;meyer

Methods Used

Hyperscale Atlas

Date & Time

Currently all of the data in our database was collected in the timeframe of a few days in late summer, 2026.

Scoring Bands and Grades

All relative grades within a silo use a relatively straightforward scoring band. Among all entries, all entries that fall within the top 20%, we assign an "A". All other four grading bands are evenly spaced using the same 20% rule. If the number of entries in a silo is 5 or less, we assign from A to F in order. If the best two results are equal, they will both receive an "A" grade.

Balanced Basket

To make datacenters comparable, we built a basket containing only items that are available in every Azure region. The basket contains the following items.

Basket Content

  • VM E2s v4, 730 instance hours
  • VM D4s v5, 730 instance hours
  • Blob Storage, Hot, 1000 GB per month
  • Blob Storage, Cold, 5000 GB per month
  • Block Storage, Standard SSD tier, 500 GB per month
  • Block Storage, Premium SSD tier, 100 GB per month
  • Azure Files, Cool tier, 1000 GB per month
  • Azure Files, Standard tier, 1000 GB per month
  • Azure Files, Hot tier, 200 GB per month

The dollar value we list, is the aggregate price of this basket. One significant caveat is the fact, that it doesn't include storage transfer costs. The prices were taken directly from the Azure API.

Internet transfer costs were omitted, because they don't factor into the basket method. Selection based on internet transfer or transfer between related regions, is a different beast.

Carbon Intensity

Carbon intensity is taken from Electricity Maps (see here). We try to map the coordinates of the datacenter into one of the zones available via the API. The data we have in the database has a few flaws.

  • All of the data is from a single point in time
  • This raises problems as to seasonal CO2 emissions, like intra-day and intra-year changes.
  • It misses macro trends, like when a country reduces CO2 output.
  • The mapping to an energy zone fails for a couple of datacenters. This means there's no data for them.

Since we use an approximation for the datacenters coordinates, it may also fall into the wrong energy zone, if it sits close to the border between two zones.

Sweat Score

The sweat score is relatively simple. We take the coordinates of the datacenter at hand and query Open-Meteo's Historical Weather Forecast Archive API for the daily maximum temperature of the last two years. Then for each day that falls into the following bands, we assign points. Finally we tally the points.

  • 1 point for a temperature >= 35 degrees C, but < 38 degrees C
  • 2 poitns for a temperature >= 38 degrees C, but < 40 degrees C
  • 4 points for a temperature >= 40 degrees C

This scoring band results in a slightly exponential characteristic of the score.

Global Rights

The Global Rights category is the only category that is not relative in a silo. For this grade we use the ITUC Global Rights index. An ITUC "1" rating maps to an "A", while an ITUC "5" rating maps to an "F".

You can find the ITUC global rights index explained on their site.

Armed Conflict

For armed conflicts we use the UCDP GED events list (link). All events starting 2020 count towards our score. For each event we compute if it occured withing a distance of 200km to a hyperscale datacenter. If all of these conditions match, we tally the event. In the end, most datacenters don't have a high tally, so we only signal with a red rifle, if there is greater than zero conflict events.

The UCDP GED data set includes both armed conflict involving states and non-state actors. For example, you will find Vienna has a non-zero score, because of a terrorist attack.

Public Data & Code

All the data and code used for the Atlas is available on Github.

Kontakt

Mark Meyer IT Beratung und Umsetzung

Breslauer Str. 17
38678 Clausthal-Zellerfeld
m@digitalmeyerei.de