Welcome to the Committee’s second panel in our inquiry into AI, business and the future workforce. We know that AI’s compute capacity is growing but remains well behind the United States and China, limiting our ability to train large frontier AI models independently. As a result, many UK firms rely on overseas cloud infrastructure. To address this, the UK Government have invested through the AI opportunities action plan and the compute road map. The UK has around 1.6 GW of data centre capacity and two national supercomputers. The UK’s strengths could be said to lie in data and semiconductor design, and it has valuable assets such as the NHS and UK Biobank for data. There are strong governance frameworks and over 200 semiconductor firms including Arm, a global leader in chip design. Access to energy infrastructure has been a huge issue, with examples of grid connections, such as that of Nscale in Loughton, which was mentioned last week, with challenges with its access to the grid and connection capacity. We know that high energy prices are affecting the ability to compete globally. The Stargate UK project has recently been shelved by OpenAI, citing energy costs and the regulation around AI. Could I please ask the panel to introduce themselves briefly?

My name is Matthew Evans. I am chief operating officer at techUK. We represent over 1,200 tech companies in the UK. About 650 to 700 of those are SMEs and the rest are either mid-cap or large multinational corporates.

Good afternoon. My name is Clinton Hasell. I am chief business officer at Kao Data, a privately owned, UK-headquartered developer and operator of data centres across the UK.

I am Hugh Milward. I am vice president of corporate affairs for Microsoft here in the UK.

Could I ask the panel for your overall assessment of the UK’s AI infrastructure today? How well positioned are we at the moment to drive that growth at scale?

It depends on what you mean by AI infrastructure. There are a growing number of data centres. Some of them are small and independent. Some of them are essentially leased data centre capacity to a number of other providers. Some of them are wholly owned by hyperscale providers such as us. We have had data centre capacity here, initially leased, for around eight years. Our model has changed and we are increasingly investing in the data centre capacity ourselves and building our own, because we can drive greater levels of environmental sustainability, efficiency and innovation in our own capacity there. Does the UK have enough capacity? It depends on what we want to do with it. It really comes down to what we want to do and whether we have the capacity to do the things we want to do. In your introduction, you mentioned model training. That is a big question as to whether we want to train models here in the UK. Is there a value to training models here in the UK? We have high energy costs. We have challenges around grid connections and the timeliness of grid connections. We are planning capacity in the north of England, but we are not going to be able to switch it on until 2033, even if we could build it within the next 18 months. There are challenges associated with that. Is there a value instead of training models anywhere where the capacity is available and the price is cheap? It then becomes, “Where is the model located? Where is it inferenced? Where can it be fine tuned?” That is a different matter. It uses different types of capacity. As a country, we need to think about what we need to do and right-size the infrastructure to do it.

Do you think that is clear enough at the moment?

It is not. It is a privilege to follow the panel before, who are some of the world’s greatest minds on AI but, even there, there were different interpretations of words such as “sovereign”. What do we mean by sovereignty? Until we have a little more clarity around what we mean, how we are thinking about resilience, how we are thinking about risk, what our risk appetite looks like and what our risk is for different types of workloads, we are still going to have some challenge around defining what we want to do here in the UK, versus what we want to have anywhere. As a good example of this, Microsoft is, as you heard about earlier, primarily an orchestration layer rather than a model developer. We have around 10,000 AI services available to any given customer. Around 1,500 are physically based here in the UK, where the data processing and the model itself is located here in the UK. Is that enough? Is it not enough? It depends on the risk tolerance around concepts such as sovereign.

That is really helpful. Thank you. Clinton Hasell, what is your assessment of our AI infrastructure here in the UK?

In answer to the question, yes, we are constrained by our infrastructure at the moment. Software is moving so quickly, and infrastructure and our ability to deliver in the UK are not keeping pace. There are a number of reasons for that. Certainly, as the previous panel said, we do not lack AI talent, ambition or the demand. We have high energy costs, which are the highest in the G7. There is the constrained grid capacity where we are waiting up to 15 years now for firm grid offers, where we could develop that infrastructure but we are constrained by our grid. Our copyright laws put us at a disadvantage compared with locating particularly the generative large language model AIs elsewhere. Those three factors are the primary factors that are constraining our ability to grow. As you said, it comes down to what kind of AI platform we are seeking to attract in this country. There is generative AI, which is what is driving a lot of the growth and demand at the moment as the big tech companies train their large language models. That makes the UK very unattractive. Those kinds of modelling and platforms are pretty much location agnostic, so the large tech companies will go where the economics make sense. At the moment that is not the UK. Then there is inference AI. Over time there will be an inflection point and that demand will overtake generative AI. That has some benefit from being located in the UK because of latency and proximity to end users, but a lot of that generative AI can still sit overseas. The challenge for the UK when that happens is that we lose control and access. For small businesses and for enterprise‑level AI, that puts the UK potentially at a disadvantage in the future. We are outsourcing productivity to other nations. The question that we pose is whether the UK wants to be an AI maker or an AI taker. If we become an AI maker, we bring a lot of economic benefit back to the UK.

Just to confirm, have you been affected by the grid connection issue? You said 15 years.

Absolutely, yes.

Is it one of your sites that you have been quoted for?

It is a number of development sites. Access to grid power is probably the No. 1 constraint for a data centre developer. It is the first question we ask when we are assessing a new site. As an example, we have developed an AI campus in Harlow to serve Cambridge, the University of Cambridge and other universities. We have AI small business platforms and larger platforms there. We have been quoted 2037 for the next tranche of power from UKPN. That may come forward as part of the NESO queue review, but you cannot plan. You cannot deliver with certainty. We have the capacity, and we have the planning and land, but we do not have certainty about the power. That could come forward, and we are waiting to hear in the next three to six months as to whether that date will move forward.

To add a bit of colour to that, most operators for sites face waits of between three and eight years, echoing what Clinton said. There is also inherent uncertainty in some of those connection dates. One of our members had a site within the M25 area where the connection dates slipped by over a decade—this is hundreds of millions of pounds’ of investment. That uncertainty, without the guarantee of that connection, makes it really hard to build in the UK.

Is that a regular occurrence?

That was probably an extreme occurrence, but that level of uncertainty where you get slips for your grid connection is happens regularly within the industry and the build-outs.

Wow, okay. What is your take more broadly about where the UK is?

Grid connections, price and copyright—I will not rehash those arguments. Reflecting on the evidence that we have just heard, we might not be in a sprint, but we are certainly in a competition for capital at the moment. The AI landscape is not a zero-sum game, we strongly believe, in terms of how we deploy and adopt AI. Where we are seeing the level of capital for AI and compute infrastructure at the moment, no one knows how long that is going to last, but it is unlikely that that is a new normal. The more we can attract to the UK now, the better we will be because, as Clinton and Hugh said, that gives you optionality further up the stack. We should say that we are really good at building compute infrastructure in the UK. We have the third largest cloud market globally and the most in Europe. Part of that is because of the demand for those services from areas such as financial services, life sciences and others. The risk is that there will be demand for inferencing and agentic AI workloads and, because of the constraints from the grid, price and copyright, we will not be able to service that demand. That will have a real-world economic impact on the UK.

Clinton and Matthew, you both spoke about copyright in your contributions, talking about us being an AI taker rather than an AI maker and losing control. Surely one of the fixes that you are then suggesting is for the UK to relax some of its copyright rules. Surely that just puts us in a race to the bottom and disadvantages those people who are taking that risk and generating that stuff and that content.

That is a choice for the UK as to what you are seeking to attract to the UK. If you want the large language models, which the inference AI platforms interface with, those large language models today are being trained in the US. They are being trained in continental Europe or the Nordics, typically where energy costs are a lot lower, where it makes more economic sense and where copyright laws enable that to happen. That cannot happen to the same degree in the UK at the moment, but the inference platforms are not subject to those copyright laws. In the UK we are benefiting from the training that happens overseas through access through the inference models. It is a choice that we have to make as a nation, but it is constraining our ability to attract large investment and those large platforms coming into the UK.

We take a slightly different view, which is that a restriction on text and data mining impacts your ability to fine tune and inference a model. If you want to do both of those, essentially you need computers to be able to read data.

Running with that a little further, could you give more detail, Matthew? Obviously, there are a lot of people wanting that power, some in data centres and some for other AI reasons, but lots of other sectors want that power. First, do you think there is a case for AI or data centres going to the top of that queue? Secondly, if you had a Government who were willing to take their magic wand out on copyright, what would you like to see disappear?

I will take those in reverse order, if I may. Part of the challenge in the UK in copyright at the moment is that we do not have a position that is now agreed and adopted, and that gives clarity to industry. For us, we would want it to be globally competitive. Our members say that, at the moment, we are legally less certain than the EU, the US, Japan and Singapore.

Can you take that apart a bit for us? There is GDPR and this is post Brexit. Where are the gaps compared with the EU that make it less certain?

At the moment, we do not have a settled position. That is the key point here. What we want to see on copyright, if we had that magic wand, is a copyright regime that is globally competitive, gives us clarity and is delivered in quite short order. We need that in order to advance investment in AI infrastructure in the UK.

The EU’s position on copyright is not globally competitive in that context. Is that what you are saying to us?

Members have some challenges with the EU’s approach. I can provide you with further detail if that would be helpful.

That would be helpful. What about on grid connection queues?

On grid connection there is an opportunity and we are working quite closely with NESO and Government to reform the grid connection process. There are a lot of speculative bids for power in that queue. We have managed to filter some of those out, but there are reasonable steps that we could take so that you are pre-qualified before entering the queue—it is called the gate 2 process. There are reforms we can do that will help make that much more stable and clearer for members, and give them better certainty to be able to invest in what are really multimillion-pound infrastructure investments.

If I have an AI data centre at gate 2 and something else there, do you think there is a case that the data centre should win through in that discussion?

We have to look at what is nationally beneficial from a strategic and economic point of view. We want to work with local communities. The tech sector is really clear and we have been clear on this in areas such as planning. We do not want to override local concerns. We want to work with local communities. As the UK, we need to look at where we best deploy capital. Where do we best deploy some of this capacity for energy? That is largely a Government decision to make. There is a good case to put AI infrastructure near the top of that queue.

Putting that into a geographical context in the UK, a lot of power is increasingly generated off the North sea, so further north. I hear arguments about latency and the need for a data centre in the south-east, otherwise it is not much use to us. Is that misconstruing the case?

There is a reason that the south-east has quite a large density for cloud computing, because that is where the demand was for cloud computing infrastructure. We are seeing more demand right across the UK. That means that there is greater opportunity to deploy AI and compute infrastructure across the UK. As Hugh and Clinton have mentioned, it depends on what you are using AI for. If we were in a position where you could use AI training or would train large AI models in the UK, geography would become less of an issue and you would probably go where the energy was cheapest. That is not the case, for a number of reasons, in the UK. Inferencing has a latency dependency, so distance ultimately does matter a bit. Where you are seeing large tech clusters develop in Manchester, Edinburgh and the central belt of Scotland, you can start to think about how you deploy compute infrastructure to service those areas of demand. Ultimately, we go where the customer and demand wants.

You are absolutely right. There is surplus renewable energy in Scotland and other parts of the UK. Having a compelling reason for the platforms to locate in those geographies makes a lot of sense, but the compelling reason is probably differentiated energy costs, so that you can drive the platforms to those parts of the country where there is surplus energy. Indeed, as consumers, we all pay for curtailment of that energy, so it is a win-win.

Going back to your answer to Charlie’s question, Mr Evans, he offered you the blank cheque of what you would like on copyright laws. You very carefully sidestepped the answer of what you would like to see. You said that you would like a competitive position and certainty. Bearing in mind how important the issue of copyright laws is to many of our constituents and businesses, I wondered whether I could press you on what that actually looks like. They are great words to use, but in reality what do they mean?

There is a question of whether we are working within the confines of what the Government are putting on the table at the moment. We were quite comfortable with an opt-out TDM—text and data mining—regime. Government have made it clear that that is no longer on the table. It is on us as an industry to work with the creative sector to say, “What is that middle ground that is going to provide clarity for both sides on that argument?” I am happy to come back to you with some more specifics on that as well.

You have probably already addressed this, but we are keen to understand what policy levers you would like to see move to accelerate the growth of data centres. I know that you have covered this in great detail, but is there anything else you want to add?

There are just two things I would add. One is that Government really need to focus on the grid connection process, how we get greater speed in that process and how we get greater certainty within it. The second is on energy prices. We have the highest industrial energy prices in the G20. A lot of that is on the non-commodity costs. The levy reform that Government have started to look at is really important in terms of the levy framework. We are part of a broad coalition of industry associations that want Government to take that forward and make sure that we are not adding additional non-commodity costs on to energy bills. At the moment, to give you some stats, if you had a 100 MW data centre in the UK, the Social Market Foundation estimates that that would give you a running cost on energy of about £220 million a year compared with £156 million in France, £67 million in Sweden and just £55 million in the US. You can see how out of kilter we are on some of these energy costs.

I would echo Matt’s comments there. When you look at what a customer would pay in the UK, 50% of their costs for being in one of our data centres would be on energy, compared with 15% in the US. That gives you the ratio. I would also like to, if I may, give you a little insight into the economics of building a data centre and how much private capital is invested by the likes of Kao. To build a medium-sized, 36 MW data centre is going to cost you £500 million just for the infrastructure—not the GPUs, computers and platforms that sit inside the data centre, but just to provide the land, the power connections, the chillers and the electrical infrastructure. That is a huge investment that companies such as Kao are making, often at risk on a three-year cycle to develop a data centre, with the customer hopefully coming on board at some time in the last year or 12 to 18 months before it goes live. Having the certainty of knowing when the power is coming is really important. Having a reasonable degree of understanding of that demand and knowing that the UK is an attractive place for the large hyperscalers or other platform providers to put their infrastructure gives us confidence that we can invest and build that critical national infrastructure.

Could I ask you all what you made of Dame Wendy’s comments about the potential for smaller, more localised data centres and whether you think that that should be pursued more vigorously?

It is not an area that we are focused on. We are that medium size. There are localised data centre companies out there that are looking at data centres at the edge. I suppose that they will have the same uncertainty of demand. I cannot really comment on that part of the market.

It is interesting. It depends on what you want to do in the data centre, I suppose. If you look at what has happened with cloud, we have gone from companies having the data in their own basements or IT rooms and large companies having their own essentially small data centre. Now we have seen that grow with the economies of scale and the efficiencies to having those in actual larger data centres with multi-customers. We are building data centres now of 300 MW. One in the US is 1 GW. You get far greater economies of scale by doing that. At the same time, we are seeing another shift in the way in which technology is operating, with a move to having more processing at a localised level as well. This edge computing points to a mixed environment of more localised data processing of different kinds, particularly as you look at a mix of chips between CPUs, GPUs and new inferencing chips. You can do different workloads at different points of what is required there.

Are there any security aspects to those considerations as well?

It is an interesting question, because it is quite easy to conclude that being local or within national boundaries or borders is more secure. Actually, more secure is more secure. The way in which you define resilience and build programmes of resilience will create your resilience. Is there a security risk associated with a mixed environment? We would say, no, there is not. We are not complacent but, at the same time, I do not think that there is a company that invests more in terms of our cyber-security within our own networks and systems. That cyber-security is universal, whether your data is in the US, in the UK, on a device or in the cloud.

Matthew and Clinton, both of you have given graphic descriptions of the costs of energy. I think that Matthew said that we are the highest in the G20, not the G7.

I meant the G7.

Okay, I just wanted to clarify that. How do the costs of compute and data centre operations compare internationally? I am quite keen to hear not just about energy costs, although I think the description that Clinton gave of 50% in the UK versus 15% in the US is pretty graphic on that. We know that the cost of doing business for all sectors in the UK is going up. Can you give me a sense of where the UK sits internationally in that and what needs to be done to make us competitive internationally?

It is worth reflecting on the impact that this has. We poll 500 business leaders, split between tech sector and non-tech sector, to look at both supply and demand each year. Energy costs have been seen as a top two barrier to greater investment in the UK in both segments for the past four years, so it is having a material impact on the UK economy. We are increasingly seen as an outlier on energy prices and that has an impact in terms of compute, but also in other areas. There are many reasons why building a large semiconductor fab in the UK might be difficult, but energy costs would be among them. It is not just one part of the network. It affects telecoms networks and the whole digital stack really, so it is a big issue. As I said, we are really focused on levy reform. The Government have opened that up as a conversation. We work across a number of different industries, because this is an economy-wide issue, not just a tech sector issue, to see how we can drive that forward. Levy reform is a top priority within that.

It is not just the data centre industry that is constrained by high energy costs. It is across many industrial sectors. That is the nature of the UK at the moment. To Matt’s point, we need to address the levies, but also shift the demand or push the larger data centre campuses that are very energy hungry to the parts of the UK where there is surplus renewable energy and we can make use of that at a lower cost. That makes economic sense across the board.

I accept everything that you have just said on energy costs. I have spoken on it many times in the House of Commons and I am absolutely behind you on this. The question I am also quite keen to understand is the wider cost of doing business. In terms of taxation and hiring people, are those costs also putting off investment in the UK?

If you were looking at the other costs associated with running or developing and building a data centre, they are probably fairly comparable across the other markets. The UK competes with Germany, France, the Netherlands and so on within Europe for attracting the large platforms. It is energy that is the key differentiator. The cost of building a data centre is fairly common across those locations.

Going back to that polling piece, this was the first time that energy cost was not No. 1—it was No. 2 this time. The regulatory cost and level of taxation were seen as the No. 1 barrier across the sector as a whole, not just on data centres, referencing Clinton’s point around build costs.

Thank you for clarifying that.

The only thing I would add is that, more than energy costs, the constraining factor, as we see it, is the supply capacity. At the moment, the pipeline of energy supply is such that we as a country cannot do all of the things that we want to do. We have some trade-offs to make, and some of those trade-offs will be around data centres and what kind of workloads we are able to do. We are not able to stand up a 1 GW data centre in the UK. There are some things that you can do in some countries that we will not be able to do in the UK. That could be just fine, but it means that we need to right-size our ambitions for what we want to do with technology and data centres to the supply of energy that we have.

The second part of the question is what we should be doing about that. I do not want to put the words in your mouth, but perhaps you can tell me what you think we should be doing to fix it.

We should be investing in a wide range of new capacity, including green—certainly for us, as a company that will really only buy green energy, that is a fundamental—but also a mix, including nuclear and other types of provision.

I wanted to follow up the point that you made, Mr Evans, on energy costs. Do the sectors that you represent benefit from the British industrial competitiveness scheme, or is that not the situation for your industries?

It is not, unfortunately. There is a glorious thing called SIC codes, which I am sure members of the Committee are more than familiar with.

Many people have talked to us about SIC codes.

They are not absolutely representative of the economy as we see it today. I cannot quite remember the date that they were updated, but we are talking pre-internet days when the last update was done. Entirely reasonably, the Government have chosen to base some of the criteria for this off industry codes, and we fall outside of those. We have engaged with the Minister on this point. I am happy to provide you with some more evidence on it. That is unlikely to have an impact on the tech sector as a whole.

Would you say that your preferred policy solution for energy costs is the removal of policy costs and loading them on to general taxation?

Broadly, yes, to some extent. We have had a number of levy control frameworks that have failed because we have continued to see levies being added to the unit price of energy. We think we need to review why they have failed and then take action to start to reduce those costs.

AI rests on super-sophisticated chips, at the end of the day. There are very specialist items, such as field-programmable gate arrays, which themselves are new and novel. What do you think about the British position for accessing these kinds of things? Mr Milward, I wanted to start with you because you mentioned the chips. Do we have access to what we need in an era when some countries are restricting access to these things?

At the moment, we would say yes. There is a global shortage of chips that is under challenge from what is going on in the middle east, and access to helium and other materials that are essential for silicon. The UK Government have actually been pretty canny in their desire to buy forward certain modern new designs of chips, including specialised inferencing chips. That makes a lot of sense. One question to your previous panel was around how you might spend sovereign. A certain amount of investment in these kinds of chips for various specialised uses is a sound concept, but there is a global shortage of chips.

How should we be addressing that? Should we be trying to onshore production of these? Is that possible? I have been to Taiwan and they are very clear in Taiwan that they are streets ahead of most people. That is not easily replicable; infrastructure like that is not something you can just create overnight. We are talking about races here. Are we at the races with this, or should we just give up on it?

I do not have a simple answer to that. We have a global interdependency in a number of different parts of the tech supply chain, but other supply chains as well. Generally, we have been quite happy to date with that global interdependency. The UK, for example, is very advanced in terms of chip design. You have some countries that are very specialised in terms of chip manufacture. If every country brought into its own borders all aspects of the supply chain, you would see massively increased costs. It becomes a trade-off as to what aspects of that a country deems to be essential to invest in. The UK is very well positioned in terms of its trade relations with the kinds of countries that are strong suppliers of this kind of infrastructure. We feel that the international agreements that are in place at the moment are sufficient for the UK’s demand.

Mr Evans, do you think that we are in quite a good place?

I would echo Hugh’s comments about the global interdependency. The UK Government have been quite forward looking for a number of years in terms of signing some quite useful trade agreements. The Hiroshima accord with Japan, for instance, focused around semiconductors, as well as other chips. We have seen some similar trade agreements. We have to recognise that we are very good in one part of that stack. We are really advanced in terms of chip design, as Hugh said, and IP. We have some niche capability around compound semiconductors. We have some very small chip manufacturers as well, but the economies of scale that you get, particularly in south-east Asia, are what drive the low cost of chips globally. There is a pent-up demand for those, which is leading to some increased costs at the moment. We have seen that flow through, particularly during covid, in the past couple of years. We have got a bit better at managing some of those supply chains as a result of some of the shocks of the last couple of years.

I wanted to clarify something. There are obviously stumbling blocks in accessing the grid. Is National Grid doing what it needs to do to support your industries?

National Grid, for a number of years, was not able to build ahead of demand as well as it would probably have wanted to. We are seeing some of the long-term impacts of that. Where we want to see more clarity around the grid—and it has gone through a number of changes over the last couple of years—is for grid connections to be given greater certainty that those dates are going to be held to. That would allow us to invest with a bit more certainty, and it would reduce the cost, ultimately, of that. Grid connection reform is the bit that we really need to crack now. We have been working very closely with Government and NESO on this. We need to see that enacted and driven through now, and be able to put that process in place so that, when bids are getting into the system for grid connections, we know that they are viable—they potentially have a customer, and maybe they have made a financial commitment of some sort. That increases the chances that it will go ahead and makes it better for everyone who is looking for a viable grid connection. We need to deliver, but I think we have got over the hurdle of what that looks like and what the solution is.

It is not as bad as it was.

The desire is there. It is the pace at which that is happening. Everybody, I think, understands the challenges and how we have got to this position, but it is being able to move forward at pace so that we can deliver the infrastructure and capture the demand that is out there.

As a good example of the challenge that we face, we are maybe £2.5 billion to £3 billion into a project that was due to come online at the end of 2026. Because an electricity substation that was on track to be upgraded has not been upgraded, we are now not able to get that full supply for another 18 months. During this next 18 months, we are obliged to turn off the data centre, in its entirety, with 20 minutes of notice. Given the investment that we have made, that is a pretty unsustainable situation for us.

This is National Grid’s fault.

On National Grid, to echo the points made earlier, the will is absolutely right. The dialogue that we have as an industry now with National Grid and NESO is extremely strong and positive. There are a series of factors that have resulted in this substation not being upgraded that are challenging, but these are very real. These are very real implications of this kind of delay.

We have scratched at something that I want to talk about more, but I know we are out of time.

Thank you to the panel for all the evidence that you have provided today and for your time. It has been extremely helpful in shining a spotlight on some of these very difficult and challenging parts to your businesses and to the sector as a whole. That concludes this panel.