How to stop a surprise cloud bill: spending caps and budgets on Vercel, Supabase and Azure
By Menelaos Vergis · · Checked against each provider's documentation on that date.
Two of these three providers can stop charging you when you reach a limit. On Vercel's Pro plan you set a budget and switch on Pause production deployments. On Supabase's Pro plan you leave the Spend Cap on. Azure is different: on a pay-as-you-go subscription a budget only sends you an email, so stopping anything means wiring up automation yourself.
None of the three caps covers everything, and none reacts the second you cross the line. The sections below show where each setting lives, what it actually stops, and what still gets billed.
Key takeaways
- Vercel: a budget with Pause production deployments on takes your production site offline once spend reaches the amount, a few minutes late. Seats, Marketplace integrations, add-ons and storage keep billing.
- Supabase: the Pro plan's Spend Cap blocks over-usage of metered items, but compute is not covered, so extra projects and bigger instances still cost money.
- Azure: budgets never stop resources. Only free accounts and subscriptions with monthly credits get a real spending limit.
- On every provider, set the limit below what you can afford to lose, and put separate limits on your AI API keys.
Which providers can actually stop spending?
Here is the short answer, provider by provider, with the delay each one admits to in its own documentation.
| Can it stop charges? | Where | Not covered | Delay | |
|---|---|---|---|---|
| Vercel (Pro) | Yes, by pausing production | Settings, Billing and Spend Management | Seats, integrations, add-ons, storage | Several minutes |
| Supabase (Pro) | Yes, for metered usage | Organization billing page, Cost Control | Compute, IPv4, custom domain, PITR and more | Not stated |
| Azure (pay-as-you-go) | No, alerts only | Cost Management, Budgets | Everything, unless you automate | 8 to 24 hours for cost data |
The trade-off is the same everywhere: a cap that stops charges also stops your app. Decide before it happens whether you'd rather have a few hours offline or a bigger invoice. For a small app that's just launched, offline is almost always cheaper.
Vercel: set a budget and pause production
Vercel calls its cap Spend Management. It is available on the Pro and Enterprise plans at no extra cost, and you need the Owner or Billing role on the team to change it (Vercel: Spend Management).
- In your team's dashboard, open Settings, then Billing and Spend Management.
- Select Add Budget and choose Team (every project) or Project (just one).
- Set the amount in US dollars.
- Switch on Pause production deployments. Without it, the budget only notifies you.
- Optionally add a webhook URL, so your own code hears about it too.
Vercel turns on web and email notifications for every budget, at 50%, 75% and 100% of the amount, and you can add an SMS at 100%. New Pro customers get notifications at a spend amount of $200 per billing cycle by default (Vercel: Pro plan). That default notifies you; it doesn't pause anything unless you've switched pausing on.
What pausing does
When a paused budget is reached, your production websites, APIs and functions stop serving traffic, and visitors see a 503 DEPLOYMENT_PAUSED error. A team budget pauses every project on the team; a project budget pauses only that project. Raising the budget afterwards doesn't bring anything back: you resume each project yourself, from the Spend Management page or the REST API (Vercel: Spend Management).
What it doesn't stop
- The delay. Vercel checks usage every few minutes, so pausing can happen several minutes after you pass the amount. Vercel itself advises setting the budget below the most you are willing to spend.
- Fixed charges. Seats, Marketplace integrations and separate add-ons are billed monthly and sit outside the budget.
- Storage. Paused projects are not deleted, so storage charges such as deployment storage and Vercel Blob keep running.
- AI Gateway and v0. Pausing production doesn't stop AI Gateway API key usage or v0 usage billed to your team.
One more detail: every Pro plan includes $20 of monthly credit, and a team budget only counts usage beyond that credit, while project budgets count all of their usage.
On the free Hobby plan there's no Spend Management at all. If you go over a usage limit, you usually wait until 30 days have passed before you can use that feature again (Vercel: Hobby plan).
Supabase: keep the Spend Cap on
Supabase's Pro plan has a Spend Cap, and it is on by default (Supabase: Billing FAQ). With it on, once you use up the quota for a usage item, further use of that item is blocked until the next billing cycle. You aren't charged for the over-usage, but your services are restricted under Supabase's Fair Use Policy (Supabase: Cost control). Those restrictions can include pausing projects, switching databases to read-only, or answering API requests with a 402 status.
To check it, open your organization's billing page in the Supabase dashboard and find the Cost Control section. You can also set it when you create an organization on the Pro plan.
What the Spend Cap doesn't cover
This is where most surprise Supabase bills come from. Supabase leaves out items that are "predictable and explicitly opted into", and keeps charging for them with the cap on (Supabase: Cost control):
- Compute, Branching Compute and Read Replica Compute
- Custom Domain and IPv4 address
- Additionally provisioned Disk IOPS and Disk Throughput
- Log Drain Hours and Log Drain Events
- Multi-Factor Authentication Phone
- Point-in-Time Recovery
Compute matters most. Paid plans include $10 of compute credits a month, enough for one project on the Micro or Nano size (Supabase: Manage compute usage). Every extra project, preview branch or bigger instance is billed on top, cap or no cap. If your bill jumped, look there first.
The cap is also all-or-nothing: Supabase says it doesn't offer budgets for a single item or notifications when a certain cost is reached. On the Free plan you are never charged; you are notified when you go over the quota, and restrictions follow if you stay over.
Azure: budgets alert you, they don't stop you
Azure has two different features, and only one of them stops anything.
The spending limit stops spending at your credit amount. It exists on an Azure free account and on subscriptions with monthly credits, such as Visual Studio Professional and Enterprise. When you reach it, your services are disabled for the rest of the billing period: virtual machines are stopped and deallocated, and storage becomes read-only (Microsoft Learn: Azure spending limit). It isn't available on pay-as-you-go or commitment plans, and you can't set a custom amount.
A budget is what you get on pay-as-you-go. Microsoft is plain about it: "Resources aren't affected, and your consumption isn't stopped" (Microsoft Learn: Create and manage budgets).
Set one up anyway, since an early email is far better than a surprise invoice:
- In the Azure portal, open Subscriptions, select your subscription, and select Budgets.
- Select Add, give the budget a name, choose a reset period (monthly fits most apps) and set the amount.
- On the alerts step, add thresholds as a percentage of the budget. Use Forecasted as well as Actual, so you hear about it before the money is spent.
- Add the email addresses to notify, then create the budget.
Mind the delay: cost data typically arrives 8 to 24 hours after use, and budgets are checked against it every 24 hours. A runaway resource can run for a day before the email lands.
Two extras worth setting
- Anomaly alerts. In Cost Management, open Cost alerts, select Add and set the alert type to Anomaly. Azure compares each day with a forecast built from the last 60 days and emails you when it spots something unusual, about 36 hours after the day ends (Microsoft Learn: Identify anomalies and unexpected changes).
- An automatic stop. A budget can call an action group, and the action group can run an Azure Automation runbook that stops virtual machines at a threshold you choose (Microsoft Learn: Manage costs with budgets). Microsoft's walkthrough stops VMs; for App Service, Functions or databases you write the stop step yourself, and you inherit the same 24-hour delay.
The bills no cap catches
The settings above protect your hosting and your database. A surprise bill often comes from somewhere else:
- AI API keys. Pausing on Vercel doesn't stop AI Gateway usage, and none of these caps reaches your OpenAI or Anthropic account. Give each key its own limit with the AI provider.
- Endpoints anyone can call. A public function that sends email or calls an AI model costs you on every request. Rate-limit it, so one script can't call it thousands of times.
- Other paid services. Email, storage, maps and search providers each have their own billing settings. Each needs a cap, or at least an alert.
These are items 17 to 19 of our free Ship Safe checklist, next to the security checks that usually go wrong first.
Frequently asked questions
Does a Vercel budget stop spending by default?
Not on its own. Vercel made pausing the default in June 2024 (Vercel changelog), but its current documentation describes pausing as a choice you make per budget and says a budget "does not stop usage on its own". Open each budget and check that Pause production deployments is on; otherwise it only notifies you.
Why did my Supabase bill go up with the Spend Cap on?
Because compute and the other opted-in items aren't covered by the cap. An extra project, a preview branch, a read replica or a bigger compute size is billed even with the Spend Cap on.
Can I set a hard spending limit on Azure pay-as-you-go?
No. The built-in spending limit only exists on free and monthly-credit subscriptions. On pay-as-you-go you combine a budget, anomaly alerts and, if you need a hard stop, an automation that shuts resources down.
Get a second pair of eyes
Every Ship Safe audit includes a hosting cost review: we check each of these settings on your actual accounts, read-only, and tell you what's still uncapped. See Ship Safe for what the audit covers and what it costs.
