Short answer: T8i is a genuine price-performance upgrade over T3 — AWS claims up to 30% better price performance and up to 70% higher compute performance — but the financial details matter more than the benchmark numbers. Unlimited mode is on by default, Savings Plans don't cover T8i yet, and the instance line tops out at t8i.medium. Here is what to check before you move anything.
Source: New low-cost burstable Amazon EC2 T8i instances are generally available — AWS News Blog, September 17, 2026.
What actually shipped
T8i instances are now generally available across 15+ regions, built on custom sixth-generation Intel Xeon Scalable processors (Granite Rapids) exclusive to AWS. Compared with T3, AWS claims:
| Metric | T8i vs T3 |
|---|---|
| Price performance | up to 30% better |
| Compute performance | up to 70% higher |
| Network bandwidth | up to 1.25x |
| EBS bandwidth | up to 2.4x |
The line is deliberately small — four sizes, all with 2 vCPUs on a single physical core:
| Size | Memory | Baseline per vCPU | Credits earned/hour |
|---|---|---|---|
| t8i.nano | 0.5 GiB | 5% | 3 |
| t8i.micro | 1 GiB | 10% | 6 |
| t8i.small | 2 GiB | 20% | 12 |
| t8i.medium | 4 GiB | 20% | 12 |
Those unusual memory ratios (1:0.25, 1:0.5, 1:1) are the interesting part: they let you pay for the memory a small workload actually uses instead of rounding up a size class to get more RAM. AWS corrected the memory sizes in a note the day after launch, so double-check current specs before planning capacity.
The three cost details that matter more than the benchmark
1. Unlimited mode is the default — again. T8i uses the same CPU credit system as T3, with Standard and Unlimited modes, and Unlimited ships enabled. Unlimited lets the instance burst past its baseline continuously and bills a flat rate per vCPU-hour for the overage. On a build server or a busy staging box, that line item grows quietly. Migrating to T8i does not fix an Unlimited-mode surprise — it re-prices it.
2. Savings Plans are not available yet. On-Demand and Spot are live; Savings Plans are "coming soon" per AWS. If your T3 fleet is covered by a Compute Savings Plan, migrating those instances today drops them back to on-demand pricing until coverage catches up. Sequence the migration: spiky, non-committed workloads first, committed ones later.
3. There is no big sibling. T8i tops out at t8i.medium, and AWS points larger workloads at M8i Flex, which scales up to 16xlarge. This is not a fleet-wide migration story — it is a targeted replacement for the small stuff: dev servers, CI runners, login gateways, freemium backends, low-traffic websites.
Who should move, and who shouldn't
Move if you run small burstable workloads that keep running out of CPU credits today — the 70% compute headroom buys real throttle relief at a better price. Also worth knowing: t8i.micro and t8i.small qualify for the Free Tier, and T8i is shared tenancy only, so Dedicated Hosts and Dedicated tenancy are out.
Stay put if your instances already sit on a different committed family, or if p95 CPU runs well above burstable baselines. That second group doesn't have a burstable workload — it has a rightsizing problem wearing a burstable costume.
What to do this week
- List every T2/T3 instance and its CPU credit trend. In CloudWatch,
CPUSurplusCreditBalanceabove zero is the smoking gun for Unlimited-mode overage. - Migrate credit-starved small instances to T8i — the price-performance claim is most credible exactly there.
- Leave committed workloads on their current family until Savings Plan coverage for T8i is announced.
- Choose Unlimited mode deliberately per workload instead of accepting the default.
This audit decays as fast as you run it — new instances arrive with Unlimited on, old ones drift into constant bursting. Continuous AI-powered cloud cost optimization exists for exactly this: flagging credit burners and mis-fitted families the week they appear, not at the next quarterly review. For the manual baseline, start with our cloud waste audit checklist.
FAQ
Is T8i cheaper than T3? AWS claims better price performance — more work per dollar — not a flat price cut. Whether your bill drops depends on utilization: throttled, bursty workloads benefit most; idle ones just cost less per hour in the same size class.
Should I use Standard or Unlimited mode? Standard if bursts are rare and short — you pay nothing extra and accept throttling. Unlimited only for workloads where an occasional overage beats being throttled. Never leave it as an unexamined default.
Why the odd memory sizes? Decoupling memory from the 2-vCPU core (0.5 to 4 GiB) lets tiny memory-bound workloads stop paying for RAM they don't use — the same logic that made the t3.nano popular, taken further.