If it's expensive to move data out of AWS, it's not just about making it hard for customers to leave AWS. It means that any third-party service that wants to sell to AWS customers must also use AWS.
For example, Snowflake can't really run its own data centers. They need to rent from AWS. If you're a company on AWS, you won't want to spend $0.08/GB sending things to Snowflake.
The high egress cost gives Amazon a lot of power to keep the third-party ecosystem within AWS and without as much negotiating power on things like EC2 costs. Snowflake can't say to Amazon, "if you don't give us a great discount, we're going to move our operations to DigitalOcean" because Amazon would just say, "no you're not. Your customers aren't going to pay $80/TB to load things into Snowflake when you're charging them $23-40/TB for storage. They'd be paying 2-3.5x your storage costs just to load the data into your system!" Yes, I'm sure that Snowflake does get a discount from AWS, but having less negotiating power can make discounts smaller.
There are other reasons to want a third party to use AWS if you're on AWS. Still, the egress pricing seems to make it very hard for third-party tech providers not to use AWS if their customers are using it.
Thank you for this. I’m surprised that CloudFlare didn’t mention it in an otherwise thorough post. After all, the reason they’re trying to pressure AWS to change is that they’d like to offer services to AWS customers without bandwidth markup. Essentially competing with existing AWS services on an even footing.
This to me makes it unlikely that AWS will change their bandwidth pricing. If it was just about this revenue source disappearing, they might bite the bullet and hope that the lower prices would attract more customers in the long term. But bandwidth pricing is the moat around their castle. If it didn’t exist, Snowflake and others would leave. But more than that, new AWS services benefit from the captive audience choosing them by default. Without the moat, each service has to compete on its merits. AWS Lambda, for example would need to be better than CloudFlare Workers or fly.io containers. It won’t win customers like it does today just because it’s “free” bandwidth.
Removing these fees is a risk AWS is currently choosing not to take. And they won’t change their pricing regardless of the number of blog posts that competitors write. The only thing that would force a change is an exodus of customers leaving AWS citing this as a reason.
And I always thought they will drop the price of their EC2 some day to balance things out. But nope.
I am wondering if Amazon could introduce a bandwidth cost rebate, where the bandwidth cost would be subtracted from your final bill at a maximum of 30% of your total VM bill.
But it is like Apple, once you have a monopoly like market they just dont need to. Amazon is buying as many TSMC capacity as they could as demand for Graviton 2 is way above their projected expectation. It seems they are still growing at such a pace despite their size.
The rebate you mention sort of already exists, just restricted to public sector research customers. Think the cap is that egress is waived as long as it is less than 15% of the bill.
Amazon is really good at leveraging its semi monopolistic presence in different markets. Once you are big enough you only need to tweak a few parameters to dominate the market.
It's interesting how AWS uses egress to this effect. Your comment is very insightful.
They employ teams of economists to model price elasticity of demand. There are likely many orders of magnitude more than a few parameters. And they look at how a change in one product feature or pricing model cannibalizes or compliments other features and prices.
I've always referred to this as a roach motel model.
Most proprietary vendors do various tactics under this to retain customers via lockin, versus actually good products.
Software file formats serve the same for proprietary software.
"Cloud" vendors provide easy API services that have no analog to on-prem, so that migrations are painful. And the Great Egress is a painful problem with AWS.
Roach motel indeed. They check in, and they don't check out.
> "Cloud" vendors provide easy API services that have no analog to on-prem, so that migrations are painful.
This is true. But just as true as "cloud vendors provide easy API services which provide value over on-prem options that companies pay for". Whether you see it as positive or negative depends on how much value you get from the service.
AWS, may be, subsidizes the costs of its loss-leaders by (what seems like) charging disproportionately for egress. I reckon, as Cloudflare gears up to take on incumbent cloud leaders, it modus operandi will be to commoditize its competitor's advantages.
Apart from CloudFront and Lambda@Edge, Cloudflare has offerings that compete with Lambda (Workers Unlimited), Web Application Firewall, AWS Elemental (Stream), and (app/network) load balancers (Argo, Spectrum, Magic Transit). Check out Cloudflare's bandwidth / usage pricing for some of those, and it isn't much different than AWS.
Cloudflare, I'd predict, will eventually arrive at a similar pricing model for its IaaS outside of the Bandwidth Alliance (which, ironically, is Hotel California with select though multiple providers at play).
I don't know for sure if they sell services at a loss, but they give away some glue infrastructure at no cost to the customer. AWS Batch, AWS Lake Formation, EC2 Auto-Scaling, Data Migration Service (free for 6 months, no-limits), AWS CodeStar, AWS BeanStalk as few examples. I believe, in almost all cases (save traffic over transits like cross-account VPCs, load-balancers, internet-gateways, NATs, etc), intra-region transfers are also free.
These costs are not only on egress traffic to the internet, but also on other ways to get data out of AWS. For example, the snowball appliance that is basically a pile of hard drives that can be shipped to AWS has free ingress, but costs a few cents per GB for egress [0].
i could see this move eventually going sour for amazon down the line, there's always that one false play that sinks a giant. it probably won't be this one or one anytime soon, but definitely seems like that strategy may eventually be ruled against. but who knows, i guess if enough people are willing to accept it then they will stay that way, i honestly was hosting on amazon for a bit, nothing major but the service for me did not justify the cost. but i am somewhat old school.
And features like AWS PrivateLink play right into this strategy. If you're a SaaS provider built on top of AWS, the egress costs heavily incentivize you and your customer to connect over PrivateLink vs over the internet.
This keeps both parties locked into the AWS ecosystem.
If it's expensive to move data out of AWS, it's not just about making it hard for customers to leave AWS. It means that any third-party service that wants to sell to AWS customers must also use AWS.
For example, Snowflake can't really run its own data centers. They need to rent from AWS. If you're a company on AWS, you won't want to spend $0.08/GB sending things to Snowflake.
The high egress cost gives Amazon a lot of power to keep the third-party ecosystem within AWS and without as much negotiating power on things like EC2 costs. Snowflake can't say to Amazon, "if you don't give us a great discount, we're going to move our operations to DigitalOcean" because Amazon would just say, "no you're not. Your customers aren't going to pay $80/TB to load things into Snowflake when you're charging them $23-40/TB for storage. They'd be paying 2-3.5x your storage costs just to load the data into your system!" Yes, I'm sure that Snowflake does get a discount from AWS, but having less negotiating power can make discounts smaller.
There are other reasons to want a third party to use AWS if you're on AWS. Still, the egress pricing seems to make it very hard for third-party tech providers not to use AWS if their customers are using it.