Explain It in 30 Seconds
Before modern cloud computing, a company needing more computing capacity often had to:
forecast demand
↓
buy servers
↓
wait for delivery
↓
install hardware
↓
configure data center
↓
maintain spare capacityAWS changed the model toward:
request service
↓
receive computing resource
↓
scale up / down
↓
pay according to service modelThe first AWS services launched in 2006. [S14][S15]
By FY2025 AWS revenue had reached $128.725B. [S1]
Where It Sits in Amazon
AWS belongs to Amazon but does not sit inside the commerce transaction loop in the same way as Marketplace, Prime or FBA.
├── International
└── AWS
Amazon reports AWS as a separate operating segment. [S1]
That separation reflects materially different customers, products and economics.
The Problem It Solved
Amazon's own teams experienced how difficult it was to provision technology infrastructure. [S14]
The internal insight was that many organizations needed the same basic building blocks:
- computing;
- storage;
- databases;
- networking;
- identity;
- security;
- messaging;
- developer tooling.
If those capabilities were standardized and exposed as services, developers could build applications without owning every layer underneath them.
How It Works — Simple Version
AWS operates physical infrastructure including data centers and networking.
On top of that infrastructure, it exposes services through software interfaces.
Customers can combine services rather than buying one monolithic product.
From Retail Infrastructure to External Platform
AWS did not begin as a web-hosting feature for Amazon sellers.
The larger idea was to make technology infrastructure programmable and available to outside developers.
The launch sequence mattered.
S3
Amazon S3 launched in March 2006 as an internet-scale storage service. [S15]
EC2
Amazon EC2 followed in 2006 and gave customers on-demand computing capacity. [S14]
The combination established two basic cloud primitives:
store data
+
run computeAWS then expanded upward into higher-level services.
The AWS Stack
DATA CENTER / NETWORK
↓
COMPUTE
↓
STORAGE
↓
DATABASES
↓
DATA / ANALYTICS
↓
SECURITY / IDENTITY
↓
DEVELOPER PLATFORM
↓
AI / ML
↓
BUSINESS APPLICATIONS / CUSTOMER WORKLOADSCustomers do not have to use every layer.
The platform's power comes from combining many services under one infrastructure and commercial relationship.
Major Service Families
Compute
Virtual machines, containers, serverless computing and specialized accelerated computing.
Storage
Object, block, file and archival storage.
Databases
Relational, key-value, document, graph and other database models.
Networking
Virtual networks, load balancing, content delivery and connectivity.
Analytics
Data processing, warehouses, streaming and business intelligence.
Security
Identity, encryption, monitoring, security tooling and compliance capabilities.
AI and machine learning
Training, inference, foundation-model access, developer tools and custom silicon.
How AWS Makes Money
AWS monetization varies by service.
Customers can pay based on:
- compute usage;
- storage consumed;
- requests;
- data processed;
- database capacity;
- model inference;
- reserved capacity;
- long-term commitments;
- enterprise agreements.
The broad model is consumption and contracted infrastructure rather than a one-time software license.
Why AWS Economics Differ From Retail
Retail revenue includes the selling price of merchandise Amazon owns.
That means product cost is embedded in the business model.
AWS sells technology services built on infrastructure Amazon operates.
Once data centers and software exist, incremental service economics are structurally different from reselling physical products.
That helps explain why AWS can contribute a much larger share of operating income than of revenue.
Scale
FY2025:
AWS revenue
$128.725B
Year-over-year growth
20%Q2 2026:
AWS net sales
$42.2B
Year-over-year growth
37%
AWS operating income
$16.6BAmazon described the Q2 2026 AWS growth rate as its fastest in 18 quarters. [S2]
Why AI Changes AWS
Generative AI is unusually infrastructure intensive.
A modern AI stack can require:
chips
↓
high-speed networking
↓
large training clusters
↓
model training
↓
model hosting
↓
inference
↓
data services
↓
agent / application toolingAWS can potentially monetize several layers of that stack.
This is why Amazon is investing in:
- custom silicon;
- accelerated compute;
- model services;
- data-center capacity;
- developer tools;
- agent infrastructure.
Custom Silicon
AWS increasingly designs specialized chips to improve economics and performance for certain workloads.
This moves AWS deeper into the infrastructure stack.
The strategic logic is:
cloud provider controls software layer
+
cloud provider increasingly controls hardware design
↓
potential optimization across full stackThe exact competitiveness of individual chips should be evaluated separately from the broader strategy.
What It Depends On
Capital investment
Data centers require enormous capital.
Electricity
Compute growth, especially AI, increases power demand.
Chips
AWS depends on both internally designed and externally supplied processors and accelerators.
Network reliability
Cloud customers depend on high availability.
Security
A cloud platform must secure infrastructure used by many independent customers.
Developer ecosystem
Technical platforms become stronger when developers know the tools and partners integrate with them.
What Depends on AWS
- external customer applications;
- startups;
- enterprise systems;
- government workloads;
- many Amazon internal technologies;
- parts of Amazon's AI development;
- Amazon Robotics data and software systems.
Why It Matters to Amazon
AWS matters for three reasons.
1. Scale
It is a $100B+ annual revenue business.
2. Profit mix
It contributes a disproportionate share of operating income.
3. Strategic position
It places Amazon at the center of cloud and AI infrastructure spending by organizations that may have no retail relationship with Amazon.
Before / After
Before AWS
Amazon technology
primarily supports Amazon businessesAfter AWS
Amazon infrastructure capability
↓
external developer platform
↓
enterprise cloud
↓
AI infrastructureWhat Most People Misunderstand
“AWS hosts Amazon.com.”
That is incomplete.
AWS is an external commercial platform with its own customers and economics.
“AWS is one server product.”
No.
It is a large portfolio of infrastructure and software services.
“Cloud means Amazon does not own physical infrastructure.”
The opposite.
Cloud abstracts physical infrastructure for the customer, but AWS itself must build and operate enormous physical infrastructure.
“AI is a separate business from cloud.”
For AWS, AI increases demand for many cloud layers simultaneously.
Key Milestones
Early 2000s
Amazon teams increasingly standardize internal infrastructure concepts.
March 2006
Amazon S3 launches. [S15]
2006
EC2 launches and on-demand compute becomes a core AWS primitive. [S14]
2010s
AWS expands into a broad enterprise cloud platform.
2020s
AI and custom silicon become increasingly central to infrastructure strategy.
Q2 2026
AWS reaches $42.2B quarterly revenue with 37% year-over-year growth. [S2]