EC2 burstable instances (T-family): cheap until you burst

Quick answer

Burstable T-family instances (t3, t3a, t4g) are the cheapest general-purpose EC2 type, designed for workloads with low average CPU that occasionally spike. They earn CPU credits while idle and spend them to burst above a baseline; if you exhaust credits, you either get throttled to the baseline (standard mode) or pay per-vCPU-hour surcharges for sustained bursting (unlimited mode). They are excellent value for bursty, low-average-CPU workloads like small web servers, dev boxes, and microservices, but a poor fit for sustained high CPU, where a right-sized non-burstable instance is cheaper and more predictable.

The T-family (t3, t3a, and the Graviton-based t4g) is the cheapest general-purpose EC2 tier, and for the right workload it is a genuine bargain. The catch is the CPU credit system: burstable instances are cheap because they assume your average CPU is low. Understand credits and unlimited mode, or you will either get silently throttled or receive a surprise surcharge.

How CPU credits work

State	What happens
CPU below baseline	Instance earns credits
CPU above baseline	Instance spends credits to burst
Credits exhausted (standard)	Throttled down to baseline
Credits exhausted (unlimited)	Bursts continue, billed per surcharge

Each T instance has a baseline CPU percentage (higher for larger sizes). Run below it and you bank credits; run above it and you spend them. As long as your average CPU stays near or below baseline, you burst for free during spikes and pay only the low hourly rate. The whole value proposition rests on that average staying low, which is why the T-family is ideal for bursty, mostly-idle workloads.

Standard vs unlimited mode

What happens when you run out of credits depends on the mode. In standard mode, the instance is throttled down to its baseline CPU, cheap but potentially slow during sustained load. In unlimited mode (the default on many launch paths), the instance keeps bursting but you pay a surcharge per vCPU-hour of sustained bursting beyond your earned credits. Unlimited mode protects performance but can quietly add cost if the workload bursts far more than expected, a classic entry on any list of hidden AWS costs to watch.

When burstable is the cheapest option

T-family wins for workloads with low average CPU and occasional spikes: small web and API servers, dev and test boxes, low-traffic microservices, build agents, and cron-driven jobs. For these, the T instance is cheaper than any comparable fixed-performance instance because you are not paying for CPU you rarely use. Choosing t4g (Graviton) over t3 adds another discount for the same reasons covered in Graviton cost savings.

When to avoid burstable

If your workload runs sustained high CPU (a busy application server, a compute-heavy service, a database under constant load), burstable is the wrong tool. You will either be throttled (standard mode) or pay ongoing unlimited-mode surcharges that can exceed the cost of a right-sized non-burstable instance. For steady load, pick a fixed-performance family from the EC2 instance families and size it to real usage. The rule: burstable for bursty, non-burstable for steady. To understand how the hourly rate itself is built, see EC2 pricing explained.

Monitor credits and mode

Watch CPU credit balance and unlimited-mode surcharge metrics. A steadily declining credit balance means the workload is outgrowing the T-family and should move to a fixed-performance instance. Recurring unlimited-mode charges are a signal the same. Set the mode deliberately (standard where throttling is acceptable, unlimited where performance matters and bursts are truly occasional) rather than accepting whatever default the launch path picked.

FAQ

What are EC2 burstable instances?

The T-family (t3, t3a, and Graviton-based t4g) of general-purpose EC2 instances designed for workloads with low average CPU that occasionally spike. They are the cheapest EC2 tier because they use a CPU credit system: the instance earns credits while running below a baseline CPU level and spends them to burst above it. They suit bursty, mostly-idle workloads rather than sustained high load.

How do CPU credits work?

Each T instance has a baseline CPU percentage. When it runs below baseline it earns credits; when it runs above baseline it spends credits to burst. As long as average CPU stays near or below baseline, you burst for free during spikes and pay only the low hourly rate. If you exhaust credits, the instance is either throttled (standard mode) or charged a surcharge (unlimited mode).

What is the difference between standard and unlimited mode?

In standard mode, when credits run out the instance is throttled down to its baseline CPU, keeping cost low but potentially slowing under sustained load. In unlimited mode, the instance keeps bursting even after credits are exhausted, but you pay a surcharge per vCPU-hour of sustained bursting. Unlimited protects performance at the risk of added cost; standard caps cost at the risk of throttling.

When should I use burstable instances?

For workloads with low average CPU and occasional spikes: small web and API servers, dev and test boxes, low-traffic microservices, build agents, and cron jobs. For these, a T instance is cheaper than any fixed-performance instance because you avoid paying for CPU you rarely use. Choosing t4g (Graviton) over t3 adds a further discount for compatible workloads.

When should I avoid burstable instances?

When your workload runs sustained high CPU, such as a busy application server, compute-heavy service, or constantly-loaded database. Under steady load you will either be throttled in standard mode or pay ongoing unlimited-mode surcharges that can exceed a right-sized non-burstable instance. For steady load, pick a fixed-performance family and size it to real usage instead.

Can burstable instances cost more than expected?

Yes, in unlimited mode. If a workload bursts far more than anticipated, sustained unlimited-mode surcharges accumulate and can quietly exceed the cost of a fixed-performance instance. Monitor the CPU credit balance and unlimited-mode surcharge metrics: a steadily declining credit balance or recurring surcharges signal the workload has outgrown the T-family and should move to a non-burstable instance.

What to do next

Price your EC2 instances, burstable or not, before you launch them. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.