T-family unlimited mode cost: when burstable CPU credits bill extra

Quick answer

Burstable T-family instances (T3, T3a, T4g) provide a baseline level of CPU and earn credits while running below it, then spend those credits to burst above baseline. Under standard mode, once credits run out the instance throttles to baseline. Under unlimited mode (the default on T3 and T4g), the instance keeps bursting and AWS charges about $0.05 per vCPU-hour for the surplus CPU used beyond what credits cover. Unlimited mode saves money for workloads that burst occasionally, but for sustained high CPU it can quietly cost more than simply running a right-sized M or C instance. The rule: T-family fits spiky, low-average workloads; sustained high CPU belongs on a fixed-performance instance.

The T-family is AWS's burstable instance line, and it is the cheapest per hour for good reason: it provides only a baseline fraction of a full vCPU continuously, letting you burst above it by spending credits earned while idle. That model is ideal for spiky, low-average workloads, but it has a cost twist under sustained load, unlimited mode, that can turn a cheap instance into a surprisingly expensive one if you are not watching.

How burstable credits work

Concept	Meaning
Baseline	Guaranteed continuous CPU fraction (varies by size)
Credits earned	Accrue while running below baseline
Credits spent	Consumed while bursting above baseline
Standard mode	Throttles to baseline when credits run out
Unlimited mode	Keeps bursting, charges ~$0.05 per vCPU-hour surplus

A t3.large, for example, has a baseline around 30 percent of two vCPUs. Run it below that and it banks credits; run it above and it spends them. As long as your average CPU stays under baseline, the bursts are effectively free, which is what makes the T-family so cheap for workloads that idle most of the time and spike occasionally.

Standard mode: throttle when credits run out

In standard mode, when an instance exhausts its credits it is throttled back to baseline CPU until it earns more. This caps cost at the instance's hourly price but caps performance too: a sustained-load workload on a standard-mode T-instance runs slowly once credits are gone. Standard mode is predictable on cost and risky on performance for anything that runs hot for long.

Unlimited mode: pay to keep bursting

In unlimited mode (the default on T3 and T4g), the instance keeps bursting past exhausted credits, and AWS charges about $0.05 per vCPU-hour for the surplus CPU used beyond what credits cover. For a workload that bursts briefly and occasionally, this is cheap insurance: you pay a few cents for the rare burst and otherwise pay nothing extra. The problem is a workload that runs at high CPU continuously: the surplus charges accrue every hour, and the total can quietly exceed the cost of a fixed-performance instance that would have handled the load better.

When unlimited mode costs more than a bigger instance

Suppose a t3.large runs continuously at 70 percent CPU, well above its roughly 30 percent baseline. In unlimited mode it pays surplus charges for the extra 40 percent every hour, on top of the base price. Add those up over a month and it can exceed the cost of a right-sized m7g or c7g instance that delivers full, predictable performance with no credit accounting. The T-family stops being cheap exactly when the workload stops being bursty, the mirror of the burstable instance value proposition.

Choosing the mode, and the family

Watch CPU credit balance and surplus charges. If an instance regularly runs out of credits and pays unlimited surplus, that is the signal to move it to a fixed-performance M or C family instance. Use the T-family with unlimited mode for spiky, low-average workloads where bursts are occasional, and use standard mode where you want a hard cost ceiling and can tolerate throttling. Prefer T4g (Graviton) over T3 for the same burstable model at a lower price. Price the T-family against a fixed-performance alternative in the resource catalog so a sustained-load workload does not silently overpay in unlimited mode.

FAQ

What is T-family unlimited mode?

Unlimited mode, the default on T3 and T4g burstable instances, lets an instance keep bursting above its baseline CPU even after it exhausts its earned credits, and AWS charges about $0.05 per vCPU-hour for the surplus CPU used beyond what credits cover. It contrasts with standard mode, which throttles the instance back to baseline when credits run out. Unlimited mode trades a hard cost ceiling for sustained performance.

When does unlimited mode cost extra?

When an instance bursts above its baseline CPU after exhausting its credits, AWS charges about $0.05 per vCPU-hour for the surplus. For occasional bursts this is a few cents of cheap insurance. But for a workload running at high CPU continuously, well above baseline, the surplus charges accrue every hour and can quietly exceed the cost of a right-sized fixed-performance instance that would handle the load better.

When should I not use a T-family instance?

When the workload runs at high CPU continuously rather than in occasional bursts. The T-family is cheap because it provides only a baseline CPU fraction and earns credits while idle; a sustained-load workload exhausts credits and either throttles (standard mode) or pays unlimited surplus charges every hour. At that point a right-sized M or C family instance delivers full, predictable performance and often costs less overall.

What is the difference between T3 standard and unlimited mode?

In standard mode, when an instance exhausts its CPU credits it throttles back to baseline performance until it earns more, capping cost but also capping performance under sustained load. In unlimited mode, it keeps bursting and pays about $0.05 per vCPU-hour for the surplus CPU. Standard mode suits workloads needing a hard cost ceiling; unlimited mode suits spiky workloads where occasional bursts should not be throttled.

How does C3X help with burstable instance cost?

C3X prices your T-family instances from Terraform against a live catalog, so you can compare a burstable instance against a fixed-performance M or C family alternative before you deploy. For a sustained-load workload that would rack up unlimited-mode surplus charges, that comparison surfaces the cheaper fixed-performance option as a concrete number in the pull request, rather than discovering the surplus charges on the monthly bill.

What to do next

Keep burstable instances cheap, not silently expensive. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.