AWS History and Timeline regarding Amazon EC2 - Overview, Functions, Features, Summary of Updates, and Introduction
First Published:
Last Updated:
This time, I have created a historical timeline for Amazon Elastic Compute Cloud (Amazon EC2), the resizable virtual server service that was announced as a limited public beta on August 25, 2006 and that effectively launched the modern public cloud era. Amazon EC2 has been continuously evolving for nearly two decades, and it remains the foundational compute layer of AWS today.
Just like before, I am summarizing the main features while following the birth of Amazon EC2 and tracking its feature additions and updates as a Current Overview, Functions, Features of Amazon EC2.
I hope these will provide clues as to what has remained the same and what has changed, in addition to the features and concepts of each AWS service.
Background and Method of Creating Amazon EC2 Historical Timeline
The reason for creating a historical timeline of Amazon EC2 this time is that Amazon EC2 is approaching its 20th anniversary since the public beta launch in 2006, and it is one of the oldest and most fundamental services in AWS.Another reason is that since Amazon EC2 was announced in August 2006, the service has expanded along multiple axes — instance families, purchasing options, networking, storage, hypervisors, and processor architectures (x86, AWS Graviton, Apple Silicon for macOS). Therefore, I wanted to organize the information of Amazon EC2 with the following approaches.
- Tracking the history of Amazon EC2 and organizing the transition of updates
- Summarizing the feature list and characteristics of Amazon EC2
There may be slight variations in the dates on the timeline due to differences in the timing of announcements or article postings in the references used.
The content posted is limited to major features related to the current Amazon EC2 and necessary for the feature list and overview description.
In other words, please note that the items on this timeline are not all updates to Amazon EC2 features, but are representative updates that I have picked out.
Amazon EC2 Historical Timeline (Updates from August 25, 2006)
Now, here is a timeline related to the functions of Amazon EC2. As of the time of writing this article, the history of Amazon EC2 spans nearly 20 years from August 2006.* The table can be sorted by clicking on the column names.
| Date | Summary |
|---|---|
| 2006-08-25 | Amazon EC2 is announced as a limited public beta. It provides resizable Linux virtual servers (instances) in the cloud, billed by the hour, with a single instance type (m1.small) in a single region (US East). |
| 2007-10-16 | Amazon EC2 begins offering Large and Extra Large instance types. This is the first expansion beyond the original m1.small. |
| 2008-03-26 | Static IP addresses (Elastic IP) and Availability Zones are introduced. Elastic IP allows remapping of public IPs across instances, and Availability Zones provide failure isolation within a region. |
| 2008-05-08 | Amazon EC2 high CPU instance types (c1.medium, c1.xlarge) are introduced. |
| 2008-08-20 | Amazon Elastic Block Store (Amazon EBS) is announced as a public beta. It provides off-instance persistent block storage that survives instance termination. |
| 2008-10-23 | Amazon EC2 reaches General Availability (GA) and exits beta. A Service Level Agreement (SLA) of 99.95% availability is published, and Microsoft Windows AMIs become available. |
| 2008-12-10 | Amazon EC2 launches in the Europe (Ireland) region (eu-west-1). This is the first AWS region outside the United States. |
| 2009-03-12 | Amazon EC2 Reserved Instances are introduced. Customers can make a one-time payment for a capacity reservation in exchange for a lower hourly rate. |
| 2009-05-18 | Amazon CloudWatch, Auto Scaling, and Elastic Load Balancing (ELB) are launched. These three services let customers monitor EC2, automatically scale instance groups, and distribute traffic across instances. |
| 2009-08-25 | Amazon Virtual Private Cloud (Amazon VPC) is announced. It allows customers to provision logically isolated virtual networks for EC2 instances and connect them to on-premises networks via VPN. |
| 2009-10-08 | High-memory Double Extra Large and Quadruple Extra Large instances (m2 family) are introduced. |
| 2009-12-14 | Amazon EC2 Spot Instances are announced. Customers can bid for spare EC2 capacity at variable prices, dramatically lowering the cost of fault-tolerant workloads. |
| 2010-02-22 | Cluster Compute instances (cc1.4xlarge) are introduced for HPC workloads. They use 10 Gigabit Ethernet and placement groups for low-latency communication. |
| 2010-07-13 | Amazon EC2 Micro Instances (t1.micro) are introduced. They provide low-cost burstable capacity for low-throughput applications. |
| 2010-09-15 | Amazon EC2 supports Resource Tagging. Tags allow customers to organize and categorize EC2 resources with key/value labels. |
| 2011-03-02 | Amazon EC2 launches in the Asia Pacific (Tokyo) region (ap-northeast-1). The first AWS Region in Japan, providing low-latency access for customers in East Asia. [Source] |
| 2011-04-04 | Amazon EC2 introduces Cluster GPU instances (cg1.4xlarge) with NVIDIA Tesla M2050 GPUs. This is the first GPU-accelerated instance family on Amazon EC2. |
| 2011-08-01 | Amazon EC2 launches Dedicated Instances. Customers can run instances on hardware that is physically isolated at the host level for compliance reasons. |
| 2011-08-04 | Amazon EC2 VM Import is announced. Existing VMware ESX, VMware Workstation, Microsoft Hyper-V, and Citrix Xen virtual machines can be imported to Amazon EC2 as AMIs. |
| 2011-08-23 | Amazon Virtual Private Cloud (Amazon VPC) becomes Generally Available (GA). |
| 2012-03-13 | Amazon EC2 Reserved Instance Marketplace is announced. Customers can sell unused Reserved Instances and buy them from other customers. |
| 2012-07-31 | Amazon EC2 Provisioned IOPS volumes for Amazon EBS are introduced. They provide consistent, predictable storage performance up to 1,000 IOPS per volume. |
| 2012-10-11 | Amazon EC2 High I/O instances (hi1.4xlarge) with local SSD storage are introduced. |
| 2012-10-26 | Amazon Glacier integration and second-generation Standard EBS volumes are announced. |
| 2012-12-20 | Amazon EC2 second-generation general-purpose instances (m3 family) are introduced. Built on Intel Xeon E5 processors with SSD-backed instance storage. |
| 2013-04-19 | Amazon EC2 High Storage instances (hs1.8xlarge) are introduced for data warehousing. They provide 48 TB of local hard-disk storage. |
| 2013-11-13 | Amazon EC2 introduces Enhanced Networking with SR-IOV on C3 instances. SR-IOV provides higher packet-per-second performance and lower jitter. |
| 2013-11-13 | Amazon EC2 C3 (compute-optimized), G2 (GPU), and I2 (storage-optimized) instance families are introduced. |
| 2013-12-04 | Default VPC networking begins replacing EC2-Classic for new AWS accounts. New AWS accounts created on or after 2013-12-04 launch all Amazon EC2 instances inside a default Amazon VPC, and accounts created from late 2017 onward cannot opt into EC2-Classic at all. |
| 2014-04-09 | Amazon EC2 third-generation memory-optimized instances (r3 family) are introduced. |
| 2014-06-30 | Amazon EC2 introduces the t2 family of burstable general-purpose instances. They accrue CPU credits during idle periods that can be spent during burst periods. |
| 2014-11-12 | Amazon EC2 announces the c4 family (next-generation compute-optimized instances). |
| 2014-11-13 | AWS Lambda is announced (built on the Amazon EC2 fleet). While AWS Lambda is a separate service, its launch marks the transition from EC2-only compute to event-driven serverless compute. |
| 2015-04-08 | Amazon EC2 Container Service (Amazon ECS) becomes Generally Available (GA). Customers can run Docker containers across managed Amazon EC2 clusters. |
| 2015-05-18 | Amazon EC2 Spot Fleet is announced. Spot Fleet lets customers request and manage thousands of Spot Instances across multiple instance types and Availability Zones with a single API. [Source] |
| 2015-06-11 | Amazon EC2 Dedicated Hosts are announced. Customers gain visibility into and control over the physical server, enabling per-socket and per-core software licensing models. |
| 2015-06-11 | Amazon EC2 introduces the M4 family of general-purpose instances. Five sizes (m4.large through m4.10xlarge, up to 40 vCPUs) powered by custom Intel Xeon E5-2676 v3 (Haswell) processors. m4.16xlarge with 64 vCPUs is added in 2016. [Source] |
| 2015-10-07 | Amazon EC2 X1 family of memory-optimized instances is announced at AWS re:Invent 2015 (preview). Designed for in-memory databases and SAP HANA workloads, scaling up to 2 TB of memory at General Availability the following year. |
| 2016-04-06 | Amazon EC2 introduces the D2 family of dense-storage instances. Up to 48 TB of HDD storage per instance for data warehousing and Hadoop. |
| 2016-05-18 | Amazon EC2 X1 instances (x1.32xlarge) reach General Availability (GA). The largest Amazon EC2 memory-optimized instance at the time, with 2 TB of DDR4 memory, 128 vCPUs on quad-socket Intel Xeon E7 8880 v3, and SSD-backed instance storage — targeted at SAP HANA, Apache Spark, Presto, and other large in-memory workloads. [Source] |
| 2016-09-29 | Amazon EC2 F1 instances (with Xilinx UltraScale+ FPGAs) are announced. First field-programmable gate array (FPGA) instance family on Amazon EC2. |
| 2016-11-30 | Amazon EC2 P2 instances (with NVIDIA K80 GPUs) are launched. |
| 2016-11-30 | Amazon EC2 R4 (memory-optimized) and T2 enhancements are introduced. |
| 2016-11-30 | Amazon EC2 Elastic Network Adapter (ENA) is announced. ENA delivers up to 25 Gbps of network bandwidth on supported instance types. |
| 2017-09-13 | Amazon EC2 Bare Metal instances (i3.metal) are previewed. Customers can run workloads directly on the physical server with no hypervisor. |
| 2017-11-06 | Amazon EC2 C5 instances are announced as the first instance family built on the AWS Nitro System. Nitro offloads virtualization, networking, and storage to dedicated hardware. |
| 2017-11-13 | Amazon EC2 G3 instances (with NVIDIA Tesla M60 GPUs) are launched. |
| 2017-11-13 | Amazon EC2 Elastic GPUs become Generally Available (GA). Customers can attach low-cost GPUs to any Amazon EC2 instance for graphics acceleration. |
| 2017-11-28 | The AWS Nitro System (formally branded) is introduced. Nitro Cards, the Nitro Security Chip, and the Nitro Hypervisor are described as the foundation for all current-generation Amazon EC2 instances. |
| 2017-11-29 | Amazon EC2 Spot Instance pricing model is simplified. Spot prices change less frequently based on long-term supply/demand trends, and bidding is eliminated in favor of a maximum-price specification. |
| 2017-11-29 | Amazon EC2 launches H1 instances for big data workloads. |
| 2018-05-17 | Amazon EC2 introduces the z1d family of high-frequency instances. Custom Intel Xeon Scalable processors run at up to 4.0 GHz sustained for high-CPU-frequency workloads (EDA, financial modeling). |
| 2018-11-26 | Amazon EC2 A1 instances are announced as the first instances built on AWS Graviton (1st generation Arm processors). Announced at AWS re:Invent 2018, A1 introduces custom Arm Cortex-A72 cores for scale-out workloads such as web servers, containerized microservices, and caching fleets. |
| 2018-11-26 | Amazon EC2 P3dn.24xlarge instances are launched. Optimized for ML training with NVIDIA V100 GPUs and 100 Gbps networking. |
| 2018-11-26 | Amazon EC2 introduces the c5n family with 100 Gbps networking. |
| 2018-11-28 | AWS Outposts are announced. AWS-managed compute and storage racks are delivered to customer-owned data centers, running Amazon EC2 instances locally. |
| 2018-11-28 | Amazon EC2 instances support Elastic Fabric Adapter (EFA). EFA enables OS-bypass networking for HPC and ML workloads. |
| 2018-11-29 | Amazon EC2 m5a, r5a, and t3a instances based on AMD EPYC processors are announced. First AMD-based Amazon EC2 instance families. |
| 2019-08-12 | Amazon EC2 Spot Instance pricing becomes more stable. Two-minute interruption notices and instance hibernation on Spot interruption are reinforced. |
| 2019-11-06 | AWS Savings Plans are announced. A flexible pricing model that offers lower prices in exchange for a one- or three-year commitment to a consistent amount of compute usage. |
| 2019-12-03 | Amazon EC2 introduces the M6g, C6g, and R6g families based on AWS Graviton2 processors (custom Arm Neoverse N1 cores). |
| 2019-12-03 | Amazon EC2 Inf1 instances (with AWS Inferentia accelerators) are announced. Purpose-built ML inference instances. |
| 2019-12-03 | AWS Local Zones are announced. Extensions of an AWS Region placed in major metropolitan areas (Los Angeles first) for single-digit millisecond latency to end users. |
| 2019-12-03 | AWS Wavelength is announced. Amazon EC2 instances embedded inside 5G telecommunications carrier networks (Verizon, KDDI, Vodafone, SK Telecom). |
| 2019-12-03 | AWS Nitro Enclaves are announced. Isolated CPU/memory environments inside Amazon EC2 instances for processing highly sensitive data. |
| 2020-11-30 | Amazon EC2 Mac instances (mac1) are announced as the first macOS instances on Amazon EC2. Built on Apple Mac mini with Intel Core i7 (Coffee Lake) and the AWS Nitro System. |
| 2020-12-01 | Amazon EC2 introduces the c6gn family with 100 Gbps networking on AWS Graviton2. |
| 2020-12-01 | Amazon EC2 R5b instances are announced. EBS-optimized variants of r5 with up to 60 Gbps of EBS bandwidth. |
| 2020-12-01 | AWS Nitro Enclaves becomes Generally Available (GA). |
| 2021-04-29 | Amazon EC2 X2gd instances are announced. Graviton2-based memory-optimized instances with NVMe SSD storage. |
| 2021-11-09 | Amazon EC2 introduces the G5 (NVIDIA A10G) and G5g (Graviton2 + NVIDIA T4G) instance families. |
| 2021-11-30 | Amazon EC2 C7g instances powered by AWS Graviton3 processors are previewed. Graviton3 delivers ~25% better compute performance than Graviton2 and adds DDR5 memory plus 200 Gbps Elastic Fabric Adapter (EFA) support on selected variants. |
| 2021-11-30 | Amazon EC2 Trn1 instances (with AWS Trainium accelerators) are previewed. Purpose-built for ML training at scale. |
| 2021-11-30 | Amazon EC2 M6a instances (AMD EPYC 3rd generation) are announced. |
| 2021-11-30 | Amazon EC2 Im4gn and Is4gen storage-optimized instances powered by Graviton2 with AWS Nitro SSDs are announced. |
| 2021-11-30 | Amazon EC2 Hpc6a instances are announced. Purpose-built for tightly coupled HPC workloads at lower cost per core. |
| 2021-12-01 | Amazon EC2 Mac instances based on the Apple M1 chip (mac2) are previewed. The first Arm-based macOS instances on Amazon EC2. |
| 2022-05-23 | Amazon EC2 C7g instances become Generally Available (GA). First general availability of an AWS Graviton3-based instance family. |
| 2022-07-12 | Amazon EC2 Mac M1 (mac2) instances become Generally Available (GA). |
| 2022-08-15 | Amazon EC2-Classic networking is retired in all AWS Regions. All instances now run inside Amazon VPC. |
| 2022-11-28 | AWS re:Invent 2022 announces Trn1n, Hpc7g, R7iz, C6in, M6in, and Inf2 (preview) instances. |
| 2023-04-13 | AWS Graviton3E variants and Hpc7g are made Generally Available (GA). Graviton3E delivers higher vector and floating-point performance for HPC. |
| 2023-04-13 | Amazon EC2 Inf2 instances (AWS Inferentia2) become Generally Available (GA). Purpose-built for large-scale generative AI inference. |
| 2023-06-23 | Amazon EC2 Trn1n instances become Generally Available (GA). They double the network bandwidth of Trn1 to 1,600 Gbps via second-generation EFA. |
| 2023-06-23 | Amazon EC2 introduces the M7a (AMD EPYC 4th gen), C7gn, and Hpc7a families. |
| 2023-08-08 | Amazon EC2 Mac2-m2 instances (Apple M2) are announced. |
| 2023-09-13 | Amazon EC2 Mac2-m2pro instances (Apple M2 Pro) are announced. |
| 2023-11-01 | Amazon EC2 Capacity Blocks for ML are announced. Customers can reserve a fixed amount of Amazon EC2 UltraCluster capacity (P5 instances) for ML training jobs starting on a future date, purchasable up to 8 weeks in advance for 1–14 day durations. [Source] |
| 2023-11-28 | AWS Graviton4 processors are announced (R8g family preview). Graviton4 delivers up to 30% better compute performance, 50% more cores, and 75% more memory bandwidth than Graviton3. |
| 2023-11-28 | Amazon EC2 Trn2 instances powered by AWS Trainium2 are previewed. Trn2 UltraServers connect multiple Trainium2 instances via NeuronLink for training trillion-parameter models. |
| 2023-11-28 | Amazon EC2 Capacity Blocks for ML become Generally Available (GA). |
| 2024-02-13 | Amazon EC2 M7i and M7i-flex (Intel Sapphire Rapids) instances are made Generally Available (GA). |
| 2024-07-10 | Amazon EC2 R8g instances powered by AWS Graviton4 become Generally Available (GA). First general availability of a Graviton4-based instance family, with up to 30% better compute performance, 50% more cores, and 75% more memory bandwidth than R7g powered by Graviton3. |
| 2024-07-23 | Amazon EC2 G6 instances (with NVIDIA L4 Tensor Core GPUs) reach General Availability (GA). Designed for graphics workloads, ML inference, and small-to-medium model training, G6 succeeds the G5 family. [Source] |
| 2024-08-29 | Amazon EC2 G6e instances (with NVIDIA L40S Tensor Core GPUs) reach General Availability (GA). Targeted at generative AI inference, 3D simulations, and high-end graphics workloads with up to 8 NVIDIA L40S GPUs per instance. [Source] |
| 2024-09-18 | Amazon EC2 X8g instances (AWS Graviton4 memory-optimized) reach General Availability (GA). Up to 3 TiB DDR5 memory and 192 vCPUs, the first memory-optimized Graviton4 family. [Source] |
| 2024-09-25 | Amazon EC2 C8g and M8g instances (AWS Graviton4) reach General Availability (GA). The first compute-optimized and general-purpose Graviton4 families. [Source] |
| 2024-11-12 | Amazon EC2 Capacity Blocks for ML expand to additional AWS Regions. Increases geographic coverage for reservation-based GPU/Trainium capacity. [Source] |
| 2024-12-03 | Amazon EC2 Trn2 instances reach General Availability (GA) and Trn2 UltraServers enter preview. Trn2 instances are powered by AWS Trainium2; Trn2 UltraServers link 64 Trainium2 accelerators with NeuronLink for training trillion-parameter foundation models. [Source] |
| 2025-01-08 | AWS Local Zone in New York City reaches General Availability (GA). Extends low-latency Amazon EC2 capacity to a new metropolitan area. [Source] |
| 2025-12-02 | Amazon EC2 C8a compute-optimized instances (5th Gen AMD EPYC Turin, up to 4.5 GHz) reach General Availability (GA) at AWS re:Invent 2025. Up to ~30% higher compute performance vs. C7a, with up to 75 Gbps network throughput and 60 Gbps EBS bandwidth. [Source] |
| 2025-12-02 | Amazon EC2 Trn3 UltraServers powered by AWS Trainium3 reach General Availability (GA) at AWS re:Invent 2025. A single UltraServer interconnects up to 144 Trainium3 chips with NeuronLink to train and serve the largest frontier foundation models. [Source] |
| 2025-12-04 | Amazon EC2 M9g instances powered by AWS Graviton5 processors are previewed at AWS re:Invent 2025. First-generation Graviton5 general-purpose family, succeeding Graviton4. [Source] |
| 2025-12-12 | AWS Dedicated Local Zones expand supported services. Includes Gen-7 Amazon EC2 instances, Amazon S3 Express One Zone, and local AMI support for sovereignty workloads. [Source] |
| 2026-02-12 | Amazon EC2 M8azn instances (AMD EPYC Turin, up to 5 GHz) reach General Availability (GA). High-frequency general-purpose option for latency-sensitive single-thread workloads. [Source] |
| 2026-04-27 | Amazon EC2 C8ine and M8ine instances (custom 6th-gen Intel Xeon Scalable, AWS-exclusive) reach General Availability (GA). AWS-custom Intel silicon for compute- and general-purpose workloads. [Source] |
Current Overview, Functions, Features of Amazon EC2
From here, we will explain in detail the main features of the current Amazon EC2.Amazon EC2 (Elastic Compute Cloud) is a web service that provides secure, resizable compute capacity in the cloud.
It allows customers to launch virtual servers — called instances — in minutes, choose from a wide variety of hardware profiles (CPU, memory, storage, networking, accelerators), pay only for what they consume, and place those instances inside isolated virtual networks (Amazon VPC).
Amazon EC2 forms the foundational compute layer of AWS. Higher-level services such as Amazon ECS, Amazon EKS, AWS Lambda, AWS Fargate, AWS Elastic Beanstalk, AWS Batch, Amazon EMR, and Amazon SageMaker all ultimately run on Amazon EC2 capacity beneath the abstraction layer.
Amazon EC2 also achieves high security through identity and network controls (AWS Identity and Access Management, security groups, network ACLs), encryption at rest and in transit (AWS KMS, AWS Nitro Enclaves), and the AWS Nitro System hardware foundation.
Amazon EC2 Use Cases
Amazon EC2 is utilized in a vast variety of scenarios and is designed to host any workload that runs on a Linux, Windows, or macOS server.The main use cases of Amazon EC2 include the following:
- Web and Application Hosting
Run web servers, application servers, and middleware tiers behind Elastic Load Balancing with Amazon EC2 Auto Scaling for elastic capacity. - Enterprise Applications
Lift-and-shift commercial software such as SAP HANA, Microsoft SQL Server, Oracle Database, and SAS onto Amazon EC2 with dedicated tenancy and BYOL licensing. - High Performance Computing (HPC)
Run tightly coupled scientific simulations using the Hpc family of instances, Elastic Fabric Adapter (EFA), and Cluster Placement Groups. - Machine Learning Training and Inference
Train large models on P-, Trn-, and UltraCluster capacity, and serve inference on Inf- and G-series instances. - Game Hosting and Streaming
Run multiplayer game servers with low latency using AWS Local Zones and dedicated Amazon EC2 fleets. - macOS Build Pipelines
Run iOS, iPadOS, watchOS, tvOS, and macOS build/test workloads on Amazon EC2 Mac instances. - Hybrid and Edge Workloads
Extend Amazon EC2 to on-premises locations using AWS Outposts, AWS Local Zones, and AWS Wavelength Zones.
Specific Examples of Use Cases
For instance, there are the following specific examples of use cases.- Auto Scaling Web Tier
A web application uses an Amazon EC2 Auto Scaling group of t3.medium or m6g.large instances behind an Application Load Balancer. Amazon CloudWatch metrics trigger scale-out events during peak traffic and scale-in events at night. - HPC Cluster for Genomics
A genomics pipeline launches a cluster of Hpc7a instances connected by EFA, runs alignment and variant-calling jobs in a few hours, and tears the cluster down to save cost. - ML Training on AWS Trainium
A deep-learning team reserves an Amazon EC2 Trn2 UltraServer through Capacity Blocks for ML, trains a foundation model for two weeks, and then releases the capacity. - macOS CI/CD
An iOS application team uses Amazon EC2 Mac (mac2-m2) instances to run Xcode builds and TestFlight uploads, integrating with AWS CodePipeline.
Amazon EC2 Conceptual Diagram
From here, I will explain the main features and characteristics of Amazon EC2, but before that, I will show the following Amazon EC2 instance family tree to make it easier to imagine the overall picture of how Amazon EC2 instance families are organized.
This diagram illustrates how the Amazon EC2 instance families are organized into five main categories — General Purpose, Compute Optimized, Memory Optimized, Accelerated Computing, and Storage Optimized — and lists the representative families within each category.
The tree continues to grow as AWS introduces new processors (AWS Graviton, AWS Trainium, AWS Inferentia, Apple Silicon) and new specialized hardware (FPGA, AI accelerators).
In the following sections, we will provide more details about these features and configurations of Amazon EC2.
Amazon EC2 Instance Families
Amazon EC2 instances are organized into families that share a common profile of CPU, memory, networking, and storage characteristics. The first letter of an instance type (m, c, r, t, x, p, g, i, d, h, f, mac, hpc, trn, inf) indicates the family, and the digit that follows indicates the generation (M3 → M4 → M5 → M6 → M7 → M8). Suffixes after the generation indicate the processor architecture or optional features (e.g.g for AWS Graviton, a for AMD EPYC, i for Intel, n for enhanced networking, d for local NVMe SSD, e for extended memory).General Purpose
Balanced compute, memory, and networking for a wide range of workloads.- M family (M3, M4, M5, M6, M7, M8): Balanced for web servers, small databases, and enterprise applications. Available on Intel (i), AMD (a), and AWS Graviton (g) variants.
- T family (T2, T3, T4g): Burstable performance using a CPU credit model — ideal for workloads with average CPU utilization below the baseline most of the time.
- A1 (Arm): The original AWS Graviton-based instance family. Largely succeeded by Graviton2 and later M/C/R variants.
- Mac: macOS instances built on Apple Mac mini hardware (mac1 with Intel, mac2 with Apple M1, mac2-m2 with Apple M2, mac2-m2pro with Apple M2 Pro).
Compute Optimized
Highest compute-to-memory ratio for compute-bound workloads.- C family (C3, C4, C5, C6, C7, C8): For batch processing, media transcoding, high-traffic web servers, and scientific modeling. Available on Intel (i), AMD (a), and AWS Graviton (g) variants.
- Hpc family (Hpc6a, Hpc7g, Hpc7a): Purpose-built for tightly coupled HPC workloads with optimized core-to-memory ratios and 200/400 Gbps EFA networking.
Memory Optimized
Highest memory-to-compute ratio for in-memory databases, analytics, and SAP workloads.- R family (R3, R4, R5, R6, R7, R8g): General memory-optimized.
- X family (X1, X1e, X2gd, X2idn, X2iedn, X2iezn, X8g) and the High-Memory u-series (u-3tb1, u-6tb1, u-9tb1, u-12tb1, u-18tb1, u-24tb1, and the newer U7i / U7in / U7inh families on 4th-gen Intel Xeon Scalable Sapphire Rapids with up to 32 TiB of memory): Extreme memory for SAP HANA, in-memory databases, and SAS Grid.
- z1d: High-frequency processors (up to 4.0 GHz sustained) for electronic design automation (EDA) and certain financial workloads.
Accelerated Computing
Instances with hardware accelerators (GPUs, FPGAs, AWS Trainium, AWS Inferentia).- P family (P2, P3, P4, P4d, P5): NVIDIA GPUs (K80 → V100 → A100 → H100) for ML training and HPC.
- G family (G2, G3, G4dn/G4ad, G5, G5g, G6, G6e): NVIDIA / AMD GPUs (K520 → M60 → T4 / Radeon Pro V520 → A10G → T4G → L4 → L40S) for graphics workloads, ML inference, and remote workstations.
- F family (F1, F2): Xilinx UltraScale+ FPGAs for genomics, financial analytics, and video encoding.
- Inf family (Inf1, Inf2): AWS Inferentia / Inferentia2 accelerators for ML inference at scale.
- Trn family (Trn1, Trn1n, Trn2, Trn2 UltraServer, Trn3 UltraServer): AWS Trainium / Trainium2 / Trainium3 accelerators for ML training of foundation models.
Storage Optimized
Local high-IOPS or high-throughput storage attached to the host.- I family (I2, I3, I3en, I4i, I7ie, I8g): NVMe SSD-backed instances for transactional databases and NoSQL workloads.
- D family (D2, D3, D3en) and H1: HDD-backed dense storage for data warehousing, Hadoop, and big-data processing.
- Im4gn / Is4gen: Graviton2-based dense NVMe SSD storage.
Amazon EC2 Purchasing Options
Amazon EC2 offers multiple purchasing models to balance flexibility against cost.On-Demand Instances
Pay per second (Linux) or per hour (Windows and commercial Linux) with no long-term commitment.Use for unpredictable workloads, dev/test, and short-term spikes.
Reserved Instances (RI)
Commit to a specific instance type / region / tenancy for one or three years in exchange for a discount versus On-Demand. Three formats: Standard RI (fixed type), Convertible RI (exchangeable for a different type), and Scheduled RI (specific recurring time windows — limited regions).AWS Savings Plans
A more flexible alternative to RIs introduced in 2019. Commit to a consistent amount of compute usage (measured in USD/hour) for one or three years.- Compute Savings Plans: Discount applies to any Amazon EC2 instance family, region, OS, or tenancy, plus AWS Fargate and AWS Lambda.
- Amazon EC2 Instance Savings Plans: Discount applies only to a specific instance family in a specific region but offers deeper discounts than Compute Savings Plans.
Spot Instances
Spare Amazon EC2 capacity available at a steep discount versus On-Demand. Spot Instances can be reclaimed by Amazon EC2 with a two-minute interruption notice. Use for fault-tolerant, stateless, batch, and CI workloads.Tools that consume Spot capacity include Amazon EC2 Fleet, Spot Fleet, Amazon EC2 Auto Scaling Groups with mixed instances policies, Amazon ECS / EKS, AWS Batch, and Amazon EMR.
Dedicated Hosts and Dedicated Instances
- Dedicated Hosts: A physical server fully dedicated to a single customer, with visibility into and control over the underlying sockets and cores. Required for some BYOL software licenses.
- Dedicated Instances: Run on hardware dedicated to a single customer at the instance level but without the per-host visibility of Dedicated Hosts.
Capacity Reservations and Capacity Blocks for ML
- On-Demand Capacity Reservations: Reserve capacity in a specific Availability Zone for any duration, with no commitment. Customers are billed at On-Demand prices whether or not the capacity is used.
- Capacity Blocks for ML: Reserve UltraCluster capacity (P5, Trn2 UltraServer, etc.) for a specific time window starting up to eight weeks in the future — designed for ML training jobs that need predictable, contiguous access to many accelerators.
Amazon Machine Images (AMIs)
An Amazon Machine Image (AMI) is the template from which Amazon EC2 instances are launched. An AMI contains a root-volume snapshot (with an operating system and pre-installed software), launch permissions, and a block-device mapping that specifies the volumes to attach when an instance launches from the AMI.AMIs are categorized by:
- Source: AWS-published (Amazon Linux 2023, Amazon Linux 2, Ubuntu, Windows Server, macOS, etc.), AWS Marketplace AMIs (third-party paid/free), and Community AMIs.
- Architecture: x86_64, arm64, or Mac.
- Virtualization type: HVM (the standard for all current-generation instances) or PV (paravirtualized, legacy).
- Root device type: EBS-backed or Instance Store-backed.
Amazon EC2 Storage
Amazon EC2 instances can use a combination of storage options.- Amazon Elastic Block Store (Amazon EBS): Network-attached block storage that persists independently of the instance. Volume types include gp3 / gp2 (general-purpose SSD), io2 / io2 Block Express / io1 (provisioned IOPS SSD), st1 (throughput-optimized HDD), and sc1 (cold HDD).
- Instance Store: Block-level storage physically attached to the host server. Provides extremely low latency and high IOPS but is ephemeral — data is lost when the instance stops, hibernates, or is terminated. Used for cache, scratch, and shuffle workloads.
- Amazon Elastic File System (Amazon EFS) and Amazon FSx: Shared network file systems (NFS, SMB, Lustre, OpenZFS, ONTAP) that can be mounted from Amazon EC2 instances.
- Amazon S3: Object storage frequently used by Amazon EC2 for backups, data lakes, and application content. See also the related AWS History and Timeline regarding Amazon S3 article.
Amazon EC2 Networking
Amazon EC2 networking is built on top of Amazon VPC.- Amazon VPC: A logically isolated virtual network where Amazon EC2 instances run. Customers control IP CIDR blocks, subnets per Availability Zone, route tables, internet gateways, NAT gateways, and security groups.
- Elastic Network Interface (ENI): A virtual NIC attached to an instance. Each instance can have multiple ENIs and multiple private/public IP addresses.
- Elastic IP (EIP): A static public IPv4 address that can be remapped between instances.
- Elastic Network Adapter (ENA): A high-performance virtual NIC (up to 200 Gbps on current instances) that uses SR-IOV.
- Elastic Fabric Adapter (EFA): A network device that provides OS-bypass networking with the Libfabric API for tightly coupled HPC and distributed ML training.
- AWS PrivateLink: Connects services across VPCs and to AWS services without traversing the public internet.
- AWS Transit Gateway: A hub that interconnects VPCs and on-premises networks at scale.
- Public IPv4 charges: Public IPv4 addresses attached to Amazon EC2 are billed per hour since 2024; customers are encouraged to adopt IPv6 where possible.
Amazon EC2 Auto Scaling
Amazon EC2 Auto Scaling automatically adjusts the number of instances in a group based on demand. Customers define an Amazon EC2 Auto Scaling group with:- A launch template (instance type, AMI, key pair, security groups, user data).
- A minimum, desired, and maximum capacity.
- One or more scaling policies (target tracking, step scaling, simple scaling, or predictive scaling).
- A health-check mechanism (Amazon EC2 status checks or ELB health checks).
The AWS Nitro System
The AWS Nitro System is the underlying platform for all current-generation Amazon EC2 instances (introduced with C5 in 2017). Nitro decomposes the legacy virtualization stack into three components:- Nitro Cards: Dedicated PCIe cards that offload VPC networking, Amazon EBS storage, instance storage, and instance management from the main CPU.
- Nitro Security Chip: A hardware root of trust that monitors the system and prevents unauthorized access to hardware resources.
- Nitro Hypervisor: A lightweight, KVM-based hypervisor that allocates only memory and CPU resources to instances — all I/O is handled by Nitro Cards.
AWS Graviton Processors
AWS Graviton is a family of 64-bit Arm-based processors designed by AWS for cloud workloads.- AWS Graviton1 (2018, A1): First-generation, custom Arm Cortex-A72 cores. Mostly used for scale-out workloads.
- AWS Graviton2 (2019, M6g/C6g/R6g): Custom Arm Neoverse N1 cores. Up to 40% better price/performance than comparable x86 instances. Wide adoption across M6g, C6g, R6g, T4g, M6gd, C6gn, X2gd, Im4gn, Is4gen, G5g, and Hpc6a families.
- AWS Graviton3 (2022, C7g): Arm Neoverse V1 cores. Up to 25% better compute, 60% better energy efficiency, and 2× floating-point performance vs. Graviton2. DDR5 memory and EFA support on selected variants.
- AWS Graviton3E (2023, Hpc7g, C7gn): Higher vector and floating-point throughput for HPC and networking-intensive workloads.
- AWS Graviton4 (2024, R8g): Up to 30% better compute, 50% more cores, and 75% more memory bandwidth than Graviton3. First applied to memory-optimized R8g, followed by C8g / M8g / X8g variants.
- AWS Graviton5 (2025, M9g preview): Next-generation Graviton announced at AWS re:Invent 2025, succeeding Graviton4. M9g is the first general-purpose family powered by Graviton5.
Amazon EC2 Mac Instances
Amazon EC2 Mac instances are macOS instances running on dedicated Apple Mac mini hardware connected to the AWS Nitro System.- mac1 (2020): Apple Mac mini with Intel Core i7 (Coffee Lake), 32 GB RAM.
- mac2 (2022): Apple Mac mini with Apple M1, 16 GB unified memory — the first Arm-based macOS instance type.
- mac2-m2 (2023): Apple Mac mini with Apple M2, 24 GB unified memory.
- mac2-m2pro (2023): Apple Mac mini with Apple M2 Pro, 32 GB unified memory.
Amazon EC2 in Hybrid and Edge Locations
Amazon EC2 extends beyond AWS Regions through several edge offerings.- AWS Outposts (rack and server form factors): AWS-managed compute and storage installed in customer data centers. Amazon EC2 instances run on the Outpost with full integration to the parent AWS Region.
- AWS Local Zones: Extensions of an AWS Region placed in major metropolitan areas (Los Angeles, Boston, Atlanta, Tokyo, Taipei, and many more) for single-digit millisecond latency to end users.
- AWS Wavelength Zones: Amazon EC2 capacity embedded inside 5G telecommunications carrier networks (Verizon, KDDI, Vodafone, SK Telecom, Bell Canada, etc.) for ultra-low latency mobile and IoT workloads.
- AWS Dedicated Local Zones: Local Zones built for a single customer or specific regulatory boundary, including disconnected and sovereign-cloud variants.
Amazon EC2 Security
Amazon EC2 security spans the network, host, and instance layers.- AWS Identity and Access Management (IAM): Controls who can launch, stop, modify, or terminate Amazon EC2 resources. Instance Profiles attach an IAM role to an instance so application code can sign AWS API calls without long-lived keys.
- Security Groups: Stateful virtual firewalls that filter inbound and outbound traffic per ENI.
- Network ACLs (NACLs): Stateless subnet-level firewalls that complement security groups.
- AWS Key Management Service (AWS KMS): Provides envelope encryption for Amazon EBS volumes, snapshots, and AMIs. See also the AWS History and Timeline regarding AWS KMS article.
- Amazon EBS Encryption by default: A regional setting that automatically encrypts all new Amazon EBS volumes and snapshots with AWS KMS keys.
- AWS Nitro Enclaves: Hardware-isolated CPU and memory partitions inside an Amazon EC2 instance for processing highly sensitive data (private keys, regulated PII, cryptographic workloads).
- Instance Metadata Service v2 (IMDSv2): A session-oriented metadata API that requires a token from PUT requests, mitigating server-side request forgery (SSRF) attacks. AWS encourages or requires IMDSv2 by default for new AMIs and accounts.
- Amazon EC2 Instance Connect and AWS Systems Manager Session Manager: Provide audited shell access to instances without exposing SSH ports. See also the AWS History and Timeline regarding AWS Systems Manager article.
Amazon EC2 Monitoring
Amazon EC2 publishes a rich set of telemetry that supports operational monitoring and security observability.- Amazon CloudWatch metrics: CPU utilization, network in/out, disk reads/writes, status checks, and instance state metrics. Detailed monitoring lowers the publication interval from five minutes to one minute. Memory and disk-utilization metrics require the Amazon CloudWatch agent.
- Amazon CloudWatch Logs and Logs Insights: Centralized log storage and ad-hoc querying.
- Amazon CloudWatch Application Insights: Detects and diagnoses problems in .NET, SAP, and SQL Server workloads.
- Amazon CloudWatch Internet Monitor and Network Monitor: Visibility into network performance across the internet and inside VPCs.
- AWS Systems Manager (Patch Manager, Inventory, Compliance, Run Command, Session Manager): Operational management of Amazon EC2 fleets at scale.
- AWS Config: Records configuration changes to Amazon EC2 resources and evaluates them against compliance rules.
- Amazon GuardDuty, Amazon Inspector, and AWS Security Hub: Threat detection, vulnerability assessment, and centralized security posture management for Amazon EC2 fleets.
Amazon EC2 Best Practices
The following best practices come from the AWS Well-Architected practical checklist and the official Amazon EC2 documentation.Instance Right-Sizing
- Use AWS Compute Optimizer recommendations to right-size Amazon EC2 instances based on observed CPU, memory, network, and disk utilization.
- Consider AWS Graviton variants when applications support Arm — they consistently provide better price/performance for compatible workloads.
Resilience
- Spread Amazon EC2 Auto Scaling groups across multiple Availability Zones.
- Use Amazon EC2 Instance Recovery (automatic recovery on hardware failure) or Amazon EC2 Auto Recovery alarms.
- Use Cluster Placement Groups for low-latency / high-throughput workloads and Spread Placement Groups for fault isolation.
Security
- Adopt IMDSv2 across all new instances and migrate legacy AMIs.
- Encrypt all Amazon EBS volumes and snapshots with AWS KMS keys, ideally with Amazon EBS Encryption by Default enabled at the account level.
- Avoid embedding long-lived AWS credentials in user data or AMIs — use Instance Profiles.
- Limit SSH and RDP exposure with AWS Systems Manager Session Manager or Amazon EC2 Instance Connect Endpoint.
Cost Optimization
- Use Spot Instances for fault-tolerant workloads.
- Combine On-Demand, Reserved Instances, and AWS Savings Plans to cover the baseline + burst pattern.
- Stop or hibernate non-production instances outside of business hours via AWS Instance Scheduler.
- Right-size storage (gp3 instead of gp2 in most cases) and delete stale snapshots and AMIs.
Performance
- Choose instance families that match the workload bottleneck (CPU-bound → C; memory-bound → R/X; disk-bound → I/D; GPU-bound → P/G; HPC → Hpc + EFA).
- Enable Enhanced Networking (ENA) and EFA where applicable.
- Use NVMe-backed instance store for shuffle / scratch I/O when persistence is not required.
References:
Tech Blog with curated related content
AWS Documentation (Amazon EC2 User Guide)
Amazon EC2 Instance Types
AWS Graviton Processor
The AWS Nitro System
Frequently Asked Questions about Amazon EC2 History
- When did Amazon EC2 launch?
- Amazon EC2 was announced as a limited public beta on August 25, 2006 with a single instance type (m1.small) in a single AWS Region (US East). It reached General Availability (GA) on October 23, 2008 with an availability SLA of 99.95% and the addition of Microsoft Windows AMIs.
- When did Amazon EC2 introduce VPC default networking?
- Amazon Virtual Private Cloud (Amazon VPC) was announced on August 25, 2009 and reached General Availability in August 2011. From December 4, 2013, new AWS accounts launched all Amazon EC2 instances inside a default Amazon VPC; accounts created on or after late 2017 cannot use the legacy EC2-Classic networking model at all. Amazon EC2-Classic itself was retired in August 2022.
- When did Amazon EC2 Spot Instances launch?
- Amazon EC2 Spot Instances were announced on December 14, 2009 with a bidding-based pricing model. The model was simplified in November 2017 so that customers specify a maximum price instead of bidding and prices change more gradually based on long-term supply and demand.
- When did Amazon EC2 Reserved Instances and AWS Savings Plans launch?
- Amazon EC2 Reserved Instances launched on March 12, 2009, allowing customers to commit for one or three years in exchange for a lower hourly rate. AWS Savings Plans launched on November 6, 2019 as a more flexible alternative that covers Amazon EC2, AWS Fargate, and AWS Lambda usage.
- When did the AWS Nitro System launch?
- The AWS Nitro System debuted publicly with the C5 instance family in November 2017 and was formally named the "AWS Nitro System" later that month at AWS re:Invent 2017. All current-generation Amazon EC2 instances are built on Nitro.
- When did AWS Graviton (Graviton1 / Graviton2 / Graviton3 / Graviton4 / Graviton5) processors launch?
-
- AWS Graviton1: Announced with the A1 instance family at AWS re:Invent 2018 (November 26, 2018).
- AWS Graviton2: Announced at AWS re:Invent 2019 (December 3, 2019) with the M6g, C6g, and R6g families.
- AWS Graviton3: Previewed at AWS re:Invent 2021 (November 30, 2021) with the C7g family, and reached General Availability on May 23, 2022.
- AWS Graviton4: Announced at AWS re:Invent 2023 (November 28, 2023) with the R8g family, and R8g became Generally Available on July 10, 2024.
- AWS Graviton5: Announced at AWS re:Invent 2025 (December 4, 2025) with the M9g family entering preview.
- When did Amazon EC2 Mac instances launch?
- Amazon EC2 Mac instances were announced on November 30, 2020 as mac1 (Intel Mac mini). Apple M1-based Mac instances (mac2) were previewed in December 2021 and reached General Availability in July 2022. Apple M2 (mac2-m2) and M2 Pro (mac2-m2pro) Mac instances were announced in 2023.
- When did Amazon EC2 introduce ENA, SR-IOV, and Elastic Fabric Adapter (EFA)?
-
- Enhanced Networking with SR-IOV: Introduced on the C3 instance family in November 2013.
- Elastic Network Adapter (ENA): Announced in November 2016 with up to 25 Gbps initially, scaling to 200 Gbps on current generations.
- Elastic Fabric Adapter (EFA): Announced in November 2018 for tightly coupled HPC and distributed ML workloads using OS-bypass networking.
- When did AWS Outposts, AWS Local Zones, and AWS Wavelength launch?
-
- AWS Outposts: Announced at AWS re:Invent 2018 (November 28, 2018) and reached General Availability in December 2019.
- AWS Local Zones: First Local Zone announced in Los Angeles at AWS re:Invent 2019 (December 3, 2019). It has since expanded to many global metropolitan areas.
- AWS Wavelength: Announced at AWS re:Invent 2019 (December 3, 2019) for ultra-low-latency 5G workloads, with the first commercial Zones launching with carriers such as Verizon (US), KDDI (Japan), Vodafone (UK), and SK Telecom (South Korea) over 2020–2021.
Summary
In this article, I created a historical timeline of Amazon EC2 and looked at the list of features and overview of Amazon EC2.Amazon EC2 (Elastic Compute Cloud), the resizable virtual server service that effectively launched the modern public cloud, was announced as a public beta in August 2006 and became Generally Available (GA) in October 2008. Over the following nearly 20 years, it has expanded along multiple axes — instance families, purchasing options, networking, storage, hypervisors (most notably the AWS Nitro System), and processor architectures (Intel, AMD, AWS Graviton 1 → 5, AWS Trainium 1 → 3, AWS Inferentia 1 → 2, Apple Silicon for macOS). It also expanded geographically through additional AWS Regions, AWS Local Zones, AWS Wavelength Zones, AWS Outposts, and AWS Dedicated Local Zones.
I would like to continue monitoring the trends of what kind of features Amazon EC2 will provide in the future.
In addition, there is also a historical timeline of all AWS services including services other than Amazon EC2, so please have a look if you are interested.
AWS History and Timeline - Almost All AWS Services List, Announcements, General Availability(GA)
References:
AWS Documentation(Amazon EC2 Documentation)
AWS Documentation(Amazon EC2 Instance Types)
AWS Documentation(Amazon EC2 Instance Purchasing Options)
AWS Documentation(Document history for the Amazon EC2 User Guide)
AWS Nitro System
AWS Graviton Processor
AWS Trainium
AWS Inferentia
AWS Compute Savings Plans
What's New with AWS?
AWS News Blog
AWS Compute Blog
References:
Tech Blog with curated related content
Written by Hidekazu Konishi