Oracle Cloud Out of host capacity: Why Ampere A1 Won't Launch and the Truth About PAYG

When you create an Ampere A1 instance on Oracle Cloud, click Create and get Out of host capacity, you have hit the step where Always Free users get stuck most often. The error means exactly what it says: there is currently no host capacity available to provision the shape you chose in the specified region, availability domain (AD) or fault domain (FD). Your account has not been banned, your credit card is fine, and it has nothing to do with your SSH key or image.

The short version: follow Oracle’s official advice in order, leave the FD unspecified, switch AD, use a smaller shape, and if that fails, wait a while and try again. To check programmatically, you can look at capacity with a capacity report first. Upgrading to Pay As You Go (PAYG) does not add physical capacity and does not guarantee a successful launch, and usage beyond the free allowance is actually charged.

This post brings together Oracle’s official documentation (checked on 12 October 2026) and my notes from setting up a Claude Code server on Oracle in October 2026. Capacity, free allowances and prices can change at any time; Oracle’s official documentation, your account’s Console and your bill are what count. Nothing here is a guaranteed way to get an instance created.

What Out of host capacity means

Oracle’s troubleshooting documentation explains that capacity is dynamic. It depends on the size of the host pool for a shape in a given region, AD or FD, and on how much OCPU and memory is left on those hosts. The Ampere A1 shape is VM.Standard.A1.Flex, and the same limit applies to it.

So the error means there is no room “right now, in this location, at this size”. Change the options or the timing and the result may be different. Signing up for a new account, switching credit cards or regenerating your SSH key will not change it.

Troubleshooting order for Out of host capacity

Leave the Fault Domain unspecified and let Oracle choose

When creating the instance, do not pick a Fault Domain by hand. According to the documentation, if you leave it unspecified, Oracle places the instance in the best available FD. If you picked a specific FD earlier, switch it back to automatic first.

Switch to another Availability Domain

Some regions have several ADs, so you can try another AD. Not every region has more than one AD, and Always Free compute resources can only be created in your account’s Home Region (explained below), so you cannot hop between regions looking for free space.

Use a smaller shape

A1 is a flexible shape, so you set the OCPU and memory yourself. Start with something smaller, for example 1 OCPU and 6 GB, and review what you need once the instance exists. After shrinking it, double-check that the size is enough for your use.

Wait a while and try again

Capacity changes, so trying again later is reasonable. The point is not to send requests over and over at high frequency with no limit; a later section covers the rules for automatic retries.

Check capacity first with a capacity report

If you need to check programmatically, the OCI CLI has oci compute compute-capacity-report create, which tells you before you create an instance whether an AD has enough capacity for a shape. The required parameters are the AD, the compartment (the documentation says to use the root compartment) and the shape configuration:

# First generate a JSON template for the shape configuration, then fill in VM.Standard.A1.Flex with the OCPU and memory
oci compute compute-capacity-report create \
  --generate-param-json-input shape-availabilities > shape-availabilities.json

oci compute compute-capacity-report create \
  --availability-domain "<YOUR_AD_NAME>" \
  --compartment-id "<YOUR_ROOT_COMPARTMENT_OCID>" \
  --shape-availabilities file://shape-availabilities.json

The report only reflects the state at the moment you query it. Capacity shown in the report does not guarantee that the launch right after will succeed.

Running out of capacity is not the same as running out of quota

Not every failed launch is a capacity problem. Read the error message first, then decide what to do:

Error or situation	Roughly means	Check first
`Out of host capacity`	No physical host capacity in the chosen location for now	FD, AD, shape size, try again later
`LimitExceeded`	Over the account’s service limit or quota	Limits, Quotas and Usage in the Console, and how much A1 the account already uses
`NotAuthorizedOrNotFound`	Insufficient permissions, or the specified resource cannot be found	Compartment, IAM permissions, whether the OCID is correct

One more check that is easy to miss: how much of the allowance your other A1 instances already use. Oracle’s Free Tier documentation says all A1 instances on an Always Free account together should not exceed 2 OCPU and 12 GB. If they do, all A1 instances are stopped and then deleted after 30 days unless you upgrade to a paid account.

Does upgrading to PAYG fix Out of host capacity

Many tutorials treat “upgrade to PAYG” as the secret to getting an A1. In reality you have to look at it piece by piece:

What PAYG is: Oracle’s paid account type, with no minimum usage commitment. Existing resources keep working after the upgrade, and resources that meet the Always Free conditions stay free.
What PAYG may change: the kinds of resources and the limits the account can request may increase.
What PAYG does not change: physical A1 host capacity. Oracle does not say that upgraded accounts get priority when creating A1 instances. PAYG accounts can hit Out of host capacity too.
What PAYG costs: usage beyond the Always Free allowance is actually charged, for example a bigger shape, extra disks, backups or other services.

Before and after upgrading, set up a low monthly Budget with email alerts in Billing & Cost Management. But be aware that Oracle’s documentation says a budget is a soft limit: it only sends notifications, alerts are evaluated every 24 hours, and charges are not stopped automatically. To keep usage under control, you still need to check your resources and Cost Analysis yourself.

The A1 free allowance: two figures in the official documentation

When checked on 12 October 2026, two official Oracle documents gave different A1 free allowances:

Official document	Free allowance per month	Equivalent running all month
Always Free Resources	1,500 OCPU hours, 9,000 GB hours	About 2 OCPU, 12 GB
Arm-based Compute	3,000 OCPU hours, 18,000 GB hours	About 4 OCPU, 24 GB

This post uses the conservative 2 OCPU and 12 GB. Do not simply assume 4 OCPU and 24 GB are free; go by the Always Free labels in your account’s Console, your usage for the month and your bill. Storage is counted separately: Always Free block volume storage is 200 GB in total (boot volumes plus block volumes), and the same page gives the minimum boot volume as both 47 GB and 50 GB.

If you must retry automatically: set a limit and back off

If you retry with a script, the rules are:

Cap the number of attempts: stop when you reach the cap and look at the situation yourself.
Lengthen the interval step by step, with some randomness: for example, start at a few minutes and increase the wait after each failure. Do not hit the API every few seconds.
Tell errors apart: retry only on Out of host capacity. Retrying errors such as LimitExceeded or NotAuthorizedOrNotFound will not help; stop and fix the configuration.
Follow Oracle’s usage policies and API rate limits.

I am not providing a retry script here: no script can guarantee a successful launch, and I have not tested a publishable version on my account. If you write your own, follow the four rules above.

Once the instance exists, check the shape, OCPU, memory, boot volume size and Always Free labels in the Console. Don’t assume everything is free just because the launch succeeded.

Home Region and idle reclamation

Home Region: Always Free compute resources can only be created in your account’s Home Region, and volumes outside the Home Region are charged at the normal rate. Before choosing a Home Region at sign-up, think about location and latency, and don’t treat it as something you can change at any time later. Also, don’t sign up for more accounts with false information just because a region is short on capacity for a while.
Idle reclamation: the official documentation states that idle Always Free instances may be reclaimed. The test is whether all of the following hold over 7 days: the 95th percentile of CPU utilization is below 20%, network utilization is below 20%, and for A1, memory utilization is also below 20%. Neither upgrading to PAYG nor keeping a program running with systemd guarantees an instance won’t be reclaimed, and don’t generate fake load on purpose to get around the test.

Once the instance is up, the next steps (SSH, installing Claude Code and the always-on setup) are in Running Claude Code Remote Control on a Free Oracle Cloud Server. If you hit No medium found while setting up systemd, see How to Fix systemctl –user No medium found. For the role this server plays in the site’s overall architecture, see Jason Finance from Zero to One.

FAQ

What does Out of host capacity mean on Oracle Cloud?

It means there is currently not enough host capacity to provision the shape you chose in the specified region, availability domain or fault domain. Oracle's documentation says capacity is dynamic and changes with the state of the host pool. It does not mean your account broke the rules, your credit card failed or your SSH setup is wrong.

What should I do about Out of host capacity?

Try these in order: leave the fault domain unspecified so Oracle picks one; switch to another availability domain; use a smaller shape, for example 1 OCPU and 6 GB for A1 to start; wait a while and try again. If you need to check programmatically, you can use an OCI capacity report first, but available capacity in the report does not guarantee the launch will succeed.

Does upgrading to Pay As You Go guarantee I can create an Ampere A1 instance?

No. PAYG may let the account request more kinds of resources, but it does not add physical A1 host capacity, and Oracle does not say upgraded accounts get priority. After upgrading, usage beyond the Always Free allowance is actually charged.

Can an Oracle Budget act as a spending cap?

No. Oracle's documentation says a budget is a soft limit used to send notifications when you may go over budget. Alerts are evaluated every 24 hours, and charges are not stopped automatically.

Is the Ampere A1 free allowance 4 OCPU and 24 GB, or 2 OCPU and 12 GB?

When checked on 12 October 2026, two official Oracle documents gave different figures: the Always Free resources page says 1,500 OCPU hours and 9,000 GB hours per month, roughly 2 OCPU and 12 GB; the Arm compute page says 3,000 and 18,000. This post uses the conservative 2 OCPU and 12 GB; your account's Always Free labels and your bill are what count.

Does Oracle reclaim idle Always Free instances?

It may. The documented conditions are that over 7 days the 95th percentile of CPU utilization is below 20% and network utilization is below 20%, and for A1, memory utilization is also below 20%. When all of these hold, the instance may be reclaimed. Do not generate fake load on purpose to get around this.