional, $29/user Organization). Legacy workspaces were force-migrated on August 1. The headline compute prices did not change that day, but every team's bill gained a potential $25 or $499 line that has nothing to do with how much compute they run.
The $29 end of the range is usually yesterday's per-seat math or a workspace fee, not compute. That is exactly why comparing starting prices across models misleads: a $7 Starter plan and a $5 usage minimum are not two prices for the same thing. One is a tier, the other is a floor under a meter.
Where the bill actually moves: four multipliers
With the sheet built, the sensitivity analysis writes itself. Four mechanisms move the total; none of them is the headline compute price:
1. Per-service multiplication. The sheet prices one web service. Add a background worker ($7 Starter), a staging copy of the stack ($7 + $7.50), or a Redis-compatible Key Value instance (from ~$10/month), and each addition lands at full tier price. Two environments of the same boring workload are ~$29 before bandwidth — the top of the entry market, reached without a single user arriving.
2. Bandwidth overage. At $0.15/GB past the included 5 GB (Hobby) or 25 GB (Pro), every additional 100 GB of traffic is $15. An app serving images or video that grows from 25 GB to 500 GB/month adds roughly $70 in overage while its compute line stays flat. The August 1 cut from $30 to $15 per 100 GB halved this slope without changing its shape.
3. The workspace fee. Move to Pro for previews, autoscaling, or more than two custom domains, and $25/month appears before any workload is priced. On the sheet's 100 GB column, Pro math reads $25 (workspace) + $7 + $7.50 + $11.25 (75 GB over the 25 GB included) — about $50.75, versus $28.75 on Hobby. The fee also buys 20 GB more included bandwidth, which offsets $3 of that gap and hides the rest.
4. Metered extras. Additional custom domains past the 15 included on Pro bill at $0.25/domain/month. Small per unit, and symptomatic: platforms that have finished competing on compute grow the bill by metering the edges — domains, snapshots, build minutes. Each new line item is individually reasonable and collectively a bill whose shape changes between quarters, which is precisely what happened twice in the last week of August.
What flat hardware changes — and what it honestly costs
The Hetzner column of the table is not an argument that hardware is free. It is an argument that flat pricing moves cost from a variable bill to a fixed one plus operational labor, and that trade has a shape worth stating plainly:
- What you gain: the monthly number stops depending on traffic. 5 GB or 500 GB, one service or four, staging or not — the CX22 costs €3.79 whether August is quiet or your launch lands on Hacker News. Budgeting becomes arithmetic instead of forecasting.
- What you pay instead: you operate Postgres — backups, point-in-time recovery, version upgrades, disk growth — instead of renting those as managed features. Render's $6 Postgres compute plus $0.30/GB storage is, itemized honestly, mostly an operations salary fractionalized across tenants. A team with no one willing to own a database runbook should keep renting that specific line item (Neon, Supabase, RDS) even after migrating compute.
- Where flat wins outright: bandwidth-heavy, always-on, multi-service workloads — the exact shape where every Render multiplier points upward at once. Where metered still wins: genuinely bursty or idle-mostly workloads, where per-second billing for a mostly-sleeping service undercuts a 24/7 box.
The migration question is therefore not "is $7 more than $4.10" — it is "which of the four multipliers will my workload trip over the next twelve months, and what is twelve months of that meter versus twelve months of a flat box plus the ops time." Recompute the sheet with your traffic, your service count, and your workspace tier. If the answer stays near the first column, stay. If it drifts toward the third, the twice-in-a-week repricing is telling you something about the direction of the slope, not just its current value.
Render's meters are current as of late August 2026 and move often — verify against the invoice before budgeting. Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.
