AWS bandwidth costs, even with their volume pricing tiers, are >10x as much as you'd pay than if you'd bought it directly. On a small enough scale, it doesn't

AWS bandwidth costs, even with their volume pricing tiers, are >10x as much as you'd pay than if you'd bought it directly. On a small enough scale, it doesn't matter so much. When you're paying millions a month, it matters a huge deal. Even on a micro scale: * Vultr, DigitalOcean are 2-4x cheaper than AWS (and they are established companies with a track record of reliability, as well as 20+ locations around the world) * Even AWS offers "Lightsail" which are significantly discounted VPSes with included bandwidth that costs more than the Lightsail price! Their ToS prevents you from using it as proxies... lol. * Want to save money on S3? Get some colo space, buy some Enterprise HDDs and set it up on RAID, and your opex is going to be 5-10% the cost of S3. Or SSDs if your access patterns require it. Set up some SMART-based email alerts and an offsite backup procedure, and it's absolutely reliable enough for the vast majority of business use cases. P.S. even S3 silently loses files... * Using CloudFront? It might be easy and integrated, but it's absolutely not going to be the cheapest or best value. It's really marked up, heavily. Spend a few hours talking to leading competitors and you'll get a quote for half your CloudFront bill, if not less; often for a better CDN. * Trying AWS Rekognition, or other ML-as-a-service products? Hire some data science intern and they can copy/paste the leading open source project from GitHub, train it on your data, and you'll quite possibly get better results than generic ML solutions... for as low as 1/100th the cost. (Real example). Yes, you gotta maintain it, but I find that people generally think maintenance is a bigger monster than it actually is. AWS has huge profit margins: where do you think it comes from? Hint: it makes it from you.