Is datacentre sovereignty really that important? - Martin Alderson

In the UK (and I'm sure elsewhere) politicians and commentators are falling over themselves to suggest that without huge fleets of datacentres built in the UK that we are going to be hopelessly left behind.

I'm not convinced this is the case, and it risks really falling into the same (mostly misguided) obsession many politicians have for heavy industry revival.

This is going to be a rare UK-centric post on my blog. Apologies for my mostly global readership; the argument may be different where you live.

Latency isn't a big deal

One of the first (and most easily dismissed) arguments I've heard is that without datacentres close to the users, the latency will be too high to use AI services. This would therefore make them too slow to use.

Clearly, this isn't the case. Nearly all AI use cases are not hugely latency sensitive. To put this in context, the time to first token (how quickly the AI responds) on Opus models is between 1.6s and 3.6s. The round trip latency introduced from the UK to the East Coast of the US is around 80ms, to Europe 10-20ms, and to Asia around 200ms. So the latency on the providers side is orders of magnitude higher than the latency for a UK based user to reach an overseas datacentre.

It is fair to say that real time voice or video applications benefit from lower latency than these typically text based use cases. But these are a tiny fraction of AI usage (at the moment) and even in that case European datacentres can provide reasonable latency for these - it doesn't have to be in the UK itself. And my personal belief is that real time audio based agents are likely to work best when they can run on device entirely (so there is 0 network latency) - so without a data centre requirement at all.

Regardless, many of these same commentators also suggest locating datacentres in the very north of Scotland (to take advantage of the excess wind power), but ironically these would have significantly worse latency for users from the densely populated south of England - Paris, Amsterdam, etc all are closer, and thus faster to respond.

"We can tax it"

The next argument that is often floated is that it becomes a tax base - in the UK business rates are applied to commercial buildings and are paid to the local authority in question.

The formulae for calculating this is in true UK tax law style overly complicated, but in essence it works on the rateable value of the building in question - what the estimated annual rent would be to rent the property - including relevant fit out. This is then multiplied by 0.508 to arrive at the annual business rate value.

To take a very rough example, my research found that buildings 5-8 in the Virtus London campus can support 100MW[1] of load. These are valued as far as I can tell at around £12m/yr of rateable value. So the local authority (London Borough of Hillingdon) gets approx £6m/yr from this in business rates.

If we scale that up to 1GW, it's fair to say that the local authority might get somewhere close to £100m/yr of business rates.

While this is not nothing - and certainly gives local authorities a valuable source of revenue - it really is a rounding error under the current system. If we moved every single datacentre under construction globally (30GW) to the UK instead, it would bring in approximately £3bn/yr, or around 0.2% of government spending.

Detractors may say that this is the current system and the tax base could be changed. But by doing that you massively reduce the attractiveness of the UK as a place to build the aforementioned datacentres. And the potential tax rates to be at all material would have to be punishingly high. This combined with the extremely high price of electricity in the UK would make it completely unfeasible to operate them in the UK.

It's a similar story with jobs. Datacentres are famously light on permanent staff - the whole point is that they're highly automated, so even a large 100MW site might employ only a few dozen people once it's running. The construction phase is more labour intensive, but temporary, and much of the capex (the chips especially) is spent overseas rather than in the UK. Even on generous assumptions the direct contribution to a ~£2.8tn economy is a rounding error.

It gives us control

The final and perhaps most plausible sounding argument is that in the event of political instability it would give us control over AI usage - which is and will be a growing national priority.

There are really two versions of this argument. The cruder one is outright seizure, which I'll come to. The more serious one is that in a global compute crunch, having the datacentres physically here means we won't be left at the back of the queue. But this doesn't survive contact either. If a hyperscaler or a frontier lab owns the racks, a datacentre in Slough serves their global demand - not ours. You can't compel a private operator to give UK users...