Dora NodaIn the spring of 2026, both Railway and Render repriced the same thing in the same window: not compute, but reliability. Railway shipped one-click high-availability Postgres to its Priority Boarding testers in March, an experimental Patroni-based cluster it explicitly warned was not production-ready yet. Render replaced its legacy per-seat workspace plans with flat-rate pricing on April 23, moving the features that make a deployment production-grade — autoscaling, preview environments, audit logs — onto clearly labeled plan tiers. And all of this landed weeks after Heroku, the platform thousands of teams were re-evaluating, confirmed on February 6 that it was entering sustaining-engineering mode with no new features and no new Enterprise sales.
The question every team doing that re-evaluation is actually asking is not "what does hosting cost." It is "what does not going down cost." So here is the number first, worked from published rates for a typical small-production setup: two app replicas at 1 vCPU and 2 GB each, Postgres with a standby, 10 GB of volume storage, 100 GB of monthly egress, and a team of five.
The number first: what HA actually costs
| Line item | Railway (Pro + metered) | Render (Pro + fixed) | Own fleet (3× Hetzner CX33) |
|---|---|---|---|
| App compute, 2 replicas | ~$80/mo (2 × ($20/vCPU + 2 × $10/GB)) | $50/mo (2 × $25 Standard) | included below |
| Postgres with standby | ~$100/mo (estimated HA stack) | ~$27/mo (Standard + HA standby) | self-managed |
| Plan fee | $20/mo (Pro, credited toward usage) | $25/mo flat (5 members incl.) | $0 |
| Egress, 100 GB | $5/mo ($0.05/GB) | ~$11/mo (25 GB incl., then $0.15/GB) | $0 (20 TB incl.) |
| Volume, 10 GB | ~$2/mo ($0.15/GB) | ~$3/mo ($0.25/GB) | included (local NVMe) |
| Machines | — | — | ~$35/mo (3 × €8.49 + IPv4 + backups) |
| Total | ~$187/mo | ~$116/mo | ~$35/mo + your ops labor |
Those are three answers to the same redundancy question, and they differ by more than 5x from top to bottom. The rest of this post earns that table: what each vendor changed, why the reliability tier rather than the entry price decides migrations, and where the math flips.
What Railway changed: HA arrives, metered and experimental
Railway's headline reliability shipment this year is high-availability Postgres, announced March 13, 2026. One click turns a single Postgres instance into a cluster: replicas managed by Patroni, an HAProxy proxy routing to the current primary, and an etcd cluster for leader election, with a dedicated UI for cluster health and replica status. Automatic failover, no manual intervention — the full stack teams previously had to assemble from Railway's Postgres HA template themselves.
Two caveats matter for the price tag. First, it shipped to Priority Boarding behind a feature flag as explicitly experimental and not production-ready, with Railway telling users not to upgrade a production database to it yet. There is no GA list price for the HA cluster as a unit; in the table above I estimated it as roughly a primary plus a replica plus the proxy and consensus overhead (~$100/mo at Railway's metered rates of $20/vCPU-month and $10/GB-month). Estimates are honest here because Railway's model makes them computable: every replica is just more metered seconds.
Second, Railway's application-layer HA story is replicas and regions, and the gating is plan-based: multi-region replicas require Pro, and the platform supports up to 42 replicas per service on Pro (50 on Enterprise). On metered pricing there is no per-replica fee — the second replica simply doubles the compute meter. That is the cleanest mental model of the three platforms: reliability costs exactly the resources it consumes, with the $20/mo Pro subscription doubling as a usage credit. The sharp edge is that "exactly the resources" cuts both ways, as the sensitivity section below shows.
What Render changed: seats go flat, reliability gets labeled
Render's April 23, 2026 change is the more structural of the two: legacy per-seat plans (Professional at $19/member/mo, Organization at $29/member/mo) were replaced with flat-rate Hobby ($0), Pro ($25/mo), Scale ($499/mo), and Enterprise (custom), with paid plans including unlimited members. Legacy workspaces were given until August 1, 2026 to migrate. The per-service compute ladder stayed familiar — Starter at $7/mo (0.5 CPU/512 MB), Standard at $25/mo (1 CPU/2 GB), Pro at $85/mo — with each additional instance billed at full price.
What changed for the reliability buyer is legibility. Render's platform features by plan now reads as an explicit reliability menu: Pro unlocks autoscaling, multi-service preview environments, the performance build pipeline, metrics streaming, and audit logs, while Scale adds SAML SSO/SCIM, HIPAA-enabled workspaces, expanded inbound IP rules, and organization-level controls. A five-person team that paid $95/mo in Professional seats alone now pays a $25 flat fee — but a team that needs SSO or HIPAA now stares at a $499/mo line item that used to be a seat multiple. Production-grade is no longer a vibe; it is a row in a table.
On the database side, Render Postgres offers a high-availability standby on paid tiers, with a small instance plus standby landing in roughly the high-$20s per month. Combined with fixed per-instance compute, Render's HA bill is the most predictable of the three: count instances, count the plan fee, add egress over the included 25 GB at $0.15/GB. The table's ~$116/mo has almost no estimation in it, which is itself a feature worth pricing.
Why the reliability tier decides migrations
Heroku's February 2026 announcement is the forcing function behind this entire comparison. When Salesforce confirmed Heroku was moving to sustaining engineering — security patches and stability work, but no new features and no new Enterprise contracts — it did not shut anything down. It did something arguably more decisive for planning: it told every team that their platform has no future capability to grow into. The migration guides Render now publishes for Heroku (and Railway) exist because that decision pushed thousands of teams into exactly this spreadsheet.
And the spreadsheet trap is comparing entry prices. A single Starter web service at $7/mo versus a metered Railway bill in the low single digits tells you nothing about what your architecture costs once it has two replicas, a standby database, preview environments for safe integration testing, and audit logs for the security review. The entry price answers "can I try this." The reliability tier answers "can I run my business on this." Those are different questions with different winners: Render's flat $25 Pro fee is nearly invisible at a team of five and a rounding error at a team of twenty, while the same team's legacy Professional seats would have billed $95 and $380 respectively. Railway's metered model is cheapest at idle and most honest under load, but every replica is a full additional meter with no volume discount on the reliability itself.
This is also why both vendors' 2026 moves point the same direction even though they look different. Railway is productizing failover (HA Postgres, replicas up to 42 per service on Pro, multi-region on Pro) while keeping the meter running on all of it. Render is packaging operational maturity into plan tiers with flat fees.
Both raised $100M rounds in early 2026 — Render's Series C extension at a $1.5B valuation alongside its Render Workflows early access, and Railway's $100M Series B in January as reported by VentureBeat — and both are spending it on the primitives that turn "deploys easily" into "stays up." The market has decided the interesting price is not compute. It is confidence.
The same redundancy on machines you own
The third column of the table is three Hetzner Cloud CX33 machines (4 vCPU/8 GB each) at €8.49/mo apiece under the June 15, 2026 price adjustment, plus €0.50/mo per IPv4 address and ~20% for backups: roughly $35/mo all-in, with 20 TB of included traffic making the egress row zero. On owned machines, the second replica is free — the hardware is already paid for — and so is the third, until the nodes fill. A Cluster-API-managed fleet declares replicas the same way on three machines as on thirty; the marginal cost of redundancy is zero until you buy another box.
The honest part of this column is the last row: your ops labor. The $35/mo buys machines, not outcomes. Patroni failover, backup restore testing, Postgres upgrades, node draining, on-call rotation — Railway and Render sell those as the product, and on your own fleet they are your evenings and weekends unless you automate them like a platform team. SNCF's March 2026 CNCF case study on building on-premises Kubernetes with Cluster API and Argo CD proves declarative fleet operations work at national-infrastructure scale — and proves it takes deliberate automation investment to get there.
Price the machines at $35/mo, then price the automation honestly against the ~$80–150/mo gap to managed. For many small teams the managed premium is the cheapest on-call engineer they will ever hire.
Where the math flips
No single worked example settles a migration, so here is the sensitivity analysis across the three variables that actually move the answer.
Replica count up (2 → 6 replicas). Railway adds ~$160/mo in pure compute meter; Render adds $100/mo in Standard instances; the Hetzner fleet adds $0 until the nodes saturate. Metered and fixed pricing both tax redundancy linearly, while owned hardware absorbs it. Teams running genuinely multi-replica, multi-region topologies are where self-hosting's flat cost curve pulls away hardest.
Egress up (100 GB → 1 TB). Railway goes from $5 to $50/mo at $0.05/GB; Render goes from ~$11 to ~$146/mo past its included 25 GB at $0.15/GB; Hetzner stays at $0 inside its 20 TB allowance. Bandwidth-heavy workloads — media, downloads, API-heavy SaaS — flip the ranking fast, and Render's $0.15/GB is the steepest of the managed curves. Check your actual egress before you check anything else.
Team size up (5 → 20 members). Render's flat $25 Pro fee does not move; under legacy per-seat pricing the same team paid $380/mo for Organization seats. This is the quietest migration incentive in the whole comparison: Render's restructure made reliability features cheaper for exactly the growing teams most likely to need them. Against that, a side project with one replica and no standby flips the other way — a $7–25/mo managed bill beats any fleet's cost floor once you value your own time above zero.
The throughline: managed platforms tax redundancy per unit, owned fleets tax it per machine, and your workload's shape — replica count, egress profile, team size — picks the winner more than any vendor's headline price.
Price the tier you will actually buy
Spring 2026 repriced reliability in public: Railway's experimental HA Postgres and replica-based app redundancy on metered rates, Render's flat-rate plans with an explicit reliability menu, and Heroku's freeze pushing everyone to read the menus. Work the table for your own replica count, your own egress, your own team size — and when the managed reliability tier costs more than the redundancy is worth to you, that is the signal to own the machines.
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.
