Most teams set up one budget alert at 100% and call it done. That alert fires the day the budget is already blown. A working setup needs three tiers (50% warn, 80% alert, 100% panic) and one in each cloud you actually use. Here is the click-by-click setup for AWS, GCP, and Azure in 2026, plus the four mistakes that quietly defeat the alerts.
If you only have 60 seconds, this is the shape:
A single late-night Bedrock deployment can spend $4,000 in eight hours. A p5.48xlarge
left running over a weekend is $1,150. The shape of cloud incidents shifted because of two changes.
AI workloads make the burn rate non-linear. Pre-2024, a misconfigured S3 lifecycle policy might add 5% to the month over weeks. Today, one forgotten GPU job hits double-digit percentage of monthly budget within hours.
Multi-cloud is now the median. Forrester's 2026 cloud survey shows 72% of enterprises run workloads in at least two of AWS, GCP, and Azure. A budget alert that only watches AWS misses the half of the bill that lives elsewhere.
The fix is not more dashboards. It is one consistent alerting setup across every cloud account, with tiered thresholds and tiered routing.
I use a three-tier framework that works the same way on every cloud. The thresholds are calibrated against typical monthly burn rates.
The three tiers do something the single-alert approach cannot: they create lead time. By the time you hit 100%, you have already had two warnings to act.
AWS Budgets is the native tool. Free up to two budgets per account, $0.02 per additional budget per day after that.
Google Cloud's Budgets and Alerts page is under Billing → Budgets & alerts. The setup is similar but the routing model is different.
Azure splits the concept into Cost alerts (anomaly-style) and Budgets (threshold-style). For a tiered setup, use Budgets.
Every "we got blindsided" cost incident I have seen comes down to one of these.
The most common pattern. The team sets one alert at 100%, gets it on day 28, and has no time to act. Always tier.
Email alerts go to one person who is on vacation. SNS, Pub/Sub, or an action group routed to Slack and PagerDuty gives the alert a chance of being seen.
A budget set at $5,000/month when the account was new fires nonstop after three months of growth. Set a quarterly review on the calendar.
Actual-only alerts are reactive. Adding a Forecasted alert at 100% catches the spike before it bills.
The honest part.
They are not anomaly detection. A 30% daily spike on a single service can stay under the monthly budget threshold and never fire. Pair budgets with cost anomaly detection (AWS Cost Anomaly Detection, GCP recommender, Azure Cost Alerts).
They lag by 1 to 2 days. Even Forecasted alerts use yesterday's data. For real-time, you need a tool that reads the billing stream directly.
They do not act. An alert tells you, it does not stop a runaway resource. Auto-remediation tools like nOps and ZopNight close that gap.
Are budget alerts enough on their own?
For accounts under $20,000 per month of spend, usually yes if you set the tiers correctly. Above that, pair with anomaly detection. Should I set budgets per project or per account?
Per project for engineering teams, per account or per billing group for finance reporting. The two views answer different questions.
Why is my Actual alert quiet but Forecasted is firing?
Forecasted reads the burn rate and projects forward. If your spend curve is bending up, Forecasted catches it before the month's Actual closes. Treat a forecasted-only fire as a real anomaly, not a false alarm.
Can I get a single multi-cloud budget?
Not natively. CloudZero, Vantage, and ZopNight join the three cloud bills into one budget object. The native consoles are per-cloud only.
How do I avoid alert fatigue?
Use the tier model. Tier 1 goes to a Slack channel nobody pings on. Tier 2 to the FinOps person. Tier 3 pages. Most fires stop at Tier 1, which is the point.
If you set up budgets a year ago and have not touched them since, the question worth asking is what the current month's actual spend is as a percentage of that old budget. Drop the ratio in the comments. I will tell you whether the budget needs a refresh or your spend curve needs a real anomaly tool.