[Update] T9g Instances Powered by Graviton5 Have Been Added to Amazon EC2
This page has been translated by machine translation. View original
Hello. I'm Takeda from the Service Development Department.
We regularly check the list of instance types internally. In that check, a new t9g has appeared.
When I looked into it, t9g is listed on the Amazon EC2 On-Demand pricing page, and it is also returned by the API. In the file that compiles pricing (Price List), the processor is listed as AWS Graviton5 Processor. Until now, the only Graviton-based T instance was the Graviton2 T4g.
As of November 20, 2026, I could not find any mention in What's New announcements, product pages, or the User Guide. This time, I looked at the information obtainable with the AWS CLI, along with the pricing.
Four sizes appear in the APIs for Tokyo and N. Virginia
When I filtered describe-instance-type-offerings by t9* in the Tokyo region, four sizes were returned. The AZs are three: apne1-az1, apne1-az2, and apne1-az4.
aws ec2 describe-instance-type-offerings \
--region ap-northeast-1 \
--location-type availability-zone-id \
--filters 'Name=instance-type,Values=t9*' \
--query 'InstanceTypeOfferings[].[InstanceType,Location]' --output text
(apne1-az1の分を抜粋)
t9g.medium apne1-az1
t9g.micro apne1-az1
t9g.nano apne1-az1
t9g.small apne1-az1
The sizes are four: nano, micro, small, and medium. This is the same lineup as the T8i that came out in September 2026. In N. Virginia as well, the same four sizes appeared in use1-az1, use1-az2, and use1-az4. The Osaka region returned 0 results and is not listed on the pricing page either.
Next, I ran run-instances with --dry-run in the Tokyo region. The AMI is the arm64 version of Amazon Linux 2023.
aws ec2 run-instances --dry-run \
--region ap-northeast-1 \
--image-id ami-0bf2a2ea0e0044148 \
--instance-type t9g.micro \
--subnet-id subnet-xxxxxxxxxxxxxxxxx
An error occurred (DryRunOperation) when calling the RunInstances operation: Request would have succeeded, but DryRun flag is set.
If you specify t9g.large, you get Invalid value 't9g.large' for InstanceType.
Tokyo region pricing is about 1.24 times that of T4g
I lined up the Tokyo region Linux On-Demand pricing against other T instances. The unit is USD/hour.
| Size | t9g | t4g | t3 | t3a | t8i |
|---|---|---|---|---|---|
| nano | 0.00670 | 0.0054 | 0.0068 | 0.0061 | 0.00816 |
| micro | 0.01339 | 0.0108 | 0.0136 | 0.0122 | 0.01632 |
| small | 0.02678 | 0.0216 | 0.0272 | 0.0245 | 0.03264 |
| medium | 0.05357 | 0.0432 | 0.0544 | 0.0490 | 0.06528 |
Comparing the hourly price for the same size, T9g is about 24% more expensive than T4g and about 1.5% cheaper than T3. Compared with T8i, it is about 18% cheaper. In N. Virginia, it is almost the same price as T3; for medium, T3 was 0.0416 and T9g was 0.04166.
t9g is listed on the pricing page in the following 12 regions. Of these, the ones I also checked in the API are Tokyo and N. Virginia.
- Tokyo, Singapore, Sydney, Malaysia
- N. Virginia, Ohio, N. California, Oregon
- Canada (Central), São Paulo, Frankfurt, Spain
Looking at the Price Lists for Tokyo and Spain, the version dated October 29, 2026 has no t9g rows, while the version published at 7:09 on November 20 (JST) does.
Only On-Demand rows are listed; there are no Reserved Instance rows. In the Tokyo region's describe-reserved-instances-offerings, t9g.micro and t9g.medium also returned 0 results.
For Spot, spot is included in the API's SupportedUsageClasses. The Tokyo region's Spot price history has records starting from 6:58 on November 20 (JST), and medium was 0.0222 USD/hour. That is about 41% of the On-Demand price for the same size.
Network and EBS limits are higher than T4g
The values from describe-instance-types are as follows. They were the same in Tokyo and N. Virginia. I have included T4g medium for comparison.
| Type | vCPU | Memory | Clock | Network | EBS bandwidth (baseline / max) | EBS IOPS (max) |
|---|---|---|---|---|---|---|
| t9g.nano | 2 | 0.5 GiB | 3.3 GHz | Up to 7.5 Gbps | 43 / 6,000 Mbps | 20,000 |
| t9g.micro | 2 | 1 GiB | 3.3 GHz | Up to 7.5 Gbps | 87 / 6,000 Mbps | 20,000 |
| t9g.small | 2 | 2 GiB | 3.3 GHz | Up to 7.5 Gbps | 174 / 6,000 Mbps | 20,000 |
| t9g.medium | 2 | 4 GiB | 3.3 GHz | Up to 7.5 Gbps | 347 / 6,000 Mbps | 20,000 |
| t4g.medium | 2 | 4 GiB | 2.5 GHz | Up to 5 Gbps | 347 / 2,085 Mbps | 11,800 |
The combinations of vCPU and memory are unchanged from the same sizes of T4g. The maximums for network and EBS have gone up, and the EBS baseline bandwidth is the same value as T4g.
Hibernation support differs from T4g. HibernationSupported is true only for medium, and false for nano through small. T4g is true for all four sizes.
BurstablePerformanceSupported is false
The BurstablePerformanceSupported value in describe-instance-types was false for all four sizes. T4g and T8i are true.
You can specify t9g in get-default-credit-specification, which returns the default CPU credit setting.
aws ec2 get-default-credit-specification \
--region ap-northeast-1 --instance-family t9g
{
"InstanceFamilyCreditSpecification": {
"InstanceFamily": "t9g",
"CpuCredits": "unlimited"
}
}
T4g is likewise unlimited. When I specified M9g, I got the error "Only burstable performance instance families support Unlimited."
The Price List also has a CPUCredits:t9g row, at 0.05 USD/vCPU-hour for Linux. T4g is 0.04 USD and T8i is 0.06 USD, and T3 has the same price.
The credit settings and pricing are handled the same as for burstable families; only BurstablePerformanceSupported is false.
Summary
Even for the same size as T4g, the price goes up, so whether to migrate will probably be decided after looking at the processing performance on Graviton5.
