AWS launches T8i burstable instances with 30% better price performance than T3
Amazon EC2 T8i instances, powered by custom sixth-generation Intel Xeon Scalable (Granite Rapids) processors and the Nitro System, became generally available on September 17, 2026 with up to 30% better price performance than T3. Low-CPU workloads should move from T3 to T8i without changing their credit model.
September 17, 2026. AWS announces general availability of the EC2 T8i instances. September 18, 2026. The post corrects the memory figures in the spec table. 2018. The T3 generation they replace had already established the burstable model. Why it matters: the cheapest entry point in the EC2 catalog changes generation, with the same credit system but a substantially newer processor.
The T8i instances run on custom sixth-generation Intel Xeon Scalable processors, codename Granite Rapids, available only on AWS, paired with the Nitro System. They sit among the cheapest EC2 instances in the catalog while claiming up to 30% better price performance than the previous T3 generation.
The burstable model, unchanged in principle
The T family rests on a simple idea: an instance offers a baseline level of CPU performance, expressed as a percentage of the vCPU, plus the ability to burst above it by spending CPU credits. Credits accumulate when the instance runs below baseline and are spent when it accelerates.
This model suits workloads whose CPU usage is low to moderate most of the time, with occasional spikes: freemium services, demo environments, staging and development, batch data processing, microservices, low-traffic websites, and login gateways. That is exactly the target AWS describes for T8i.
The mechanism does not change between T3 and T8i: the same Standard and Unlimited modes, with Unlimited still enabled by default on T8i. That is what makes migration nearly transparent for operators.
What T8i brings
Compared to T3, the announced gains break down into four points:
- Up to 30% better price performance, lowering total cost of ownership at equivalent workload.
- Up to 70% higher compute performance, on a newer vCPU.
- Up to 1.25× the network bandwidth of T3.
- Up to 2.4× the EBS bandwidth of T3.
These figures rest on the pairing of the Nitro System and Granite Rapids. Nitro offloads networking, storage, and security from the hypervisor to dedicated hardware, freeing the processor for the application, while Granite Rapids delivers the generational core gain.
The lineup is limited for now to four sizes, all with two vCPUs presented as a single core, with distinct memory ratios:
| Instance | vCPUs | Memory (GiB) | Baseline /vCPU | Credits /hour | Network burst (Gbps) |
|---|---|---|---|---|---|
| t8i.nano | 2 | 0.5 | 5% | 3 | up to 6.25 |
| t8i.micro | 2 | 1 | 10% | 6 | up to 6.25 |
| t8i.small | 2 | 2 | 20% | 12 | up to 6.25 |
| t8i.medium | 2 | 4 | 20% | 12 | up to 6.25 |
As on T3, these sizes offer vCPU-to-memory ratios that are rare elsewhere in EC2 — 1:0.25, 1:0.5, and 1:1. For workloads that outgrow the medium format, AWS points to M8i Flex, announced with the same 30% price-performance gain and scaling up to 16xlarge.
Standard or Unlimited: how the credits work
The credit model is worth spelling out, because it is what makes the T family cheap. Each instance size earns a fixed number of credits per hour — from 3 on the t8i.nano to 12 on the t8i.medium — and spends them when it exceeds its baseline. One credit corresponds to one minute of vCPU at 100%, and unused credits accumulate up to a cap.
In Standard mode, an instance that exhausts its credits falls back to its baseline, regardless of what the workload demands. In Unlimited mode, the instance can keep bursting beyond its balance, at a per-use surcharge. Unlimited is the default on T8i: the right choice for a service that must never slow down, and the wrong one for anyone who wants a perfectly predictable bill. Knowing which mode an instance runs in is part of cost discipline, because unprovisioned burst always shows up on the invoice.
T8i or T4g: x86 versus ARM
The T generation is not the only burstable family in the catalog. The T4g instances, powered by AWS Graviton processors on ARM, occupy the same niche at an often lower price. The choice between T8i and T4g comes down to compatibility and ecosystem: a workload whose toolchain is ready for ARM — multi-architecture containers, compiled libraries, native images — will often find better price-performance on T4g.
T8i serves those who cannot or will not leave x86: legacy binaries, dependencies without ARM builds, or a simple compatibility preference with an existing fleet. AWS deliberately keeps both paths open, and the right reflex is to test your workload on both families before committing. A T3-to-T8i move is frictionless, but a T3-to-T4g move can be just as smooth if the application is portable — in which case the savings often land on the Graviton side.
What T8i does not solve
The T8i lineup is deliberately narrow: four sizes, two vCPUs, memory capped at 4 GiB. For a workload that needs more cores or memory, the burstable model stops making sense, and AWS points to M8i Flex. It is also worth remembering that a T instance, even in Unlimited, bills the burst: a workload in near-permanent peak ends up costing more than a fixed-capacity instance. Burstable rewards a low average and a short peak, not the reverse.
A migration checklist
Moving from T3 to T8i is mostly a lift-and-shift exercise, but three steps make it clean. First, review the credit mode on your current instances: an instance in Standard that occasionally throttles is a candidate for Unlimited on the new generation, and vice versa. Second, right-size the memory ratio — T8i offers the same 1:0.25, 1:0.5, and 1:1 ratios, so match by workload profile rather than by habit. Third, test in a staging region before cutting over production, since T8i launched in a limited set of regions and your burst pattern may differ on a newer core. Above all, revisit the choice after a month of real traffic rather than trusting the benchmark: the 30% figure is an average across workloads, and your own peak pattern decides whether it materializes.
Availability and purchasing
The T8i instances are available now in US East (N. Virginia, Ohio), US West (Oregon, N. California), Asia Pacific (Hyderabad, Malaysia, Mumbai, Seoul, Singapore, Sydney, Tokyo), Canada (Central), and Europe (Frankfurt, Ireland, London, Paris).
Purchasing is available through On-Demand and Spot, with Savings Plans coming soon. The instances support shared tenancy only, with no Dedicated tenancy or Dedicated Hosts. The t8i.micro and t8i.small sizes qualify for the AWS Free Tier, making them an immediate entry point for new accounts.
Verdict
If you run T3 instances for light or intermittent workloads, moving to T8i is one of the rare no-risk modernization moves: same sizes, same credit system, same Unlimited mode, but a faster vCPU and lower cost per unit of work. If you are starting a new project on AWS, pick T8i directly for low-utilization components — API services, cron jobs, staging environments — rather than a fixed-capacity instance you will never saturate. If your workload is continuously intensive, look elsewhere: T8i is built for bursts, not sustained compute, and an M or C family will cost less over time.