top of page
Won by turning Amazon's own internal infrastructure-scaling pain into a product years before any other tech company realized renting out raw compute and storage by the hour would become larger than Amazon's retail business itself.
1
MODEL
BUSINESS MODEL
Infrastructure Platform
model bm
HOW THEY BUILT IT
- Launched 2006, growing out of Amazon's internal need to standardize infrastructure across its own rapidly scaling e-commerce operations — engineers realized the internal APIs they'd built to provision servers could be sold externally as a product.
- Pioneered the pay-as-you-go cloud infrastructure model (compute via EC2, storage via S3) at a moment when nearly every company still bought and managed its own physical servers, a capital-intensive and inflexible model AWS eliminated entirely.
- Grew into the dominant cloud infrastructure provider globally, becoming Amazon's most profitable division and effectively subsidizing the company's famously thin retail margins for years.
- Continuously expanded from raw compute/storage into hundreds of adjacent managed services (databases, machine learning, IoT), letting customers build increasingly complex applications entirely on AWS infrastructure.
HOW TO ARCHITECT IT
1. Look inward at your own company's hardest internal infrastructure problem — if you've built genuinely reusable internal tooling to solve it at scale, that tooling may be sellable as an external product to every other company facing the same problem.
2. Move the entire industry from capital expenditure (buying servers) to operating expenditure (renting compute by the hour) at the exact moment cloud connectivity and virtualization technology made that shift technically reliable enough to trust with production workloads.
3. Continuously expand from foundational infrastructure into higher-level managed services once you own the underlying compute/storage layer, since each new managed service both deepens customer lock-in and captures more of the value chain.
DISTRIBUTION MODEL
Self-Serve Website, Enterprise Sales, Partnership Distribution
dm
HOW THEY OPERATIONALIZED
- Distributed via instant self-serve sign-up with a credit card for developers and small teams, combined with direct enterprise sales and a vast partner/consulting ecosystem (AWS Partner Network) for large migrations.
- Free-tier and credits programs for startups (AWS Activate) seeded adoption among the next generation of technology companies before they had meaningful infrastructure budgets.
HOW TO REPLICATE WHAT WORKED
What worked: recognizing that internally-built infrastructure tooling, developed to solve Amazon's own scaling pain, was itself a sellable product — a genuine 'eat your own dog food' origin story that gave AWS deep technical credibility from day one. Trap if copied blindly: building cloud infrastructure at AWS's scale requires truly massive, sustained capital expenditure in data centers — a founder without Amazon's balance sheet cannot replicate this model at the infrastructure layer, though the broader lesson (productizing internal tooling) applies far more broadly.
| PATTERNS OF THIS MODEL
PATTERNS IN PLATFORM ECOSYSTEMS BUILT ON COMMODITY PRIMITIVES:
1. START WITH THE MOST GENERIC PRIMITIVES POSSIBLE (compute, storage) BECAUSE GENERALITY MAXIMISES THE ADDRESSABLE WORKLOADS. Specialisation comes later, once usage reveals demand.
2. MOVE UP THE STACK CONTINUOUSLY INTO MANAGED SERVICES. Each higher-level service captures more value and deepens dependence than the layer below it.
3. LET CUSTOMER USAGE DICTATE THE ROADMAP. In infrastructure, observed workload patterns are a better product plan than any market research.
4. THE ECOSYSTEM AROUND YOU BECOMES A DEFENCE. Consultants, certifications and third-party tooling built on your primitives make displacement an industry-wide problem, not a vendor decision.
What companies with this model reveal
| OPPORTUNITY INTELLIGENCE
GOLDMINE 1 — CLIMB FROM RAW INFRASTRUCTURE INTO MANAGED SERVICES.
Standard: once you own compute and storage, each higher-level managed service (databases, ML, IoT) captures more of the value chain and deepens dependency. The margin migrates upward; the moat is the accumulation.
GOLDMINE 2 — LET THE ECOSYSTEM BUILD YOUR BREADTH.
Standard: partners, marketplaces and certified architects extend capability without proportional internal headcount, and their livelihoods become an incentive to keep you entrenched.
GOLDMINE 3 — MAKE CERTIFICATION THE PROFESSION'S CREDENTIAL.
Standard: engineers certified on your platform carry it to every future employer.
THE PIT — HUNDREDS OF SERVICES BECOME A COMPLEXITY TAX CUSTOMERS RESENT.
Service sprawl and opaque billing are now the most-cited enterprise complaints, and they are the wedge for simpler providers, FinOps vendors and multicloud mandates. Breadth compounds until it becomes the reason to leave.
THE SECOND PIT — THE PROFIT ENGINE OF A RETAIL PARENT ATTRACTS PERMANENT ANTITRUST ATTENTION.
MOVE WITH CAUTION — SOVEREIGN AND DATA-RESIDENCY RULES FRAGMENT A GLOBAL ARCHITECTURE.
Note: this row duplicates entry 210 in the source list; the standards above deliberately cover the ecosystem and managed-services dimension rather than repeating the origin and pay-as-you-go analysis.
Untapped Business Model / Gaps / Goldmines / Pits
Patterns & Insights
2
MARKET
mkt mt es
MARKET TYPE
Blue Ocean
WHY THEY WON
Pay-as-you-go, self-serve cloud infrastructure barely existed as a category in 2006 — companies bought and managed physical servers themselves. AWS created the category rather than displacing an existing cloud-infrastructure competitor. Transferable principle: internal infrastructure tooling built to solve your own company's scaling pain can define an entirely new market category if no external vendor yet serves that same underlying need.
ENTRY STRATEGY
Greenfield Entry
EXECUTION
AWS entered a functionally undefined category — self-serve, pay-as-you-go cloud infrastructure — building both the product and market education around the pay-as-you-go model simultaneously, since virtually no company understood this consumption model as viable for production infrastructure in 2006.
FOOTHOLD STRATEGY
fs
Beachhead Strategy
The beachhead was startups and developers who couldn't afford or didn't want to manage physical server infrastructure — a segment with acute pain (capital constraints, unpredictable scaling needs) and low switching cost since they had no legacy infrastructure to migrate away from. From there, AWS expanded upmarket into large enterprises migrating existing data centers.
GROWTH CAMPAIGN
CAMPAIGNS THAT WORKED
The 2006 EC2/S3 launch itself, timed to virtualization technology maturity; AWS Activate startup credits program, seeding adoption among the next generation of venture-backed companies; continuous expansion into hundreds of adjacent managed services (RDS, Lambda, SageMaker) deepening lock-in within existing accounts.
KEY LEARNING
If your company has built genuinely reusable internal infrastructure tooling to solve its own hardest scaling problem, consider whether that tooling itself is a sellable product to every other company facing the same problem — some of the most durable infrastructure businesses started as an internal-only solution before anyone realized its external market value.
gc
Market Context
| MARKET INTELLIGENCE
THE STANDARD: Internal tooling built to solve your own scaling pain can define a new market when no external vendor serves the same underlying need.
RULE 1 — THE FIRST CUSTOMER BEING YOURSELF DE-RISKS THE HARDEST PHASE. Product-market fit is proven internally before a single external sale.
RULE 2 — SELLING SPARE CAPACITY IS A MARGIN STORY THAT BECOMES A BUSINESS STORY. What starts as utilisation efficiency ends as the parent's profit engine.
RULE 3 — PRIMITIVE-FIRST DESIGN LETS OTHERS BUILD THE USE CASES. Shipping building blocks rather than solutions is what allows an ecosystem to form on top.
RULE 4 — CATEGORY CREATION DEMANDS TOLERANCE FOR EARLY SCEPTICISM. The market questioned why a retailer sold servers, which is precisely why the head start lasted.
MARKET TYPE: Blue Ocean (cloud infrastructure), created from internal capability.
| MARKET ENTRY PLAYBOOK
THE STANDARD: WHEN THE CONSUMPTION MODEL IS THE INNOVATION, MARKET EDUCATION IS THE PRODUCT WORK.
RULE 1 — TEACH THE BUYER A NEW UNIT OF PURCHASE.
Pay-as-you-go for production infrastructure contradicted capital-expenditure planning. The category could not exist until finance functions accepted a variable line item.
RULE 2 — PUBLISH PRICES AND CUT THEM REPEATEDLY.
Transparent, falling pricing was the credibility mechanism that made an unproven model safe to adopt at scale.
RULE 3 — DEFAULTS BECOME STANDARDS.
Whoever defines the vocabulary of an undefined category — instances, regions, buckets — sets the terms every later entrant must translate into.
How to enter
| FOOTHOLD STRATEGY PLAYBOOK
THE STANDARD: A beachhead of experimenters becomes an enterprise business only if reliability is proven publicly before the enterprise asks.
RULE 1 — LET THE FIRST SEGMENT STRESS-TEST THE PLATFORM AT LOW REPUTATIONAL RISK. Developers tolerate immaturity that regulated buyers never would.
RULE 2 — TRACK RECORD IS THE ONLY CREDENTIAL FOR MISSION-CRITICAL MIGRATION. Years of uptime, certifications and reference architectures are what unlock regulated and government workloads.
RULE 3 — BREADTH OF PRIMITIVES CREATES SWITCHING COSTS NO CONTRACT COULD. Applications built on many managed services cannot be lifted out.
RULE 4 — EARLY DOMINANCE INVITES SCRUTINY OF EGRESS AND LOCK-IN. Regulatory attention is the mature-market tax on an infrastructure position.
How to get the first strong position
MARKET PATTERNS & PLAYBOOK
3
MONEY
money rev pri
REVENUE MODEL
Usage-Based
PRICING MODEL
Usage-Based Pricing
WHY THEY WON
Pure consumption-based pricing across compute, storage, data transfer, and hundreds of managed services, billed per hour/second/GB used, with volume discounts and reserved-capacity pricing for predictable, sustained workloads.
Granular, metered pricing by exact resource consumption removes upfront capital cost entirely, targeting developers and IT buyers who evaluate cost against the capital expenditure and inflexibility of owning physical infrastructure.
TARGET AUDIENCE
CUSTOMER BUYING BEHAVIOUR
tg cb
Startups and developers (buying flexible, no-upfront-cost infrastructure to scale unpredictably); enterprises migrating data centers (buying cost and operational efficiency at scale); ML/AI teams (buying specialized compute and managed AI services).
Self-serve, instant sign-up for individual developers; committee-driven, multi-year enterprise agreements for large migrations involving IT, security, and finance stakeholders evaluating total cost of ownership against on-premise infrastructure.
| PRICING INTELLIGENCE
What makes this model effective & make customers pay
Commitment discounts convert an elastic, switchable relationship into a multi-year lock. The pay-as-you-go promise is the entry, not the endpoint.
RULE 1 — RESERVED CAPACITY AND SAVINGS PLANS TRADE FLEXIBILITY FOR PRICE, AND FLEXIBILITY IS THE MOAT YOU SOLD.
Customers accept because the discount is large. They then cannot leave for years.
RULE 2 — EGRESS PRICING IS WHERE INFRASTRUCTURE LOCK-IN ACTUALLY LIVES.
Data is cheap to store and expensive to remove. Competitors attack precisely this line.
RULE 3 — MANAGED SERVICES ARE STICKIER AND HIGHER-MARGIN THAN RAW COMPUTE.
Proprietary databases, queues and functions cannot be lifted to another provider without rewriting.
RULE 4 — ENTERPRISE DISCOUNT AGREEMENTS ARE NEGOTIATED, NOT PUBLISHED.
List pricing exists to be discounted against commitment. Assume every large customer pays something different.
A CTO is buying the option to grow without a procurement cycle, and paying for it later in migration cost. Optionality sold cheaply becomes dependency priced expensively.
PRICE & REVENUE
| Revenue Risk - The biggest threat to revenue stability
Consumption billing means every customer efficiency programme reduces your revenue without any churn event, permanently.
Committed-spend and reserved pricing trade realised rate for visibility, concentrating your best economics in your least profitable contracts.
AI-era demand concentrates revenue in a few very large counterparties whose financing decisions become your delivery schedule.
Capacity is built years ahead of revenue, turning growth into a power, land and construction problem rather than a software one.
Public parent (AMZN); backlog and segment operating margin are what to verify.
Where the model can break
4
MOTION
GROWTH EXPANSION MODEL
COMPETITIVE STRATEGY
motion ge cs
Platform Expansion
HOW THEY EXPAND
AWS expanded from foundational compute (EC2) and storage (S3) into hundreds of adjacent managed services spanning databases, serverless computing, machine learning, and IoT, sequenced to progressively capture more of a customer's total technology stack rather than remain a pure infrastructure provider.
First-Mover Advantage
HOW THEY COMPETE
AWS's dominant market position rests substantially on being years ahead of Microsoft Azure and Google Cloud in launching a genuinely reliable, self-serve cloud infrastructure product, a head start that let it build unmatched service breadth and enterprise trust before credible competitors emerged.
GROWTH ENGINE
GTM
ge n gtm
Platform Expansion, Ecosystem Expansion
Growth compounds as more managed services are added to the platform, since each new service both deepens lock-in for existing customers (harder to migrate a multi-service architecture elsewhere) and expands the addressable use cases that draw new customers to the platform in the first place. It would break down if a sufficiently portable, cloud-agnostic architecture standard reduced the switching cost of moving workloads to a competing cloud provider.
Self-serve developer sign-up combined with direct enterprise sales and a vast partner/consulting ecosystem, reinforced by startup credit programs that seed adoption among the next generation of technology companies before they have meaningful infrastructure budgets.
SUSTAINING MOATS
Switching Costs, High Customer Lock-In, Brand Power, Technology Advantage (complex enterprise scenarios)
moat
AWS's moat combines massive economies of scale in data center infrastructure (letting it price below what most competitors can match) with deep architectural lock-in once a customer has built complex applications across dozens of interdependent AWS-specific managed services, a combination that makes both cost and migration effort prohibitive for most customers to leave.
| MOAT INTELLIGENCE
THE STANDARD: An enterprise standard becomes self-reinforcing when the skills market makes it the safe choice for every buyer.
RULE 1 — THE TALENT POOL IS PART OF THE MOAT. When most infrastructure engineers know your platform, choosing an alternative means a harder hiring problem — a cost that appears in no vendor comparison.
RULE 2 — CERTIFICATION AND PARTNER ECOSYSTEMS ARE DISTRIBUTION YOU DO NOT PAY FOR. Consultancies recommend the platform their staff are trained on, because their utilisation model depends on it.
RULE 3 — MULTI-CLOUD IS A NEGOTIATING POSTURE MORE OFTEN THAN AN ARCHITECTURE, which is why the primary provider's share of workload rarely matches the stated strategy.
THE SIGNAL: model providers and AI-native infrastructure are the first entrants in a decade to win workloads on capability rather than price. Where the scarce resource is compute access rather than service breadth, incumbency counts for less than it used to.
Why this company remains defensible
ARR & TAKEAWAY
ARR Journey - what to do at each stage
PRE-$1M — DUPLICATE ENTRY, DIFFERENT LESSON: EXTERNALISE THE INTERNAL PLATFORM
This name appears twice in the dataset. The first row covers category creation; this one covers the model of turning internal infrastructure into a product.
The precondition is that the internal platform was built with real service boundaries and APIs, not as shared code. Most companies cannot externalise because their internals were never productised.
$1–5M — CHARGE YOUR OWN TEAMS FIRST
Internal metering and chargeback is what proves the unit economics before an external customer exists.
$5–10M — PUBLISH THE PRICE LIST AND NEVER NEGOTIATE EARLY
Transparent, uniform pricing scales without a sales organisation and sets category expectations.
$10–50M — COMMITTED-USE DISCOUNTS TRADE PRICE FOR PREDICTABILITY
Reserved and savings plans lock in multi-year revenue and give customers a reason to consolidate spend with you.
$50–100M — EGRESS AND DATA GRAVITY ARE THE REAL LOCK-IN
Data is cheap to put in and expensive to move out. That asymmetry is the commercial moat — and increasingly a regulatory target.
$100M+ — CAPACITY COMMITMENTS REPLACE SALES AS THE CONSTRAINT
AI demand has shifted the business toward forward capacity contracts and enormous capital expenditure, concentrating revenue in fewer counterparties.
Rule: you can only sell your internal platform if you built it as a product for yourself first.
COPY PLAYBOOK : What Worked → What Failed → What to Replicate → What to Avoid
THE STANDARD: Internal tooling built to solve your own scaling pain can be a sellable product — and the credibility of having run it yourself is the strongest technical proof there is. [Same company as row 210; different lesson.]
SEQUENCE:
1. Identify the infrastructure you built for yourself that others would pay for.
2. Lead with the fact that you run your own business on it.
3. Sell to developers first; procurement follows adoption.
WORKED: A genuine dogfooding origin conferring deep technical credibility from day one, in a category where engineers distrust vendors.
CAUTION:
1. THE INFRASTRUCTURE-LAYER VERSION OF THIS REQUIRES MASSIVE SUSTAINED CAPEX no ordinary founder can match. The transferable lesson is productising internal tooling, not competing at the infrastructure layer.
2. MOST INTERNAL TOOLS ARE INTERNAL FOR GOOD REASONS — assumptions baked in for one company rarely survive contact with a second.
bottom of page