Buyer constraint
Development against production-HA cost
What resilience actually adds to the bill, per provider.
The rule, stated before the list
A comparable record must price both a single development instance and a production floor of two instances across failure domains plus a load balancer.
A plan reaches the table below only if all of this is true of it:
- The vendor publishes a monthly price for it.
- It is a paid plan, not a free tier.
- Its vCPU and its memory are both published.
- The vendor publishes a load-balancer price.
- The vendor publishes a cross-zone transfer rate.
- The vendor publishes a health-monitoring charge.
- The vendor states, on a page we can cite, that the plan runs across more than one failure domain.
Caveats and plans that just miss
These came close enough to be worth knowing about. Some meet the broad rule above but fail a narrower use-case check; the failed column names that distinction. Publishing those exclusions makes it possible to tell a deliberate caveat from a plan nobody evaluated.
| Provider | Plan | Failed | Price |
|---|---|---|---|
| HostBrr | EPYC-4GBrr | did not meet the rule | $4.25 |
| Hetzner | CX23 | did not meet the rule | $6.49 |
| DigitalOcean | s-2vcpu-4gb | did not meet the rule | $24.00 |
| Render | Standard | did not meet the rule | $25.00 |
What actually decides this
The dataset cannot honestly render a production-HA total. It can price one development instance and it can double that compute number, but no paid comparison record supplies both a load-balancer price and sourced failure-domain support. Cross-zone transfer and health-monitoring cost are also absent. The correct production total is therefore null for every row, with a visible two-instance compute floor where useful.
Development is allowed to optimize for speed and reversibility. The lowest current single-instance row can be a reasonable disposable environment when downtime and rebuilds are acceptable. It is not evidence for a production recommendation. A single disk, host, network path, and maintenance event remain single points of failure.
Production HA starts with an availability objective and failure model. AWS makes the distinction explicit: its single-instance EC2 SLA is 99.5 percent, while the 99.99 percent region-level commitment requires instances across at least two availability zones. The architecture usually adds a second instance, health-based routing or load balancing, replicated state, backups, monitoring, and possibly cross-zone traffic. Managed platforms may internalize some of those pieces, but the dataset must source that before setting their cost to zero.
Use the development price as one line, the two-instance compute floor as a second line, and the unresolved HA inputs as a third. Do not call the second line production cost. For each candidate, verify zone placement, health checks, failover timing, state replication, maintenance behavior, load-balancer pricing, and transfer between failure domains. When those inputs arrive, the formula can execute. Until then, this page should teach the cost boundary and return no production winner. That is more useful than a precise total built from missing infrastructure.
How to check this yourself
Every price here is read from the vendor's own page and carries the date we read it, most recently unknown. The rule above lists every check applied, so you can disagree with the threshold rather than guess at it. The full method is here, and the dataset is downloadable if you would rather run your own.
Compare every provider on your own workload ·All VPS plans ·Report a wrong price