Amazon's system design

0/3
☰
Amazon · system design

Amazon's store is hundreds of services, each owning its own data: every page is assembled from 150+ calls, and the shopping cart sits on a database designed never to reject an 'add to cart'.

+ The whole system in one paragraph

Amazon's store isn't one program with one database. It's hundreds of services, each owning its own data, and every page is assembled from many of them. The shopping cart sits on Dynamo, a database built to never refuse an 'add to cart', even while machines fail. Placing the order is the opposite: careful, and done exactly once.

192M
requests a second to DynamoDB, AWS's database service, at the peak of Prime Day, Amazon's yearly sale (Jun 2026)[7]
59T+
DynamoDB requests over the four days of Prime Day 2026[7]
150+
services one page request typically called (2007)[1]

In this lesson

3 chapters · ~7 min · a quick check in each

Everything is on this page — just scroll down ↓

Chapter 1 of 3

The whole system

1
Requirements

What it must do — and how well

The cart must accept every 'add' even while machines fail; an order must happen exactly once. 192M is the Prime Day peak of DynamoDB, inspired by Dynamo but a different design.

ⓘ Details & sources

Yellow = must do. Blue = how well. SLA = the response time a service promises; 99.9th percentile = the time 999 of every 1,000 requests beat.[1][7][14]

Build every pagefrom 150+ services,fast.Add to cart workseven while disks,networks or datacenters fail.Take each orderexactly once.Charge the card onlywhen the orderships.Every service insideits SLA at the99.9th percentile,not the average.Lose a whole datacenter: no dataoutage.Prime Day 2026: 192Mrequests a second.Never lose data.
2
Architecture

The store: a page made of services, a cart that never says no

Top: every page is assembled from many small services, each owning its own data — search, catalog, recommendations, cart. Bottom: placing an order, from checkout to the warehouse.

ⓘ Details & sources

Top: the page and the services behind it. Bottom: the order. Dashed = our model. Click any piece; play a flow.[1][6][12][14][17]

Watch it happen:
Services, each owning its data112223buy4reads the cart45566indexedYoubrowser, appPage renderingbuilds the pageAggregatorsstateless, cachedCatalogitems, offers, pricesRecommendationssimilar items, offlineCart servicenever says noDynamoN=3 · R=2 · W=2Checkoutplace your ordermodelWalletcards on fileFraud checkBuyer FraudOrder queueasync, retriedmodelWarehousesfulfillment centersSearchquery → ranked list
clientserviceworkerdatabasequeue dashed border = simplified model
123456
Step 1 of 6

You open a product page. A page-rendering service builds it, asking aggregator services for the pieces.

Curious? Go deeper+ 3 optional
4
Pipeline

'Customers who bought this also bought': built offline, read online

No comparing you with millions of shoppers on each visit: which items sell together is worked out ahead of time, and the page only looks up your few items. Stages 1–4 run offline.

ⓘ Details & sources

Item-to-item collaborative filtering (recommending from what other customers did), from the 2003 paper by Linden, Smith and York — the expensive table is built ahead of time.[13][20]

Purchase histories29M+ customers (2003)1offlineCount co-purchasesitem by item2pairsScore each pairbought together more than chance?3scoresSimilar-items tablethe expensive part, done ahead4page loadsYour itemscart, purchases, ratings5table lookupLook up + combinemost related first6listOn the pagecart, homepage, email7

7 steps. Follow the flow one step at a time — what happens first, and what comes next.

5
Architecture

One product, many sellers: which offer you get

More than 60% of sales come from independent sellers (2025). Many can offer the same product; Amazon features the offer you're most likely to buy.

ⓘ Details & sources

Amazon's own retail business and independent sellers list offers on the same product page, and one is featured. Sellers can also keep their stock in Amazon's warehouses (FBA).[29][30][26][27]

Watch it happen:
Offers on one product2223details3you buy4stores its stock4Amazon retailits own offerSeller Aships it themselvesSeller Bstock in Amazon's FCs (FBA)Catalog itemone product, one pageFeatured offermost likely to be boughtProduct pagefeatured + more choicesAmazon fulfillmentpicks, packs, ships
clientserviceworkerdatabaseexternal dashed border = simplified model
1234
Step 1 of 4

Many sellers can offer the same product, each with its own price, shipping fee, delivery promise and returns policy. More than 60% of the sales in Amazon's store come from independent sellers (2025).

Quick check

DynamoDB served 59 trillion+ requests over the four days of Prime Day 2026. Roughly how many is that per day?

Pick one — or just keep scrolling, it's not graded.

Keep going — it's free

Create your free account to unlock the rest of Amazon's system design, keep your progress and pick up right where you left off.

  1. 2Cart & Dynamo
  2. 3Orders & delivery
Create my free account →I already have one — log in
Questions & comments

Ask anything about this system, or share how you'd design it.

Log in to join the conversation