IPv4 trading by the three big Cloud providers

I wrote this as an assignment for The Economist's Data storytelling and visualisation executive education course, so it's different from my usual writing. :) But some of the data are interesting (suprising even, for me), so I thought I'd republish it here.

The data used in this article comes from Scott Seligman. I downloaded it on 2023-08-04. Thanks!

It's a known fact that IPv4 addresses were exhausted between 2011 and 2019 (depending on the continent), and since then, IPv4 address ranges are a traded commodity with no further supply - no wonder one IPv4 address' price went up from around $8 in 2014 to over $40 in 2022 (with peaks above $60).

Since IPv4 is still the de facto standard for Internet servers that need to be accessible by everybody, owning IPv4 addresses is important for cloud providers, and trading activity in them could be an indicator for their own assessment of growth in number of servers/applications hosted with them.

Which rival companies did you choose to compare? #

I decided to compare the three big Cloud providers by a somewhat technical measure: the IPv4 addresses they owned over time.

1. Amazon AWS - 32% market share by revenue
2. Microsoft Azure - 23% market share
3. Google Cloud Platform (GCP) - 10% market share

Where did you find the data for your comparison? #

Originally I thought that I could be able to download trading data from the Regional Internet Registries (RIR), but this proved problematic: they don't all seem to publish them. (ARIN does, but RIPE doesn't seem to, for example.)

Fortunately, cloud providers provide their CURRENT ip range data publicly, and I found that a software developer named Scott Seligman from Seattle, have for the past few years been scraping this data, and stored each day's state in a GitHub repository. Not the most usual of data formats (but a great idea), and nicely workable with some scripting! :)

After a quick and dirty validation (the latest address counts were aligned with ipinfo.io's current data for Google and Amazon), the data seemed trustworthy.

The data covers the last two years (From Aug 2021 to Aug 2023, as of today), this is long enough to see some interesting patterns. (Mr Seiligman actually has 8+ years of AWS data in a different git repository, but only two years for the other two (and some smaller) cloud providers.)

Was anything in the data misleading? #

Not really misleading, but it's an interesting property of this data that IP ranges are of different sizes: the biggest ones ("/8" ranges) can contain 16.777.216 addresses, while the smallest range (a "/32") has only one single IPv4 address.

Also, some of the IP range amount decreases could come either from selling, or from acquiring some last piece of a bigger range and then unifying it.

What did you learn from the data? #

The IPv4 address ownership somewhat reflects the providers market share by revenue, but not perfectly: Amazon AWS owns the most IPv4 addresses by far, Microsoft Azure comes second, and Google comes third. But while by revenue Amazon AWS is only ~50% bigger than Microsoft Azure, it owns more than twice the IPv4 addresses.

It's interesting to see that Microsoft Azure actually owns more IPv4 ranges than even Amazon AWS, but this amounts to less than half as much IPv4 addresses. Presumably as a latecomer to the Cloud Computing market, it had to build up its stock from smaller, still available ranges.

The companies manifest different trading behaviors. Google's and Microsoft's behavior is close to what I expected to see in the data: they are scooping up IPv4 addresses at a slow and steady rate, always increasing, (almost) never decreasing. On the other hand, Amazon AWS seems to be actively trading: opportunistically buying but also disposing of IPv4 ranges. (I never expected the Cloud providers to sell, too!)

Google goes weeks without adding new stock, while AWS's trader guy almost seems to be day trading. :) I can even identify a trend of him systematically divesting smaller ranges, while keeping/buying the larger ones (which is rational for a big provider): for example, AWS dumped more than 150 ranges (more than 10% of his 1379 owned at the time) in July 2023, while the number of its owned IPv4 addresses barely budged.

The three Cloud providers have different "appetites". In the past two years, they have increased their stock of IPv4 addresses by the following amounts:

1. Amazon AWS: 65.128.534 => 72.524.132 (+11.3%)
2. Microsoft Azure: 20.439.974 => 38.525.147 (+88.5%) (!)
3. Google GCP: 9.092.864 => 12.244.736 (+34.7%)

AWS clearly feels that he "owns enough", while Microsoft almost doubled its stock.

In the chart we can see that Microsoft went on a real "shopping spree" in September/October of 2022, scooping up several fairly large ranges, and didn't increase that much since, so its appetite seems to be satisfied for now.