AWS EC2 T8i burstable instances bring Granite Rapids and up to 70% more compute

What AWS launched

Amazon's newest burstable EC2 generation, the T8i family, runs on the AWS Nitro System with custom sixth-generation Intel Xeon Scalable processors, the Granite Rapids line. According to a dev.to post by AWS Builders, the instances are aimed at workloads with low to moderate CPU demand: microservices, development and test environments, low-traffic websites and batch processing.

How T8i compares to T3

The T3 family launched in 2018, and the dev.to analysis quantifies the generational gap: T8i delivers up to 70% more compute performance, 1.25 times the network bandwidth (up to 6.25 Gbps on t8i.medium versus 5 Gbps on t3.medium) and 2.4 times the EBS bandwidth.

AWS frames the update as up to 30% better price-performance. The write-up cautions that the announcement leaves out a detail: T8i carries roughly 20% higher on-demand rates, with t3.medium listed at $0.046 per hour and t8i.medium at $0.055. Two further caveats: T8i is not yet covered by Savings Plans, and availability may vary by region. There is also a hard ceiling on sizing, since T8i currently ships only in nano, micro, small and medium, with nothing at large or above.

Migration paths

The article lays out four strategies depending on usage pattern and size.

Direct replacement (nano through medium): CPU and memory allocations are unchanged, so the swap carries little performance risk. Where CPU is the bottleneck, the extra headroom can remove it, and if an instance is consuming credits in Unlimited mode, a faster processor burns fewer of them, which can offset the 20% price increase.

Dedicated families for sustained CPU: T3's credit model can become a cost trap. In Standard mode, an exhausted credit balance throttles the instance to its baseline, 30% of a vCPU on t3.large. In Unlimited mode, which is the default, sustained usage above baseline incurs $0.05 per vCPU-hour; a t3.large pinned at 100% CPU accrues $0.10 per hour in surcharges and lands at roughly $0.18 per hour in total. The article points to AMD-based c6a.large at $0.082 per hour and m6a.large at $0.096 per hour, plus Intel's c7i-flex.large at $0.091 per hour, all offering fully dedicated vCPUs with predictable cost.

Flex instances for memory-heavy, CPU-light loads: AWS recommends Flex families for sizes above T8i's ceiling. A t3.xlarge (4 vCPU, 16 GiB, $0.182 per hour) can become an r8i-flex.large (2 vCPU, 16 GiB, $0.148 per hour). The vCPU count is halved, but the write-up notes the R8i processor is about 70% more powerful than T3-era CPUs, and advises monitoring CPU after the switch to confirm the trade holds.

Managed services: The author describes a client running Redis on a t3.micro. The instance cost $0.011 per hour, but its public IPv4 address, which AWS began charging for in 2024, added $0.005, bringing the total to $0.016 per hour, more than the $0.0136 per hour of a managed t4g.micro equivalent. With AWS launching lower-priced Valkey in 2025, a server originally built to save money ended up costing more than the managed alternative.

Why it matters

Burstable instances are where most small workloads land, and the T8i launch redraws the options for teams still running T3 fleets. The right move depends on workload shape rather than a blanket upgrade: nano-to-medium workloads get a real performance lift, large and above are pushed toward Flex or AMD families, and anyone routinely burning credits in Unlimited mode is probably already overpaying relative to a dedicated instance. The gaps in the T8i rollout, namely no large sizes and no Savings Plans coverage yet, mean many architectures will end up split across several families rather than consolidated onto one.