Amazon EC2 turns 20: from one m1.small instance type in one region to more than 1,200 across 39
Two decades of EC2 milestones double as a map of the concepts, from hypervisors to custom silicon, that underpin every cloud platform.
Amazon Web Services marked the twentieth anniversary of Amazon Elastic Compute Cloud on 25 August 2026 with a retrospective on the AWS News Blog written by Channy Yun. EC2 launched in beta on 25 August 2006, announced by Jeff Barr, offering resizable Linux virtual servers billed by the hour. At launch there was a single instance type, m1.small, and a single region, US East.
Two decades later, the service offers more than 1,200 instance types across 39 AWS regions, and has extended beyond regions into AWS Outposts, Local Zones and Wavelength locations. The post traces the milestones that shaped the platform: Amazon Elastic Block Store in 2008; Elastic Load Balancing, Auto Scaling, CloudWatch and Virtual Private Cloud in 2009; the AWS Nitro System in 2017; and Graviton processors in 2018.
The period from 2019 to 2026 is summarised as a run of silicon and form-factor additions, including Inferentia chips, Mac instances, Trainium accelerators, Capacity Blocks for ML, Graviton5 and the Nitro Isolation Engine. The post highlights Graviton5 as offering 192 cores, a cache five times larger than its predecessor and up to 33 percent lower inter-core latency. It also notes that Trn3 UltraServers can interconnect up to 144 Trainium3 chips.
The retrospective frames these developments as AWS continuing to push what cloud computing can deliver, particularly through custom silicon for general-purpose and AI workloads and through deployment models that customers in 2006 could not have anticipated. Taken together, the post presents the twenty years as a progression from a single virtual server type to a catalogue spanning general-purpose, Arm-based, Apple and AI-accelerator instances.
Why it matters
The original EC2 idea, renting a server by the hour with no upfront commitment, reshaped how software is built and paid for, and the anniversary shows how far the abstraction has travelled. The most striking shift is that a cloud provider now designs its own CPUs, AI accelerators and isolation hardware rather than assembling commodity parts. That vertical integration is becoming the competitive axis among hyperscalers. For learners, the milestone list doubles as a map of the concepts that underpin every modern cloud platform.
The original EC2 idea, renting a server by the hour with no upfront commitment, reshaped how software is built and paid for, and the anniversary shows how far the abstraction has travelled.
2006: EC2 beta
Resizable Linux virtual servers billed by the hour; one instance type (m1.small) in one region (US East)
2008: Elastic Block Store
Persistent network-attached storage decoupled from the life of an instance
2009: ELB, Auto Scaling, CloudWatch, VPC
Load balancing, metric-driven scaling and private networking arrive together
2017-2018: Nitro and Graviton
Offload hardware for networking, storage and security, then AWS-designed Arm CPUs
2019-2026: AI silicon and new form factors
Inferentia, Trainium, Mac instances, Capacity Blocks, Graviton5, Nitro Isolation Engine
What you can learn from this
- Virtualisation is the foundation. A virtual machine is a software-defined computer that shares physical hardware with others, with a hypervisor enforcing isolation between them. EC2's original promise was that you could request one in minutes and give it back when done, which only works because the hypervisor can carve up and reclaim resources on demand. Every later service, from containers to serverless functions, sits on top of this capability. The hypervisor also meters usage, which is what makes hourly billing possible. Understanding it explains both the flexibility and the noisy-neighbour concerns of shared infrastructure.
- Offloading to dedicated hardware. The Nitro System moved networking, storage and security functions off the main processor onto dedicated cards, leaving nearly all of the host CPU for the customer's workload. This matters because a traditional hypervisor consumes a share of every server just to manage it, and that overhead grows with I/O. Putting those functions in purpose-built hardware also shrinks the software attack surface on the host. The Nitro Isolation Engine mentioned in the post continues that direction.
- Separating compute from storage. Early EC2 instances lost their local disks when they stopped, which is why Elastic Block Store in 2008 was such an important addition. EBS provides network-attached volumes that persist independently of any instance and can be snapshotted or moved. Snapshots also give you a point-in-time copy that can be restored in another availability zone. Once storage is decoupled, compute becomes disposable: you can replace a failed server without losing data. This separation is a recurring pattern in cloud architecture.
- Elasticity and the autoscaling loop. Hourly billing only becomes economically powerful when capacity can follow demand automatically. Auto Scaling watches a metric from CloudWatch, such as CPU utilisation or request count, and adds or removes instances behind a load balancer to hold that metric near a target. The loop works because instances are interchangeable and stateless, which pushes state into databases and object storage. This is the mechanism behind paying for what you use.
- Custom silicon changes price-performance. Graviton processors are Arm-based CPUs designed by AWS, and the post's figures for Graviton5, including 192 cores and lower inter-core latency, reflect a focus on many efficient cores rather than a few fast ones. Server workloads that handle many parallel requests often benefit more from core count and cache than from peak single-thread speed. Owning the design lets a provider tune the chip for its own data centres and price it accordingly.
We teach this
How to use this in practice
- Launch, snapshot and terminate an instance from the command line. Using the AWS CLI with a free-tier-eligible account, run
aws ec2 run-instanceswith a small instance type, attach an EBS volume, write a file to it, create a snapshot withaws ec2 create-snapshot, then terminate the instance. Record the time from the run command to the instance reporting as running; that latency is the elasticity EC2 introduced. Done looks like a snapshot ID in your account and zero running instances, verified withaws ec2 describe-instances. - Compare two CPU architectures. Launch one small x86 instance and one Graviton instance, run
lscpuand a short single-threaded and multi-threaded benchmark of your choice on each, and record the results next to the on-demand hourly price for each. Check that any software you rely on has an Arm build before you assume the cheaper option is usable. Done looks like a four-row table showing performance per dollar for both architectures. - Draw the autoscaling loop. Diagram a load balancer, an auto scaling group, a CloudWatch metric and the scaling policy that connects them, with arrows showing the direction of data and control. Add a note on where application state lives so that instances remain replaceable. Then mark the point at which a scale-in event would terminate an instance and what must already be saved elsewhere by then. Done looks like a one-page diagram you could hand to someone new to the cloud.
- Audit one security group. Pick any security group in your account and list its inbound rules with
aws ec2 describe-security-groups. Flag any rule that allows port 22 or 3389 from 0.0.0.0/0 and restrict it to your own IP address. Repeat for any rule that opens a database port such as 3306 or 5432 to the internet, which should almost never be necessary. If you find one, note who created it and when, using CloudTrail if it is enabled. Done looks like a security group with no administrative or database ports open to the whole internet.
Sources
- Happy 20th Birthday, Amazon EC2 — AWS News Blog
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report