Desperately seeking UK data centre data

Data centres might be economically significant, so it’d be nice to know what they are and what they’re doing. There are two longstanding problems with that, at least in the UK:

1. We don’t know what they are.
2. We don’t know what they are doing.

We don’t know what they are.

We don’t know what they are doing.

These issues have forced economists to interpret UK data centre activity as reflected in other measures, such as computer part imports.

Naively, we’ve assumed this might be one of those statistical pictures that, over time, becomes clearer as official statistics evolve. However, that is apparently not the case: data centre data ambiguity is now baked into statistical policy in the UK.

The Office for National Statistics, in a comprehensive article on this topic released a couple of weeks ago, says “we are currently unable to separately identify either the investment in or output of data centres”. Instead, they offer a review of data centre count estimates (tl;dr: highly variable and of questionable provenance) and point out that the data is there, it’s just not all in one place.

We’re not usually hesitant to criticise the ONS, but having read the full article: honestly, fair enough. For something that has assumed such a major position in the current economic, political and environmental discourse, data centres really don’t make for good data.

Data centres do not have a unique standard industrial classification of economic activities under the current schema, a situation that is not set to improve this decade. As the ONS explains:

The earliest planned use is in the Blue Book 2031 (UK National Accounts). However, this will not improve the visibility of data centres within the National Accounts, as their classification will still depend on the dominant activity of the reporting unit.

The earliest planned use is in the Blue Book 2031 (UK National Accounts). However, this will not improve the visibility of data centres within the National Accounts, as their classification will still depend on the dominant activity of the reporting unit.

The fundamental problem they point to is that even once a data-centre-specific SIC does exist — 63101, capturing everything under “Provision of cloud infrastructure and platforms, including AI facilities and neocloud” — there’s no guarantee that the overall company (as a statistical reporting unit) will actually use it.

After all, many companies that do things like operate data centres (eg BT, CapitaLand and SDC Capital Partners) also do other things (eg telecoms, real estate development or private equity investment) to a greater degree.

And even if every company that offers some 63101 services was identified as a 63101 company, there would be more issues.

For instance, if a company runs a co-location data centre (where they might handle the energy supply and keep the toilet paper stocked while other companies do the expensive work of installing server racks), then capturing the actual investment into that data centre would require the service users to also be 63101s. That is possible, but also offers another way that the data picture can become partial. And that’s before we even get into things like where the chips come from and how that affects the overall growth picture.

All of which is a roundabout way of saying: you should question all data, but especially if it’s data about UK data centres.