What are the barriers to the development of a homegrown AI industry in Britain?  

What are the factors holding back data centre build?  

And how might the sector in Britain fare if we create the right conditions for growth?  

Data Centre: Demand sat down with CUDO Compute founder Matt Hawkins to discuss all of this, and more. (CUDO is a UK-founded provider of scalable, sustainable high-performance computing infrastructure designed to support AI workloads globally.) 

1.There are some considerable blockers to the development of AI infrastructure in the UK. What are the biggest challenges? 

The biggest one is access to grid connections. It is taking years to get approvals. So the question becomes, where can you find capacity? Fortunately, there are a lot of windfarms, solar plants, batteries and industrial sites, such as for steel production, where you can go in off the back of someone else’s infrastructure.  

If we can match up people generating energy with people consuming energy, we can reduce the cost of capacity in the UK, and it means the grid connections are already in place. We are already working with renewable energy partners to deliver data centres without straining the grid. We want to use as many renewable energy resources as possible.  

2.Is the model where data centres generate their own power on site likely to become more prevalent? 

Definitely. That’s good for the grid because it doesn’t create constraints, and it means other people aren’t paying for the upgrade. In you look at the US, for example, a lot of money goes into upgrading the grid to cope with data centre build. The cost goes onto everybody’s bills. But if you pair the two together [data centre and onsite generation] you are not affecting everyone else.  

3.What could grid operators do to make the connections process smoother? 

This grid has been built over a long period, and it does take time to upgrade locations. The key is large generators of power working with large consumers of power. On the power generation side, there are windfarms, solar farms, and other companies generating energy, but there is no real link or correlation with consumers. 

If we could work with government and energy companies to link the two, that will enable the UK to expand projects much faster. 

4.If Britain becomes a consumer of AI rather than a creative force, what would that look like? 

It would mean not creating local employment, not expanding local infrastructure, and not keeping data sovereign. There is also an incentive for talent to move overseas. We need to try and keep that talent in Britain.  

If the environment for infrastructure development is sustainable, on the other hand, people will learn the engineering skills and the data centre infrastructure skills. That means jobs for thousands of people.  

But it’s not easy to get people with the right skills in the UK. That is something we should be driving.  

5.How do you view the development of indigenous AI right now?  

If you look at most of the foundational models, they are being built in the US and China. There are some teams and projects trying to do the same in Europe. It is very expensive to build a foundational model. There is a huge amount we can do in the UK, but it is more likely that we have models fine-tuned to the economy and environment. 

6.Have you seen evidence of AI workloads going overseas? 

Yes, partly because of the lack of data centre capacity. We are building capacity now to try and keep up with demand, because at the moment we are having to put clusters offshore, which we don’t want to do – we want the UK to compete. But there just isn’t the capacity available, and it’s taking time to build.  

In general, you have overdemand for GPUs in the market that outstrips supply. There isn’t enough compute capacity.  

7.Where are your customers coming from when it comes to sustainable high-performance computing? 

 
Most of them come from the US and Europe. It is partly US companies looking for a European presence for their infrastructure. Pretty much any successful company in this space just keeps on building and needing additional capacity.  

Inference is directly proportional to the amount of customers you have, so anyone that is growing continuously needs more capacity. 
 
For the last data centre news, sign up to our newsletter.