Why AWS Announces Regions in Advance

Generally, AWS doesn’t pre-announce things years in advance. When they do, it’s usually a misstep—such as with Timestream’s GA being two years after its announcement.

A notable exception to this policy is geographic expansion of AWS regions.

Last week, AWS pre-announced a new region in Zurich, with an eye towards it launching in late 2022, and a second new region in Hyderabad coming in mid-2022.

Why on earth would they do that? How can it possibly take two years to build a data center—even assuming they’re dealing with Dell’s shipping times for all of their servers? What’s going on in that world?

This week, let’s delve into what drives these pre-announcements and why they’re seemingly premature.

What is a Region?

AWS’s global infrastructure page goes into some detail about what they’re building. It consists of multiple physically separate availability zones, which are themselves “one or more data centers.”

Think about that for a moment. Multiple buildings need to be acquired (or perhaps more realistically, built) close to one another but not TOO close to one another. Then, multiple redundant power lines need to be run to each of those buildings. You’ve also got to run massive amounts of fiber not just between those buildings, but also to the larger internet backbone and Amazon’s internal networks as well.

Once all of that is done, you’ve got to fill these things with (extremely conservative guess) tens of thousands of servers, provision those servers, and deploy the starting set of AWS services that are going to be supported in those regions. The cost for all of this easily climbs into the billions of dollars per region and requires massive investments in real estate, computer hardware, networking, and hiring. Their Hyderabad region apparently resembles a $2.8 billion investment.

AWS has done all of this to General Availability 24 times as of this writing–with five more on the way.

Real estate deals leak like Equifax

It takes time to buy a few square miles of industrial property. There are a lot of moving parts, and a lot of different people involved. Once the deal is signed and construction starts, it’s very hard to keep things quiet.

“Who do you work for?” asks the taxi driver. “Oh, some random small company,” replies the Amazonian festooned with company swag and still rocking their badge.

Secrets are hard.

By pre-announcing, an entire swath of this problem goes away. Journalists won’t be digging into real estate rumors in an effort to be the first to break the story, and that buys Amazon some space.

This is, of course, far from the primary reason for pre-announcements.

Customer obsession

Whenever AWS announces a region, they aren’t doing it because they think it’d be neat to have a region in a given place or because they landed a screaming deal on some real estate in the Greater Duluth area. Regional expansions are instead driven by customer demand.

When customers ask for regions near them (either for latency or regulatory reasons), it invariably means that they’re either at or fast approaching a decision point about which vendor to use. By announcing an upcoming region, AWS is saying that while there isn’t an acceptable region for those customers today, there will be in the near future. It reassures folks that their pain hasn’t been forgotten or ignored and informs longer-term enterprise planning.

AWS Regions are different

I wrote above about what’s involved in an AWS region. This isn’t what other cloud provider regions look like at all.

Given some late 2019/early 2020 capacity constraints, you could be forgiven for assuming that an Azure region is a data closet next to the boiler room in a 90-year-old building with flaky power. Azure’s region table indicates that only some of their regions have multiple availability zones; meanwhile AWS has no regions with fewer than three AZs (excluding their mainland China regions, which are themselves operated by partner companies).

GCP has regions that are similar to AWS; it’s easier for them to expand, since when they need more GCP capacity they simply turn off another beloved Google product to make room.

IBM “Cloud” is not a real cloud provider.

For most folks, having new AWS regions available is just a curiosity; but for customers in those locations it can be a game-changer.

Personally, I find the announcements fascinating because they give a glimpse into what the world of cloud might start to look like in a few years’ time. And they also give me a better idea of which companies are most likely to call me up to complain about their AWS bills, and where they’re coming from.