AWS for Thnkrr: A Beginner's Guide

10 October 2026. A reference for anyone at Thnkrr. No AWS or technical knowledge needed.

What this document is for

Thnkrr is a professional network for high-school students, a bit like LinkedIn for teenagers. Students build a profile, connect with classmates, post about what they've done, and get advice from mentors (university students). Teachers and school staff use it too. We're starting with a pilot: a trial with a few US schools, beginning with about 20 students and growing to about 1,000.

Thnkrr will run on Amazon Web Services (AWS), Amazon's business of renting out computers, storage and software over the internet. AWS sells over 200 products, with short names like EC2 (Elastic Compute Cloud), S3 (Simple Storage Service) and RDS (Relational Database Service) that mean little to most people.

This document explains, in plain words:

1. What each job behind Thnkrr is, and which AWS product will do it.
2. What other AWS products could do the same job, what they cost, and when we'd use one.
3. What other companies do, from small startups to LinkedIn, and what we can learn from them.

Nothing is running on AWS yet. The app is built and tested on our own laptops. This document describes the launch plan: the setup we'll create on AWS before the first students join.

Contents

1. The one-minute version
2. Words you'll see
3. What AWS is, in two minutes
4. How Thnkrr will run on AWS
5. How a social network grows
6. Why we skip most of AWS
7. The jobs, one by one (20 short sections, plus products Amazon is closing)
8. Lessons from other companies
9. The whole list of AWS products
10. Sources

1. The one-minute version

• The launch setup is small and simple on purpose: one rented computer runs Thnkrr, one database stores everything, and Amazon's file storage holds photos and videos.
• It will cost about $66 a month at the start and about $103 a month at 1,000 students. Amazon has given us about $5,000 of free credit through its startup programme: enough for about 6 years at the starting cost, or about 4 years at 1,000 students, if it doesn't expire first (checking its expiry date is one of the launch steps).
• Most of AWS's 200+ products are for much bigger companies. Section 6 explains why we skip them, and each job section says at what size one would start to make sense.
• The biggest companies started like this too. Instagram once ran on a handful of rented computers with three engineers. LinkedIn started as a single program.
• Safety comes first at every size: a person on our safety team reviews every photo and video before anyone else sees it. An AI (artificial intelligence) model checks all writing first and passes anything it isn't sure about to a person.

2. Words you'll see

"Copy" means different things, so this document always says which:

• a second running copy of the app: if one breaks or is being updated, the other keeps working;
• a standby database: a live twin of the database that takes over if the main one fails;
• a backup: a saved snapshot for going back in time.

3. What AWS is, in two minutes

Think of AWS as renting instead of buying. A company used to buy its own computers, put them in a cold room, and hire people to fix them. With AWS, you rent exactly what you need, by the hour, and Amazon owns and repairs the machines.

Three things make it useful for us:

• You pay only for what you use. Thnkrr's whole setup will cost about as much a month as a phone plan. If we grow, we rent more; if we shrink, we rent less.
• Amazon does the boring, risky work. Replacing broken disks, keeping copies of data in more than one data centre, installing security fixes.
• It can grow with us. The same company that will run our one small computer also runs Netflix and Airbnb. We won't outgrow it.

About the credits. Amazon's startup programme gave us about $5,000 of AWS credit. It has an expiry date we still need to check. It's separate from AWS's "free tier", a small extra allowance (up to $200) that every new account gets.

About the names. Most AWS products have short names like those above. The first time each appears, this document gives its full name and says what it does, so you never need to remember the codes.

Why Virginia? Every region has outages now and then (Lesson 3). We chose Amazon's Virginia region because it's closest to our first schools around Pittsburgh, has every product we need, and checks pictures fastest.

4. How Thnkrr will run on AWS

Here's what will happen when a student posts a photo, and which AWS product does each step.

1. The student types thnkrr.com. Route 53, Amazon's internet address book, tells their phone which computer to talk to.
2. The pages load quickly from CloudFront, a network of computers around the world that keep copies of our pages close to every user.
3. The student signs in. The app, running on one rented computer (EC2), checks their password against our database (RDS).
4. They upload a photo. It goes into a private, locked folder in Amazon's file storage (S3), where only our safety team can see it.
5. The photo and its caption are checked by machines:
   • Amazon's picture check (Rekognition) looks for things like nudity or violence;
   • an AI model (run by Bedrock) reads the caption for bullying, phone numbers, self-harm and similar;
   • it will also be compared against official lists of known child-abuse images, once the organisations that keep them approve us (section 7.6).
6. A person on our safety team reviews it. Nothing goes live until they approve it.
7. Once approved, classmates can see it, and they get an email from Amazon's email service (SES, Simple Email Service).

In the background:

• CloudTrail records every action anyone takes in our AWS account;
• GuardDuty watches for signs of a break-in;
• WAF (Web Application Firewall) blocks known attack patterns;
• AWS Backup keeps locked copies of the database;
• CloudWatch sends an alarm to a phone if anything breaks.

What it will cost each month:

Two of these are estimates. GuardDuty will cost between about $5 and $13; its first month is free and shows the real figure. Storage grows every month because nothing is deleted yet.

5. How a social network grows

The right setup depends on size. What's sensible for 1,000 students would be reckless for 10 million, and what's needed for 10 million would waste money at 1,000.

In the job sections, "Growing" and "Large" refer to these sizes.

6. Why we skip most of AWS

Almost every AWS product we don't use falls into one of three groups:

1. It has a minimum price even when nobody uses it. Amazon's recommendation engine costs at least about $389 a month, a dedicated search engine at least $350, and one networking product (NAT Gateway, short for Network Address Translation Gateway, section 7.17) about $33 before it does anything. The first two each cost several times our whole setup.
2. It solves a problem we don't have yet. A standby database, a cache or a second copy of the app all matter once thousands of people use Thnkrr at once. With 20 to 1,000 students, one computer and one database are plenty.
3. Every extra product is one more thing to set up, watch and pay for. Each is a new way for something to break. Big companies have teams for this; we have a few people.

When we reach a size where a product makes sense, its job section says so.

7. The jobs, one by one

Each section explains one job behind Thnkrr: what it is, what we'll use, the other AWS options and their prices, and, where useful, what bigger companies do.

Reading the tables: "Price" is what Amazon charges. Where Amazon charges per use (per GB, per minute, per million), the column also gives what that would cost us, where we know.

7.1 The computer that runs the app

What this is. The app has to run all the time: answering every tap and click, running the safety team's review list, converting videos. It needs a computer that never switches off.

What we'll use. One rented computer from EC2 (Amazon's computer rental). It has 2 processors (the parts that do the work) and 4 GB of memory, about as powerful as a good laptop. It costs about $31 a month, including its hard drive and a fixed internet address that never changes. It uses Amazon's own chip (called Graviton), which costs less for the same work.

Other AWS options:

Outside AWS: DigitalOcean rents the same size of computer for about $24 a month.

How bigger companies do it. LinkedIn began as one program, like ours. At about 100 million members it became unstable, so in 2011 the company stopped building new features for two months and split the program into many smaller ones. Splitting only makes sense with many engineers, so each small program has an owner.

7.2 The database

What this is. The database is the app's memory: every account, post, message, review decision and safety case. It must never lose anything, and must find any record in a fraction of a second.

What we'll use. RDS, Amazon's managed database service, running Postgres (the most popular free database software). A small one costs about $14 a month. Amazon handles repairs, daily backups and small updates. Big version upgrades, roughly once a year, are ours to schedule. Because Postgres is free software, we could move to another company without rewriting anything.

Other AWS options:

Watch out: if we skip a big upgrade for too long, Amazon automatically charges an "old version" fee of about $146 a month on our database, ten times its normal price. Upgrading on time avoids it.

Outside AWS: Supabase (from $25 a month) also runs Postgres.

How bigger companies do it. Instagram, Pinterest, Figma and Notion all ran on one ordinary database for years. When it finally got too big, they split their data across many machines, for example one group of customers per machine. LinkedIn eventually built its own database, called Espresso. Lesson 2 has more.

7.3 Speeding things up (the cache)

What this is. Some things are asked for thousands of times: a popular profile, the number of unread messages. Instead of asking the database every time, apps keep a copy in very fast memory: a cache. It's like keeping your most-used things on your desk instead of in a cupboard.

What we'll use. Nothing separate. The database keeps popular data in its own memory, and that's enough at our size.

Other AWS options:

How bigger companies do it. Facebook's cache answers billions of requests a second. The hard part isn't speed; it's keeping copies correct when the original changes. Almost every growing social network adds a cache before anything else.

7.4 Storing photos and videos

What this is. Every uploaded photo and video must be stored safely: private until a reviewer approves it, and kept for years.

What we'll use. S3, Amazon's file storage. It keeps every file on many disks in several data centres, so files are practically never lost. Under $1 a month at the start, about $14 at 1,000 students after a year, because nothing is deleted yet.

S3 can also lock a file so nobody, not even us, can delete it before a set date. We'll use that for records we must keep when a safety case is open, for example for the police.

Cheaper shelves for older files. Like moving old boxes from the living room to the attic, S3 can move files nobody opens any more to cheaper storage:

We'll start moving older files once the founder decides how long we keep things.

Outside AWS: Backblaze B2 is one of the cheapest storage companies; we plan to keep a second copy of our backups there.

How bigger companies do it. Facebook built its own photo storage in 2010 to hold 260 billion photos. Dropbox moved its files off Amazon onto its own machines and reported saving about $75 million over two years. Both only make sense at enormous size; everyone else rents.

7.5 Our web address, and getting pages to phones quickly

What this is. Two related jobs:

• thnkrr.com must point to the right computers, with the padlock that shows a secure connection;
• a phone in Pittsburgh shouldn't wait for a computer in Virginia every time it loads a photo, so a delivery network keeps copies of our pages, photos and videos in hundreds of places worldwide.

What we'll use.

• Route 53, Amazon's internet address book, for thnkrr.com: about $1 a month.
• ACM (AWS Certificate Manager) for the padlock certificate that proves the site is really ours: free.
• CloudFront, Amazon's delivery network: the first 1 TB a month is free, which covers us. Our website's own files sit in S3 and go out through CloudFront.

CloudFront can also be told to forget a file instantly. We'll use that when a photo is taken down. Our safety rules give us 2 hours to remove some images, such as private photos shared without permission.

Other AWS options:

Outside AWS: Cloudflare, a popular rival delivery network, is free to start.

How bigger companies do it. Netflix built its own delivery network, placing its machines inside internet providers' buildings, because sending video is its biggest cost.

7.6 Checking content and keeping students safe

What this is. Every post, photo and video must be checked for bullying, sexual content, contact details, self-harm and child abuse. Machines help; a person decides.

What we'll use:

Who decides:

• Photos and videos: a person reviews every one before anyone else sees it. The picture check only puts worrying ones at the top of the list; it never approves anything.
• Writing: the AI lets through writing it's confident is fine and holds back writing it's confident is harmful. Anything it isn't sure about goes to a person, and so does anything about self-harm, so a person can respond with care.
• If the AI is ever unavailable, all writing waits for a person. Nothing goes out unchecked.

Our firm rule on AI: the AI model runs inside our own Amazon account, and Amazon is set not to save anything it reads. Students' writing is never sent to an outside AI company.

Known child-abuse material is checked a different way. Child-safety organisations keep lists of "fingerprints" of known abuse images and videos, and each upload is compared against them. This catches only material already on a list, so human review still matters. Each organisation only shares its list with platforms it has approved, so the founder has to apply to each one. Our app is ready to connect to the lists; which ones we connect is decided before launch.

Other AWS options:

Outside AWS: WebPurify supplies human reviewers at 2¢ an image. Thorn sells tools that spot new abuse material, at about $30,000 a year.

How bigger companies do it. Meta works with over 15,000 human reviewers, while its AI removes most harmful posts before anyone reports them. Roblox built an AI that spots signs of grooming in chats, and in 2025 gave it away free to other companies. Small platforms have people decide and machines sort the list; huge platforms have machines decide obvious cases and people handle the hard ones.

7.7 Video

What this is. Phones record video in many formats. Each uploaded video is converted so it plays everywhere. The reviewer gets a cover image and a sheet of still frames, and the speech is turned into text so it can be checked like a post.

What we'll use. Two free programs running on our rented computer: ffmpeg converts the video and Whisper writes down the speech. Nothing extra to pay. A person watches every video before it goes live.

Other AWS options:

Outside AWS: Mux handles video from upload to playback, but keeps the videos on its side, which makes leaving harder.

How bigger companies do it. Video is the most expensive thing any network handles. Large companies convert videos on separate computers, so a burst of uploads never slows down the rest of the app.

7.8 Email, texts and notifications

What this is. Sending consent emails to parents, sign-in codes, alerts to our staff, and "someone messaged you" notices.

What we'll use.

• SES, Amazon's email service, for emails to students, parents and teachers: about 50¢ a month at 1,000 students.
• SNS (Simple Notification Service), which sends one message to many places at once, for our staff alarms by email and text: nearly free.

Other AWS options:

How bigger companies do it. At large size, notifications become their own system: it decides what's worth sending, bundles messages into daily summaries, and respects quiet hours. The challenge becomes not annoying people, rather than cost.

7.9 Signing in

What this is. Letting students, mentors and teachers sign in safely, and letting our own staff sign in to AWS.

What we'll use.

• For students, mentors and teachers: our own sign-in, already built into the app. It knows our special rules: parental consent, school email checks, suspended accounts.
• For our staff: each person gets their own AWS login with a code from their phone (two-step login), through IAM Identity Center (IAM: Identity and Access Management), which is free.

Other AWS options:

How bigger companies do it. Sign-in becomes a whole security team: spotting stolen passwords, logins from unusual places, and fake accounts.

7.10 Search

What this is. Finding a person by name, even misspelt, filtering by school and interests, and never showing anyone the searcher isn't allowed to see.

What we'll use. Search built into our database. We tested it with a million made-up profiles: names come back in a few thousandths of a second. Free.

Other AWS options:

How bigger companies do it. LinkedIn built its own search engine in 2014. In 2025 the grocery app Instacart went the other way: it moved its search off a separate engine and back into its database, because keeping two copies of the data in step caused problems. Databases can do more than people think.

7.11 Background work

What this is. Some work is too slow to make a student wait: converting a video, sending 100 emails, nightly clean-ups. The app puts these jobs on a queue, a to-do list that small background programs (workers) work through.

What we'll use. A free queue that lives inside our database. One less thing to run.

Other AWS options:

How bigger companies do it. Around 2010, LinkedIn found every system was building its own pipe to every other system, a growing tangle. So it built Kafka: one central stream of events that every system writes to and reads from. Kafka is now used almost everywhere; Netflix and Uber each pass about a trillion events a day through it.

7.12 Suggestions and "people you may know"

What this is. Suggesting classmates to connect with, mentors to follow, and posts worth seeing.

What we'll use. Simple rules in the database: same school, shared interests, friends of friends. Free.

Other AWS options:

How bigger companies do it. LinkedIn built its own database for "people you may know", holding about 270 billion connections. After moving to it, LinkedIn could serve 150 times more suggestions a second. The lesson for us: store connections neatly from day one, so they're easy to move later.

7.13 Reports and analytics

What this is. Answering questions like "how many students posted this week, by school?" without slowing down the app, and keeping old records searchable for legal requests.

What we'll use. Simple counts from the database.

Other AWS options:

How bigger companies do it. LinkedIn built its own reporting database, which powers features like counting who viewed your profile across a billion members.

7.14 Knowing when something breaks

What this is. When something goes wrong we need to know fast, know what broke, and have someone woken up if it's serious.

What we'll use.

• CloudWatch keeps the app's logs (a diary of everything it does) and sends alarms to a phone and an email.
• A Route 53 health check visits the site every 30 seconds from outside our computer and raises the alarm if it doesn't answer.

CloudWatch costs about $1 to $3 a month, and the health check $1.50. Both are Amazon products, so we also plan a free second check run by a company other than Amazon. That way we'd still hear about it if Amazon itself had problems.

Other AWS options:

How bigger companies do it. In 2017, an Amazon outage stopped Amazon's own status page from updating, because the page was stored on the very thing that broke. In 2021, Slack's dashboards went down at the same moment as Slack. Your alarm system must not depend on the thing it watches.

7.15 Security

What this is. Blocking attacks, spotting break-ins early, and protecting what we know about children.

What we'll use:

GuardDuty spots things like our passwords being used from another country, our computer contacting known criminal addresses, or someone switching off CloudTrail's records. It only warns; a person has to act.

Other AWS options:

How bigger companies do it. Large companies run a security team around the clock. The first steps are the same as ours: two-step logins, no shared passwords, a record of every action, and alarms that reach a person.

7.16 Accounts, permissions and passwords

What this is. Deciding who can do what in our AWS account, and keeping passwords and keys safe.

What we'll use:

• The account owner's login (the one created when the account was opened, which can do anything) is locked away with two-step login and never used day to day.
• Each person gets their own, more limited login through IAM Identity Center (free).
• One account-wide setting stops Amazon using our data to improve its AI products, set through AWS Organizations (free).
• Our computer gets a built-in permission (Amazon calls it a "role"), so no Amazon password ever has to be stored on it.
• Passwords the app needs, like the database password, are kept in a protected file on our computer, with a copy in Parameter Store, Amazon's free settings store, so a replacement computer can fetch them.
• Engineers reach the computer through Session Manager: typing commands in a browser window, with Amazon checking who they are. The computer has no other way in, and every session is recorded. Free.

Other AWS options:

7.17 The private network

What this is. Our computer and database sit on a private network, so only the right things can talk to each other. The database accepts only our app. The app accepts visitors only through CloudFront.

What we'll use. Amazon's standard private network (VPC, Virtual Private Cloud), free.

Two costs to watch out for:

• NAT Gateway, a common surprise on small AWS bills. Some setups hide their computers completely from the internet (no "public address" anyone can reach them on). NAT Gateway lets those hidden computers still reach out to the internet. It costs about $33 a month before any use. Our computer is reachable only through CloudFront and a firewall, so we don't need it.
• Data leaving Amazon for the internet costs 9¢ per GB, except through CloudFront's free amount. That's why everything goes out through CloudFront.

How bigger companies do it. When Facebook bought Instagram, the two companies' private networks used clashing addresses, and Instagram had to rebuild its network before it could move. Small choices early can matter years later.

7.18 Backups

What this is. Never losing data: undoing mistakes, surviving a broken database, and even surviving losing our whole AWS account.

What we'll use:

• Daily database backups kept for 35 days, which can rewind the database to any minute: free at our size.
• An extra backup before every change to how the database is organised.
• A monthly backup kept for 7 years in a locked vault nobody can empty early, through AWS Backup: a few dollars a month.
• A practice restore every three months, to prove the backups work.
• A copy at a different company (Backblaze B2), so losing our Amazon account wouldn't lose our data.

Other AWS options:

How bigger companies do it. Netflix runs a tool called Chaos Monkey that switches off its own computers during working hours, to prove the site survives. The rule everyone follows: a backup you've never restored is not a backup.

7.19 Building and releasing new versions

What this is. Testing every change automatically, then putting the new version live without breaking anything.

What we'll use.

• GitHub, where our code lives, runs the automatic tests on every change.
• A short set of commands (a "deploy script") puts each new version on our computer without taking the site down.
• CloudFormation, free, sets up parts of our AWS account from written instructions instead of clicks, so they can be rebuilt exactly.

Other AWS options: Amazon's own build tools (CodeBuild, CodePipeline) make sense mainly for teams that do everything inside AWS. Three of Amazon's older code tools have closed to new customers since 2024.

How bigger companies do it. LinkedIn tries every new version on a small slice of real users for 30 minutes, and compares it with the old one before everyone gets it. This is called a canary, after the birds miners took underground as an early warning.

7.20 Keeping the bill under control

What this is. Never getting a surprise bill.

What we'll use. Budgets emails us when the month's spending passes $80, and again at $95. Before we reach 1,000 students, we'll raise these to $120 and $140, so they don't go off every month. Cost Anomaly Detection warns about unusual spending. Both are free.

Paying less later: promising to rent for 1 or 3 years (Savings Plans) cuts costs by 30 to 60%. It isn't worth it while free credit pays the bill.

How bigger companies do it. Big companies sign multi-year deals. Pinterest has committed $4 billion to AWS through 2031. Snap committed $2 billion to Google and $1 billion to AWS. The discount comes from the promise.

7.21 Products Amazon is closing

So nobody plans around them:

8. Lessons from other companies

Real stories, mostly from the companies' own engineering blogs and outage reports, grouped by what they teach. Some are told in more detail in the job sections above.

Lesson 1: Keep it simple until a real problem forces a change.

• Segment (2018): split its product into 50+ small programs, then found three engineers spent most of their time keeping them running. It merged them back into one.
• Amazon's Prime Video team (2023): rebuilt one internal monitoring tool, made from many small pieces that Amazon charged for at every single step, as one simple program, and cut its cost by over 90%.
• Instagram and LinkedIn (sections 7.1 and 7.2): both grew huge before splitting anything.

For us: one computer and one program is the right choice at launch, not a shortcut.

Lesson 2: One database lasts a long time, and when it doesn't, change it slowly.

• Figma (2024): its database workload grew 100 times in four years before it had to split it. The split took nine months, with no disruption for customers.
• Notion (2021 and 2023): split its database by customer workspace, then split it again two years later, both times without taking the site down.
• Discord (2015 to 2022): its storage for chat messages broke at 100 million messages, then again at trillions. Each time it copied, compared, then switched.

For us: if we ever split, the natural unit is the school.

Lesson 3: Outages happen, even at Amazon.

• Amazon's Virginia region, where we'll be, had major outages in 2011, 2017, 2021, 2023 and 2025.
• In 2017, an engineer's typo took Amazon's file storage down for hours.
• In October 2025, a hidden fault in one of Amazon's internal systems caused problems for about 14 hours.

For us: keep a backup at a second company, keep a second alarm check outside Amazon, and accept that a few hours of outage a year is possible at our size.

Lesson 4: Practise for failure.

• Netflix (section 7.18) switches off its own computers on purpose.
• GitHub (2018): an automatic switch-over that had never been properly tested turned a 43-second network problem into 24 hours of disruption.

For us: the practice restore every three months.

Lesson 5: Child safety is shared work, and on by default.

• Microsoft's PhotoDNA (2009) and free tools from Meta (2019 and 2022) let any platform check for known abuse material without building it themselves.
• Discord (2023) switched on safety features for teenagers by default.
• Yubo (2022), a social app for teenagers, checked every user's age.

For us: use the shared tools, have a person look at every photo, video and unclear case, and make safety the default.

Lesson 6: Big moves take years, even for giants.

• LinkedIn planned in 2019 to move onto Microsoft's cloud (Microsoft owns LinkedIn), and paused the move in 2022.
• 37signals, the makers of Basecamp, left AWS for their own computers and expect to save over $10 million in five years. But they had a steady business and engineers to look after the machines.

For us: use tools we could leave if we had to (free database software, standard file storage), and don't plan a move we don't need.

9. The whole list of AWS products

Every AWS product in this document. When: Launch (part of the launch plan), Later (when we grow), No (not expected).

10. Sources

All AWS prices are for Amazon's US East (N. Virginia) region, read from Amazon's own pricing pages on 9 and 10 October 2026. More detail is in the production and supplier research notes in the same folder.

AWS prices

• EC2: https://aws.amazon.com/ec2/pricing/on-demand/
• Lightsail: https://aws.amazon.com/lightsail/pricing/
• Fargate: https://aws.amazon.com/fargate/pricing/
• Lambda: https://aws.amazon.com/lambda/pricing/
• EKS: https://aws.amazon.com/eks/pricing/
• Savings Plans: https://aws.amazon.com/savingsplans/compute-pricing/
• RDS for PostgreSQL: https://aws.amazon.com/rds/postgresql/pricing/
• RDS "old version" fee: https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/extended-support.html
• Aurora: https://aws.amazon.com/rds/aurora/pricing/
• DynamoDB: https://aws.amazon.com/dynamodb/pricing/on-demand/
• Neptune: https://aws.amazon.com/neptune/pricing/
• ElastiCache: https://aws.amazon.com/elasticache/pricing/
• MemoryDB: https://aws.amazon.com/memorydb/pricing/
• S3: https://aws.amazon.com/s3/pricing/
• Route 53: https://aws.amazon.com/route53/pricing/
• ACM: https://aws.amazon.com/certificate-manager/pricing/
• CloudFront: https://aws.amazon.com/cloudfront/pricing/
• Amplify: https://aws.amazon.com/amplify/pricing/
• Bedrock and Guardrails: https://aws.amazon.com/bedrock/pricing/
• Rekognition: https://aws.amazon.com/rekognition/pricing/
• Comprehend: https://aws.amazon.com/comprehend/pricing/
• Textract: https://aws.amazon.com/textract/pricing/
• MediaConvert: https://aws.amazon.com/mediaconvert/pricing/
• Transcribe: https://aws.amazon.com/transcribe/pricing/
• IVS: https://aws.amazon.com/ivs/pricing/
• SES: https://aws.amazon.com/ses/pricing/
• SNS: https://aws.amazon.com/sns/pricing/
• End User Messaging: https://aws.amazon.com/end-user-messaging/pricing/
• Cognito: https://aws.amazon.com/cognito/pricing/
• OpenSearch: https://aws.amazon.com/opensearch-service/pricing/
• SQS: https://aws.amazon.com/sqs/pricing/
• EventBridge: https://aws.amazon.com/eventbridge/pricing/
• Step Functions: https://aws.amazon.com/step-functions/pricing/
• MSK: https://aws.amazon.com/msk/pricing/
• Personalize: https://aws.amazon.com/personalize/pricing/
• SageMaker: https://aws.amazon.com/sagemaker/ai/pricing/
• Athena: https://aws.amazon.com/athena/pricing/
• QuickSight: https://aws.amazon.com/quicksight/pricing/
• Redshift: https://aws.amazon.com/redshift/pricing/
• CloudWatch (including Synthetics): https://aws.amazon.com/cloudwatch/pricing/
• X-Ray: https://aws.amazon.com/xray/pricing/
• WAF: https://aws.amazon.com/waf/pricing/
• Shield: https://aws.amazon.com/shield/pricing/
• GuardDuty: https://aws.amazon.com/guardduty/pricing/
• CloudTrail: https://aws.amazon.com/cloudtrail/pricing/
• Inspector: https://aws.amazon.com/inspector/pricing/
• Security Hub: https://aws.amazon.com/security-hub/pricing/
• Config: https://aws.amazon.com/config/pricing/
• Macie: https://aws.amazon.com/macie/pricing/
• Secrets Manager: https://aws.amazon.com/secrets-manager/pricing/
• KMS: https://aws.amazon.com/kms/pricing/
• Parameter Store and Session Manager: https://aws.amazon.com/systems-manager/pricing/
• Control Tower: https://aws.amazon.com/controltower/pricing/
• NAT Gateway and data transfer: https://aws.amazon.com/vpc/pricing/
• AWS Backup: https://aws.amazon.com/backup/pricing/
• Elastic Disaster Recovery: https://aws.amazon.com/disaster-recovery/pricing/
• CloudFormation: https://aws.amazon.com/cloudformation/pricing/
• CodeBuild: https://aws.amazon.com/codebuild/pricing/
• Budgets: https://aws.amazon.com/aws-cost-management/aws-budgets/pricing/
• AWS free tier for new accounts: https://aws.amazon.com/free/
• App Runner closure: https://aws.amazon.com/apprunner/
• Pinpoint closure: https://docs.aws.amazon.com/pinpoint/latest/userguide/migrate.html
• Kendra closure: https://docs.aws.amazon.com/kendra/latest/dg/kendra-availability-change.html

Other companies' prices

• DigitalOcean: https://www.digitalocean.com/pricing
• Supabase: https://supabase.com/pricing
• Backblaze B2: https://www.backblaze.com/cloud-storage/pricing
• Cloudflare: https://www.cloudflare.com/plans/network-cdn/
• Mux: https://www.mux.com/pricing
• Google's child-safety tools: https://protectingchildren.google/tools-for-partners/
• Meta's free matching tools: https://github.com/facebook/ThreatExchange

Company stories

• LinkedIn's history: https://engineering.linkedin.com/architecture/brief-history-scaling-linkedin
• LinkedIn and Kafka: https://engineering.linkedin.com/distributed-systems/log-what-every-software-engineer-should-know-about-real-time-datas-unifying
• LinkedIn's database, Espresso: https://engineering.linkedin.com/blog/2017/08/migrating-from-oracle-to-espresso
• LinkedIn's search engine: https://engineering.linkedin.com/search/did-you-mean-galene
• LinkedIn's "people you may know": https://www.linkedin.com/blog/engineering/graph-systems/how-liquid-connects-everything-so-our-members-can-do-anything
• LinkedIn's traffic tests: https://engineering.linkedin.com/blog/2017/05/trafficshift--load-testing-at-scale
• LinkedIn's paused cloud move: press coverage of CNBC's report of 14 December 2023
• Instagram: https://instagram-engineering.com/what-powers-instagram-hundreds-of-instances-dozens-of-technologies-adf2e22da2ad
• Instagram's move to Facebook: https://instagram-engineering.com/migrating-from-aws-to-fb-86b16f6766e2
• Figma: https://www.figma.com/blog/how-figmas-databases-team-lived-to-tell-the-scale/
• Notion: https://www.notion.com/blog/sharding-postgres-at-notion
• Discord: https://discord.com/blog/how-discord-stores-trillions-of-messages
• Instacart: https://tech.instacart.com/how-instacart-built-a-modern-search-infrastructure-on-postgres-c528fa601d54
• Segment: https://www.twilio.com/en-us/blog/developers/best-practices/goodbye-microservices
• Prime Video (news coverage; the original post was removed): https://www.infoq.com/news/2023/05/prime-ec2-ecs-saves-costs/
• GitHub's 2018 outage: https://github.blog/2018-10-30-oct21-post-incident-analysis/
• Amazon's outage reports: https://aws.amazon.com/message/41926/ (2017) and https://aws.amazon.com/message/101925/ (2025)
• Slack's 2021 outage: https://slack.engineering/slacks-outage-on-january-4th-2021/
• Netflix's Chaos Monkey: https://techblog.netflix.com/2011/07/netflix-simian-army.html
• Facebook's cache: https://www.usenix.org/node/172909
• Facebook's photo storage: https://www.usenix.org/conference/osdi10/finding-needle-haystack-facebooks-photo-storage
• Dropbox's savings: Dropbox's stock market filing (Form S-1, February 2018)
• 37signals: https://world.hey.com/dhh/our-cloud-exit-savings-will-now-top-ten-million-over-five-years-c7d9b5bd
• Meta's free matching tool: https://about.fb.com/news/2022/12/meta-launches-new-content-moderation-tool/
• Roblox's grooming detection: https://corp.roblox.com/newsroom/2025/08/open-sourcing-roblox-sentinel-preemptive-risk-detection
• Meta's review teams: https://transparency.meta.com/enforcement/detecting-violations/
• Discord's teen safety: https://discord.com/safety/safer-place-for-teens