What even is an 'AI growth zone', anyway?

I figured it out so ChatGPT doesn't have to.

Britain in 2030 if Keir Starmer somehow makes it to a second term.

POD! If you’ve not been checking out the recently renamed YIMBY Pod then you’re missing out. On this week’s show, Martin explores why the new Universal theme park is having water woes, and I talk about my big day speaking at the London Assembly. Plus we speak to Eamonn Ives from The Entrepreneurs Network about how to make Britain’s immigration regime less awful for highly skilled founders – and why this is important for the economy. Listen here, or wherever you get your pods.

---

Sometimes it can be hard to separate out government buzzwords and catchphrases, from anything actually, materially happening.

For example, when Boris Johnson was in power we heard a lot about ‘levelling up’, and though most politics-watchers had a sense that the government intended it to mean something about addressing regional inequality and the north’s lack of prosperity relative to the south, it didn’t really translate to much in concrete policy terms.

The same can be said at the moment for the much talked about ‘National Data Library’, which is less a concrete plan, and more the government’s attempt at a Jedi Mind Trick to silence any questions it is asked about data policy.¹

I could go on. But what I want to talk about today is another catchphrase that is regularly used by government figures in Parliament and the media as a signal that they are taking action: ‘AI Growth Zones’.

For example, back in November, the government published a press release announcing that an ‘AI Growth Zone’ would be created in South Wales. And a few days ago, another was announced for Lanarkshire in Scotland. (Let’s just call them ‘AIGZ’ for ease.)

In total, five have been announced so far. The other three are in the north-east of England, North Wales, and Culham in Oxfordshire.

But there is an obvious question to ask. What exactly is an AIGZ in practical terms? What is actually going to happen in these places?

In all honesty, it’s pretty confusing if you just read the press releases, as they are less clear on what actions the government is taking, and are more slanted towards touting impressive-sounding numbers about economic uplift, or the number of jobs that will be created.

For example, apparently the South Wales AIGZ will be a £10bn investment, and will create 5,000 jobs. And it will apparently have something to do with a ‘Sovereign AI Unit’, which will be backed by a £500m investment, and a further £137m that is being put into AI-assisted scientific research.

But this still isn’t particularly clear. In fact, those latter programmes – the sovereign AI unit and extra cash for scientific research – though important and interesting initiatives in their own right, aren’t really anything specifically to do with AIGZs at all.

So I’ve been trying to disentangle it all. And now I think I’ve got a pretty good handle on what the policy is actually supposed to be. So let’s dig into it, and find out what an AI Growth Zone really is.

If you enjoy nerdy policy deep-dives, you’ll like my newsletter, so make sure you subscribe for free to get more of this in your inbox!

Two for one

The phrase ‘AI Growth Zone’ originally came from the government’s AI Opportunities Action Plan that was written by Matt Clifford and published just over a year ago. If you don’t remember it, it’s probably because, inexplicably, the big story out of the launch wasn’t about AI at all – but some lobby-brained nonsense about Keir Starmer declining to say if Rachel Reeves will be Chancellor at the next election.

Anyway, in his report, Clifford called for AIGZs to “facilitate the accelerated build out of AI data centres”, and he sets out how they could be an opportunity to “crowd in” private cash to pay for them. Then following this recommendation, the Department for Science, Innovation and Technology (DSIT) turned this from an idea into a policy that could actually be delivered.

So what exactly is the plan? Essentially DSIT is attempting to combine two problems Britain faces and tackle them both at the same time – while sidestepping around an awkward third problem.

The first problem is the obvious one. Britain simply needs more data centres.

This is because if we want to remain economically competitive in a post-AI world, we’re going to need our own ‘sovereign’ compute capacity, so that British companies can crunch data and train models at a similar scale to rivals in other countries.

And this could be a serious economic problem if we don’t do something about it. According to the UK Compute Roadmap, we’re going to need 6GW of data centre capacity by 2030 – three times as much as we have now.² So we need many more server racks stacked with GPUs, for basically the same reason we need well-maintained motorways or a highly skilled workforce. This is just the cost of doing business in the twenty-first century.

Some computers doing computer stuff. Photo by Taylor Vick on Unsplash

The top-level goal of the policy then is to attract tech companies to build data centres. But what about that second problem? Well, if we’re going to build data centres somewhere, why not build them so that they can help reduce regional inequality – or (as we used to say) level-up the parts of the country outside London.

This is a good intention both in its own terms, and for cynical electoral reasons for the coming battle with Reform. So the thinking is that one way to uplift these areas would be to, say, attract a bunch of tech companies to build data centres in our former industrial heartlands.

So the plan is to pair this build out with a bunch of other measures to support the local areas in the vicinity of the new data centres. For example, the government has said that it will invest £5m in each AIGZ area, to support local businesses adopt AI.

And more significantly, it has said that local authorities hosting AIGZ data centres will “retain 100% of business rate growth” for the next 25 years, which the government reckons could boost local authority budgets by £5m-10m every year. Given that seemingly almost every local authority in the country is at breaking point, that’s extra cash that will be useful.

A little scepticism

So that’s the premise of the plan, but I must admit I’m a little bit sceptical here.

I’m sure if a government minister were asked about it on Politics Live, they would carefully explain that attracting high-tech industry to specific locations could drive the growth of the local tech industry in each area, by creating high-skilled jobs and acting as a magnet to cluster tech companies together, and so on.

I’m sure this is true to an extent, but what I can’t quite get past is that the specific physical location of a data-centre doesn’t matter all that much, for most use cases, at least once you’re inside the same country.³

This is because data centres are designed to operate lean – operators want the minimum number of staff present on site, and many operations can be handled remotely. The same is true downstream of this as, er, that’s the point of a data centre.

This is where coders really want to hang out. Photo by Sandro Cenni on Unsplash

So this isn’t quite the same as clustering, say, trendy tech start-ups or big tech firms like Apple or OpenAI. For those sorts of companies, it makes far more sense to locate offices somewhere like London or Manchester, as both cities have existing pools of engineers to hire, and trendy bars to go for drinks in after work. Ping time back to the server racks simply doesn’t matter, when the difference is a few hundredths of a second.⁴

Similarly, I’m sure local authorities will appreciate the £5m for “AI adoption” and the extra cash from business rates, but realistically the uplift is relatively small. For example, this year the budget for Bridgend County Borough Council, where the South Wales AIGZ is located, is £383.3m – so the projected data centre bonus only adds an extra 1% or so. Better than nothing, for sure – but I’m not entirely sure it will be revolutionary.

So this is all to say that I think, in reality, combining the two problems is a small sleight of hand. But this isn’t to say it isn’t a useful deceit. That £5m will presumably improve some local businesses and make them more competitive.⁵ And the promise of cash will help the government win the political argument over building large data centre campuses in specific locations.

And as I said above, if we want the national economy to grow we do ultimately need these data centres for broader economic reasons. If framing it as levelling up and throwing a little cash helps win support, then that all sounds good to me.

One size doesn’t fit all

Anyway, cynicism aside, let’s get back to the plan. As the next obvious question is “how exactly is the government planning to attract new data centres to AIGZs in the first place?”. After all, simply declaring somewhere an ‘AI Growth Zone’ doesn’t mean that it will just happen.

The slightly complicated answer here is that, as I understand it, no two AIGZs will be exactly the same. Instead, DSIT is essentially acting like a salesperson on behalf of Britain, and will be negotiating bespoke agreements with investors to get AIGZs started. So you can imagine this might include all sorts of negotiated inducements, like tax breaks or promising to build infrastructure to support them, and so on.

But despite this, there will still be commonalities across all of the AIGZs. And we now know what they are, as they were published last November.

First off, there’s tweaks to the planning system. In line with broader planning reforms, the government says that, if necessary, it will fast track Nationally Significant Infrastructure Projects (you know, like data centres). And it will even use its “call-in” powers to take planning decisions away from local councils, to speed data centres to approval, if need be.

And similarly, AIGZs will also give grid queue priority to data centres. This is really important, as it can take years to hook up major projects to the national grid normally.

Then to grease the wheels even further, apparently the government will also appoint a team of data centre planning experts, who will provide data centre companies with a combination of advice and grant funding, to help get their planning applications over the line.

And this brings us neatly to perhaps the single most important consideration for running a data centre – the cost of energy. To make AIGZs particularly attractive, the plan is to offer data centres a discount on their energy bills.

How are they going to do this? Well, it’s all to do with where the AIGZs are actually going to be…

Proximity to power

Now we reach the third problem. The one that the AIGZ policy is carefully stepping around.

In principle, building out Britain’s data centre capacity is a very good idea. So why are we creating AIGZs? Why can’t we just… offer the same incentives to companies to build data centres… everywhere in Britain?

The answer is because some areas are significantly more energy constrained than others. And Britain’s historic inability to think ahead has come back to bite us again.

For example, in some places, like London, lack of grid capacity has created a frustrating zero-sum conflict, where planners have to effectively choose between housing and data centres, even though we really need both.

So this is the secret at the heart of the ‘growth zones’ – they are being created where they are not just because we want to help Tyneside or South Wales, but because these are the places in the country that are either next to a significant source of energy, or are already well connected to the grid.

For example, the first AIGZ announced was at the Culham science park in Oxfordshire – not exactly somewhere struggling economically. The reason it is here is because helpfully it is the site of JET, the now decommissioned Joint European Torus – an experimental nuclear fusion reactor that required an ungodly amount of grid energy during tests. So it already has the serious grid connectivity that a bunch of AI data centres might require.

The inside of the tokamak at JET. I went on a tour once, and barely understood a word the nuclear scientists were saying.

The other sites are similar too in their proximity to power. In the North East, an AIGZ has been announced at the site of the former Blyth Power Station, and in South Wales it is on the site of the former Ford Bridgend plant. So both are already deeply plumbed into the grid. And the planned AIGZ in North Wales, in Anglesey, is close to the planned Wylfa small modular reactor.

The only site I’m not entirely sure on is the AIGZ that was announced in Lanarkshire, just a few days ago. The press release mentions that it will be generating 500MW of on-site power with renewables, which is in theory enough power to run a data centre, but it will presumably require significant battery storage for the, er, rare occasions when the sun doesn’t shine in Scotland.⁶

And there could still more locations to come. There’s apparently a rolling application process, so I’m sure more will be announced over the next few years.⁷

That’s the plan

So there you have it. This, in a nutshell, is the government’s plan for AI Growth Zones.

And frankly, despite my cynicism about how they are being sold, they sound like a good idea – and like far more than just a buzzword. Ultimately, we need these data centres, and they need to go somewhere. Where they can be located is limited. So doing everything we can to make sure data centres gets built in those specific places makes a lot of sense.

Phew! If you read this far, then you must like my stuff. Subscribe for free to get more policy deep dives direct to your inbox!

1. Though I understand that deep within the Department of Science, Innovation and Technology, that there is a team trying to work out what it is going to be, and maybe one day they’ll tell us.

2. 2030 sounds like the distant future, but really it is now less than four years away, because we’re all really fucking old now.

3. One of the reasons we need more data centres in Britain, as opposed to other countries, is because for both security and legal reasons, it is often unwise to send data flying around the world. Would we want our health data off-shored to who-knows-where, for example?

4. This footnote is to acknowledge that there are circumstances where proximity does matter. Like, if you’re Netflix or YouTube shifting a lot of data to viewers, you want your servers relatively close as bandwidth is cheaper at enormous volumes. But if you’re training an AI model and just need to set a bunch of servers chugging away for a while? Perhaps not so much.

5. It’s not exactly clear how the money will be used for this purpose. So I’m not sure if it refers to training, or capital spending, or whatever. But hey, that’s an extra £5m sloshing around the local economy!

6. Maybe there is a great reason to build an AIGZ here, but it does smell a bit of people in government thinking “we really need to have one in Scotland to stop people complaining”.

7. Perhaps we’ll hear about plans for a Gorton and Denton AI Growth Zone in the next few weeks?