I advise members of the constitutional requirement that members must be physically present within the confines of the Leinster House complex in order to participate in public meetings. The evidence of witnesses physically present, or who give evidence from within the parliamentary precincts, is protected by absolute privilege in respect of the presentations that they make to this committee. However, they are expected not to abuse this privilege and it is my duty as the Leas-Chathaoirleach to ensure that this privilege is not abused. If their statements are potentially defamatory in relation to an identifiable person or entity I will direct them to discontinue their remarks. It is imperative that they comply with any such direction.

Members are reminded of the long-standing parliamentary practice to the effect that they should not comment on, criticise or make charges against a person outside the Houses or an official either by name or in such a way as to make him or her identifiable.

We have apologies from Senator Gareth Scahill. Senator Alice-Mary Higgins is substituting for Senator Lynn Ruane.

This is our second meeting in a series on the topic of AI and the environment. This is a very important topic and I am glad we can discuss it today. It is also one year since this committee held its first public meeting. We have been very fortunate to be able to engage with expert individuals and organisations who have shared their knowledge with the committee. Today we are very pleased to be joined by five more experts. We welcome: Professor Lisa Ryan, professor of energy and economics at the school of economics and the energy institute; Professor Andrew Keane, professor in the school of electrical and electronic engineering and director of Nexus UCD and the energy institute; Professor Brian Mac Namee, associate professor at the school of computer science; and Dr. Patrick Brodie, assistant professor in the school of information and communication studies, University College Dublin, UCD; and in case people thought we had a bias towards UCD, finally we have Dr. Patrick Bresnihan, associate professor in the department of geography at Maynooth University.

They are welcome and I invite the witnesses to make their opening statements.

I thank the Leas-Chathoirleach and members for the opportunity to speak to the committee today on the important topic of the impact of data centres on the electricity system in Ireland. As the Leas-Chathoirleach has said, the UCD energy institute is represented here today by Professor Keane from electrical engineering and by Professor Lisa Ryan and myself from energy economics. Professor Keane is director of the energy institute in UCD, and Nexus UCD, which is an all-island strategic partnership programme on energy system decarbonisation.

I am co-director of Rinn Energy, Research Ireland's national energy research centre.

The provision of electricity for Ireland's data centre sector has reached a critical stage. While providing economic value, as demonstrated through the recent report for the Department of enterprise, the sector's unprecedented electricity demand poses severe challenges to grid infrastructure, resource adequacy, consumer costs and climate targets. This statement outlines the impact of this growth on the Irish electricity system with accompanying recommendations for policy measures.

Ireland's electricity system is undergoing rapid change with the transition to renewable energy and increased demand but that change is becoming increasingly unbalanced. Data centres now account for around 22% of national electricity demand, which is an unprecedented share on a per capita basis in a developed economy. This was recently flagged in a United Nations report. This is up from only 5% of our energy demand in 2015. Over the past decade, data centres have accounted for 88% of Ireland's electricity demand increase and crucially, 96% of these facilities are concentrated within the greater Dublin area. EirGrid forecasts continued sectoral growth with data centre electricity demand rising to 31% by 2034, which would be double 2023 levels. That is going to continue to pose significant system and cost challenges to Ireland. The demand from data centres is large, concentrated and relatively constant. It does not match well with the Irish electricity mix of wind and gas, which is more suited to flexible demand. Data centre energy demand contributes to more frequent dispatch of higher-cost gas-fired generation. In practical terms, with higher demand, this means that more expensive and less efficient gas generation is called upon more often and these costs flow through to consumers in the form of higher electricity prices. We note that data centres do mitigate systems costs directly and indirectly via connection costs and power purchase agreements but the full extent of this requires further investigation.

Renewable generation has not been delivered at a pace sufficient to match this demand growth. While the expansion of data centres can reasonably be viewed as a success in terms of foreign direct investment and industrial development, it has not been accompanied by a commensurate increase in renewable supply and much of our new wind and solar output has simply been absorbed to meet this additional load. On current trends Ireland is set to miss the 2030 target of up to 80% renewable electricity as set out in our climate action plan. I am now going to hand over to my colleague Professor Keane.

I thank Professor Ryan. Electrification is a central pillar of our decarbonisation strategy but rising demand from large energy users without aligned renewable supply and system measures undermines the emissions reductions we expect from electrification. The grid has physical limitations. Power transmission capacity has not been expanded in several years. Electrification is the right path but it requires investment and action in a timely manner. Ireland's ambitions to host further data centres are equally constrained by this failure to expand power transmission and capacity.

Our regulatory framework has struggled to keep pace. The Commission for Regulation of Utilities's, CRU new electricity connection policy mandates that facilities build dispatchable local generation or storage and that it be made available to the single electricity market. It also requires that data centres source 80% of energy from new domestic renewables after six years and that grid constraints are taken into account when assessing data centre locations. This represents a step forward in technical and regulatory terms but does not fully address the issues at hand.

ESB Networks and EirGrid are mandated to provide connections but there is no explicit mandated framework for prioritising scarce grid capacity between large data centres and essential public infrastructure such as housing and transport. In practice, this means that data centre projects are effectively competing with other projects for limited network capacity. Network capacity is a limited resource and takes time to increase. If large energy users use the bulk of any available capacity, it poses a risk that key public transport and other critical infrastructure that support our society and economy are delayed. This further steepens the challenges Ireland faces in the delivery of climate, social and other economic actions.

Given their scale and operational characteristics, data centres should be treated as active participants in the electricity system, where they invest appropriately in providing flexibility and system support services that enhance grid stability and security but this will not happen without a clear policy framework. Data centres must be required, incentivised, or both, to align their demand with system needs and to support rather than hinder our decarbonisation pathway. Equally, reporting on data centre use, grid connections and emissions to foster public trust and support robust policy formation is required. We welcome the committee's intention to contact Uisce Éireann and recommend that experts on the water energy nexus and water usage and management in relation to data centres also be engaged.

If Ireland is to continue to develop data centres, we must be honest that this is a deliberate policy choice, not an inevitability.

That choice carries consequences for our ability to meet climate targets, keep bills affordable and allocate grid capacity in line with the public interest. If we choose to proceed, then the terms of that development have to change. Data centres should only be connected and expanded where they are clearly aligned with decarbonisation, system security and wider societal priorities.

I thank the committee very much for the invitation to come here today. I am honoured to act as a witness.

I am a professor at the school of computer science in UCD and have been working in AI for over 20 years. As well as my own work on AI teaching and research, I have held several roles that are relevant to the work of the committee. I am co-director of the Research Ireland Centre for Research Training in Machine Learning, known as ML-Labs, where we support over 120 PhD candidates working in machine learning and AI. I was one of the founders of the Enterprise Ireland and IDA funded CeADAR centre for applied AI research. I am one of the co-directors of the Research Ireland Insight Centre for Data Analytics and I am now part of the leadership team for the newly announced Research Ireland Rinn AI centre. I am also a director of Krisolis, an Irish company that delivers AI and data science training.

From my involvement in all these different AI activities across these different roles, I would like to highlight three points that are especially relevant to the work of the committee and the topic of AI and the environment. The first is to highlight positive uses of AI in this area. The committee heard from my colleague in UCD, Professor Andrew Parnell, last week about using AI for weather forecasting. Other examples from UCD and the insight and ML-Labs centres include using AI to process satellite imagery to measure tree canopy coverage in cities, using AI to measure coastal erosion rates, model air and water exchange in rainforests and provide live air quality measurement and forecasting using sensors and parcel delivery vehicles. Modern sensor technologies, for example, satellite imagery for earth observation and networks of small sensors to measure environmental characteristics, allow us to measure our environment with massive spatial coverage and at very high temporal granularity. AI techniques are a key tool for interpreting this volume of data and making it useful. We are already doing a lot of this important work in Ireland. There is great potential to do more of it. There is real potential for Ireland to be a leader in this area.

The second issue I wish to discuss is generative AI and large language models, LLMs, for example, Chat GPT, Claude and Gemini, that have emerged in the past decade. The capability of these approaches has been driven by bigger and bigger models, more and more training data and massive amounts of computation, with associated consumption of energy and other resources. This is often referred to as "the scale is all you need approach" to building AI systems - bigger models, more training data and increased computation to achieve more capable models. As these generative AI models have become more capable, we have seen a parallel push for their deployment in a huge range of applications, leading to the era we are in today, of AI-enabled everything. This has led to parallel increases in demands for computation and other resources. It has also led to problematic uses of generative AI technology. The volume of AI-generated content online now eclipses human-generated content, deepfake and similar technologies have become commonplace and there are potentials for significant harmful impacts when young people interact with some AI systems. The same generative AI techniques, however, also give us automated translation systems, automated close captioning systems and novel approaches to drug discovery that are starting to have a big impact. The use of generative AI requires careful consideration and regulation to balance its potential with its impacts. It would be great to see Ireland continue and reinforce the regulation and policy efforts to protect against inappropriate uses of generative AI, such as the use for misinformation and disinformation and the harmful impacts that young people can have interacting with AI systems, and to promote careful, responsible and transparent use of AI.

The last issue I would like to raise is that we could think of the current scale-dominated approach to AI as the adolescence of AI. There is, however, an emerging set of alternative AI approaches, known as lean AI, that use less computation, less, more carefully curated data and careful interactions with people during model training to result in smaller, less computationally and resource expensive models. For example, the open source, small language models released by the French AI lab, Mistral AI, have shown comparable performance to leading larger models with a fraction of the computational cost. In the new Rinn AI centre looking at AI, this approach to lean AI is one of the key research techniques that we are working on. There is an opportunity for Ireland to become a leader in this alternative way of doing AI. It is fantastic that we are being supported to do this work in the Rinn AI centre. It gives us an opportunity to lead, to develop new, more responsible approaches to AI and to continue to train graduates in Ireland that we need.

I thank the committee for the opportunity to come here today, and I am happy to answer any questions.

I thank the committee for having me. I am an assistant professor in information communication studies at UCD. For the past nine years, I have studied the environmental and political implications of data centre growth. I am a recognised expert in the political ecology of digital infrastructure and data centres specifically. Colleagues and experts have already briefed the committee today and in other sessions about the considerable environmental impacts of data centres in Ireland, especially in terms of energy usage. It is key to bring up that it has been the case that data centres in Ireland have developed in this way since before the AI boom. The closest country comparison to Ireland's 22% of national grid capacity is Singapore, which was in the range of 7% to 9% in the same year. In Europe, the Netherlands was somewhere between 3% and 5%. The recent UN report on AI and sustainability corroborated the fact that Ireland is an outlier - "a cautionary tale" was the term used. As Professor Hannah Daly told the committee at the previous session, before and since AI expansion, data centres have represented an explicit challenge to Ireland's climate goals.

According to a recent KPMG report favourable of data centre growth and commissioned by the Government, Ireland's installed data centre capacity is projected to grow by 19% between 2025 and 2030. This is based on 300 MW worth of existing contracts for further grid connections as of the end of 2025. This would mean a total installed capacity of 1.84 GW. However, a recent market intelligence exercise undertaken by EirGrid indicates that additional demand by data centre operators and developers over the 15 years to 2040 will be of the order of 5.8 GW. This growth comes alongside EirGrid's acknowledgement that the operator will be in a challenging position in the near term, until at least 2028, but probably beyond.

While these impacts on the environment and the consumer milieu are important to recognise and mitigate, probably more important are the ways in which these impacts are shaping the transformation of our energy systems through new regimes of infrastructural provision and planning policy. The research Dr. Bresnihan and I have done has shown that, from an historical and a contemporary perspective, data centres map onto a trajectory of Irish economic development that I think we are all aware of, specifically the strategy of industrialisation by invitation, forged as a sovereign industrial development policy in the postcolonial and intervening period. Understanding this, specifically the infrastructural imbalances that it has produced due to these recognised trade-offs with the environmental impacts of these industries that have exacerbated over time, is essential to meaningfully govern and address environmental impacts and their implications for the Irish public. The point we want to make clear is that the environmental implications of AI are not just the impacts in terms of resource use and pollution but, rather, about who is responsible for, who benefits from and who owns the productive capacities of industrial transitions that are taking shape through our public infrastructure in Ireland.

Given the constraints on Ireland's energy grid for the further build-out of data centres, we have seen at least a decade of efforts to resolve these limits through innovative policy and technological solutions. What we have seen is a model of green transition as a transfer of functions from the State to the tech sector, which is a form of privatisation by stealth being pushed by the sector. Due to the scale of data centre growth in the period from 2015, the State, for reasons already cited, has been unable to provide the necessary infrastructure through this existing model of State provision, so technical solutions have had to take the place of infrastructural assessment, focused on delivering energy security and decarbonisation. For example, we have discussed corporate power purchase agreements, which will not necessarily be enough to contribute to decarbonisation with the expected growth of data centres.

If data centres are going to continue drawing power from the grid, the idea is that we let them build their own off-grid solutions. This has become a strong arguing point, and not just here. There are what are called behind-the-meter solutions in other countries. However, this is a clear case of energy privatisation I encourage the committee to push back against. If these tech companies are able to build off-grid, bespoke infrastructure alongside our creaking public grid, what is to stop the presentation of these solutions as offering a disruptive privatisation solution to energy shortages? This locks us into these energy futures being dictated by the demands of data centre growth. There are a few examples of these companies developing grid solutions through Ireland, which they have then sold elsewhere.

From grid flexibility to district heating, these energy solutions can be sold back to us as a service once they are tested on the public grid and, if push comes to shove and the solution does not come through renewables, fossil gas and even nuclear energy are proposed as potential transition solutions. This is all while taxpayers and consumers are footing the bill for grid upgrades.

These contradictions need to be taken seriously rather than dismissed as necessary trade-offs or the costs of doing business. These are mechanisms of backdoor privatisation and fossil fuel lock-in in our energy sector. A truly just and secure energy and environmental future when it comes to AI requires a strong, meaningful rejection of an AI arms race that proposes we need to develop these things rapidly and without consideration for environmental impacts necessarily, and a refocusing of priorities on climate action and digital policies, which are explicitly in service to the needs and democratic mandate of the Irish public.

I thank the Leas-Chathaoirleach and the committee for the invitation to appear today. My background is research in different areas of environmental governance and policy in Ireland for around 20 years. The last seven of those years has focused on the so-called twin transitions of digitisation and decarbonisation. A lot of that work has been alongside Dr. Brodie. Colleagues, and Dr. Brodie in particular, have emphasised the risks associated with locking our energy transition into the business strategies of multinational tech companies and AI growth. The focus of my contribution will be on the costs of this infrastructural dependency for Ireland. I am going to briefly touch on three forms that these costs may or are taking: financial costs to households and the public; opportunity costs associated with not pursuing alternative transition pathways; and social costs, as the energy transition becomes negatively linked with AI growth.

I will start with financial costs, some of which have already been covered by colleagues. A recent report by Friends of the Earth outlined how the provision of energy for continued AI growth via the public energy system involves financial costs for households. The research estimated that between 2015 and 2023 households in Ireland paid an estimated €715 million extra for their electricity. At the same time, recent figures from the Commission for Regulation of Utilities, CRU, show that a record 500,000 households are now in arrears over energy bills. That is for electricity and for gas. In April, Eurostat found that households in Ireland had the highest electricity prices in the EU for the second half of 2025. It has already been mentioned that as continued AI growth poses challenges to Ireland's climate goals, there are also potential financial costs in terms of EU fines. A joint report from the Irish Fiscal Advisory Council, IFAC, and the Climate Change Advisory Council in 2025 warned that Ireland could face what it described as staggering costs of between €8 billion and €26 billion if it fails to meet legally binding climate targets by 2030. Meeting those climate targets is going to be made a lot harder by the growth of data centre demand.

Moving to opportunity costs, one of the arguments made by the industry is that Ireland's energy transition depends on AI growth. This relates to the point about the twin transition. Whether that is financial investment in renewables through corporate power purchase agreements, CPPAs, the scale of data centre demand, which allows for the build out of, say, offshore wind energy, ancillary grid services, as referred to by Dr. Brodie, or district heating, all of these are seen as innovative solutions not just to data centre energy demands but to energy transition and decarbonisation. However, in shaping Ireland's energy transition around the needs of AI and this specific sector, what alternative transition pathways are we foregoing? Renewable energy generation capacity in Ireland reached 8 GW in March 2026. This is up from 1.2 GW in 2015. This is a significant achievement, yet most of this additional capacity has been absorbed by data centre demand.

Last month, the Climate Change Advisory Council reported additional renewable energy generated is being "cannibalised" by data centres instead of displacing fossil fuel generation. In other words, the opportunity cost of allowing tech companies to finance and supply their own renewable energy is that the rest of society finds it considerably harder to decarbonise and reap the benefits of that decarbonisation. To compare this to Spain as an example, since 2019 that country has doubled renewable energy generation, with the share of renewables increasing from 40% to 56% in terms of the total energy make-up on the grid. Spain's energy transition has been lauded not only for reducing energy related emissions but also for reducing electricity bills and insulating the country against recent energy price volatility.

Spain's energy transition is in no way radical. It simply involves long-term state planning and an independent industrial policy, which contrasts with the case of Ireland where our energy transition is increasingly tethered to AI growth and foreign direct investment.

The deployment of resources for AI growth creates additional opportunity costs when we consider that housing and other critical infrastructure require resources and labour. This was touched upon by colleagues already. In particular, we can think about labour in construction. The primary source of jobs in the data centre industry is construction, both in Ireland and through Irish-owned companies building data centres abroad. At the same time, the Irish construction sector faces a severe shortage of workers. Property Industry Ireland recently estimated that Ireland will need between 95,000 and 110,000 additional workers to meet Government targets for housing and critical infrastructure up to 2030. The report also highlights that Ireland is exporting trained talent to international markets where returns on projects like data centre construction are a lot higher.

A third area of costs is social costs. This is a bit harder to quantify but it is no less significant. We must consider that there is a social cost for those living close to the physical infrastructures of AI. We are seeing this manifest in acute forms in parts of the US where so-called gigafactories are being built in places with long histories of environmental injustice. We have not seen anything similar in Ireland, but this may change as more and larger data centres are built, particularly outside of Dublin.

A second social cost associated with the current model of the AI-led twin transition is that it can fuel a more general disaffection directed towards State-led climate action and infrastructural development. In our research on opposition to wind farms, battery storage and other essential grid infrastructures in the midlands and north west, a recurring refrain was "We are not against the infrastructure, but who is it for? Is it for data centres or is it for us?" We hear a lot now about how the social contract is strained because of a lack of social and public infrastructures combined with increased costs of living. The flip side of this is that the State appears to be doing a lot for multinational tech companies, in particular through planning and infrastructural interventions for data centre growth.

It will become increasingly difficult to sell the story that we need more data centres because they bring investment, deliver climate action and address societal issues. The current model of twin transition is led by AI companies that openly admit their aims are exponential growth, profit maximisation and the cutting of jobs, and that this can only be delivered by wide-scale adoption and scaling up of infrastructure for their needs. The risk is not just that this will lead to growing mistrust or resentment of State policy and institutions, but that it will turn people against the necessity of large-scale energy transition and decarbonisation, which is potentially the most significant cost.

I thank all of the witnesses very much for their opening statements. We will now go to questioners. Each questioner has five minutes.

I thank the witnesses very much for their opening statements and the additional information as well. I will ask questions of each of the witnesses. We regularly hear two sides of this argument. The witnesses have articulated it well between them in terms of the economic opportunities involved. We certainly hear it from the Government. We also hear it from local government. There are local authorities that say our data centre policy has transformed local authority funding and their ability to raise rates and deliver local services. There are then the very real environmental, social and financial implications, as have been outlined.

I have two questions. Is there a happy medium that can be struck in relation to it and what does that look like? Is there a sustainable trajectory and what does that look like in terms of policy and regulation? Separately, and a number of speakers touched on this, there is the question of a private wires policy. It was not stated explicitly but I took it from a number of the contributions that there are some concerns about what might happen in that regard. It is a live issue in this place.

There is no doubt Ireland has been successful in attracting data centre investment and establishing an unprecedented number of data centres in Ireland, and from that has flowed economic benefit. What I am speaking to is that in terms of a sustainable path, we need to match that with renewables roll-out and decarbonisation. The grid has limitations. At 22% as against 7% in the other international comparator, we are way ahead so we have been successful in that. We need to now invest in our public grid infrastructure to support all grid activities, which can include data centres, but it also needs to include public transport, housing and so on.

The average in Europe is 3%. You expect that, if we have 22,% our grid should be that much better than the rest of Europe if you are going by direct comparison. It is not. If anything, we have more difficulty because we are on a more difficult trajectory in the context of our renewables. We do not have hydro, by contrast with other countries, which provides a stable supply of renewables.

Do the witnesses think nuclear provides a fix? My sense is that is where some of that argument originates.

It will in the future perhaps, but it is not really worth considering now. Even if you were to get into small modular reactors, which is probably the only way you would consider it for Ireland, they are not ready and are not commercial. There are only two in the world in China and Russia. That tells you something. It will be 2040. We are dealing with an emergency right now, so it is not worth getting distracted with it.

I echo that. We have just produced a paper on nuclear for Ireland and our assessment is post 2050 for some of the reasons Professor Ryan outlines. There is a range of aspects, but small modular reactors are not as advanced as we maybe would like. We will just not be able to support it at this stage. I will also address the private wires point. Private wires is an interesting technical solution, and they are part of the solution. However, to use a roads analogy, it will allow a private road to be built on a localised level to connect a project to another project. What we need to do is upgrade our motorways or our national grid. That will provide the grid capacity to stop the constraining and containment of renewables and to allow grid capacity for data centres and other aspects. Private wires are relevant, but the real issue is the public route.

I will briefly touch on the issue of a happy medium and where private wires comes into that. The Deputy asked if there is a happy medium in terms of policy and regulation. The key thing is recognising industrial policy. This is not just an Irish issue with AI; it is EU-wide. Industrial policy is not mapping well onto existing models of digital regulation. These two things are not corresponding. Looking at industrial policy, we need to look at things that are developing by and for the Irish public. It is about economic benefits as well as mitigating environmental impacts. In terms of private wires, we are looking at a private solution. We are not looking at something being directed toward the public benefit. That is the key issue there.

I thank the witnesses. I am trying to choose where to go with my questions. These were really interesting presentations. I was struck by a couple of things. There was the piece on the cost, but also that point wider than the economic cost. People might want to touch on that economic cost and why the Spanish model is so different. What would it look like if the increase in renewables we have had, instead of being cannibalised, as the Climate Change Advisory Council put it, had been put into electrification? What would that look like? It gives us a bit of a picture of what we are not choosing and what impact that may have had on an economic level, if the witnesses have some sense of that - I know this is not the finance committee - and on our being closer to meeting our carbon and climate targets. I want to get a sense of what is not happening that could be happening. I know a couple of examples were given of areas like housing and others where we are hearing of direct tensions.

I think the lack of a prioritisation mechanism was touched on by a couple of speakers. There is a lack of a prioritisation mechanism about things that will give us decarbonisation and things our society needs. There are dangers in that, which are not just a failure to prioritise the public good over the private but in fact, with some of the planning measures and changes that we have had, almost giving it priority.

We have the Critical Infrastructure Bill coming through now which proposes exemptions from the climate Act for certain kinds of critical infrastructure, which may well be critical infrastructure that serves data centres or large energy users. It may be the danger of not just a failure to prioritise what we need to deliver but also giving preference to this model.

I am interested in the other cost mentioned by Dr. Bresnihan because it is one of the crucial ones, which is the huge disaffection. Frankly, it can only be imagined what it does to just transition telling people to run their washing machine at night, brush their teeth with the tap switched off and all of the rest of it while at the same time people can see huge unimaginable uses of energy and water, and not just here because they see it happening in the United States, for example, where we see climate change. I was struck by that idea of rates, because we should not really be going that way. Just transition funding is the kind of thing that should be allowing local areas to move forward. It can be imagined what it does to confidence in Ireland, but also on an international level, when we look to the developing world and how those industrial countries which cause climate change are now adding a brand new, very greedy industry right on top at a time when basic needs are not met, what that does to climate transition globally in terms of confidence in that model. I have another question or two but there might be a second round.

The reports show that increased demand is pushing our energy prices up because the whole demand curve is higher and more gas is coming on the system when we would not have it if we did not have such high demand. That is in answer to the Senator's question of what would it look like otherwise.

Spain is not perfect but it has got a lot of things right. They had a big blackout last year. When you are playing with a lot of renewables, it is very challenging.

In terms of the grid costs, data centres and the industry would argue they are paying a disproportionate amount for the grid costs and contend that they are saving us money. There are no numbers to back that up and that is a question that should be asked, how much is being invested by the data centres. Apparently they pay €20 million when they put on a 100 MW data centre, but I have never seen any numbers to back that up, so it would be interesting to find that out. As we can see, the infrastructure is still behind. Even if they are paying a lot for infrastructure, we still have not got a North-South interconnection. There is a lot of infrastructure that needs to be done.

On the private wires Bill, I emphasise the point that there seems to be step-change happening in terms of how the State responds to AI and the tech industry. Dr. Brodie mentioned that there is a longer history of the Irish State IDA providing infrastructure for multinationals. Intel is a good example but even going back to the sixties. What is happening now with AI and the tech sector is that we cannot keep up. We are trying to find solutions that allow them to build their own infrastructure. There is a general sense that that seems like a no-brainer and a win-win, but we are entering uncharted waters, because what happens if there is a crash? What happens if they decide to leave and our energy system is locked into their infrastructure, technologies and investments? We hear a lot about the precarity of corporation tax. The same kind of argument needs to be had around their role in shaping our energy grid and energy transition. That is a point I would like to emphasise on the private wires Bill. That is where all of that comes to the surface.

I thank the witnesses. I have a bit of a problem in terms of the argument that some of the witnesses are making in that they are presenting data centres as something completely divorced from all of our daily lives, that they are set aside and totally separate from everything we do, as if we as citizens have no interaction with data centres. I would love if one of the witnesses even put up their hand and said they do not use a data centre, but the reality is we watch Netflix, listen to Spotify, we may have a bank account where bank transfers are made, hospital data is stored in data centres, and for those of the witnesses with universities, both the institutional and research data are stored on data centres. Part of the difficulty I have is that people who present data centres as evil and consumers of all of this energy do not present the alternative. If we got rid of all the data centres in the morning, where are we going to store all of the data? Do we go back to servers based in individual institutions?

Part of the debate we had last week related to the difference between what is good AI and bad AI. I am a strong supporter of small language models that have very specific purposes as opposed to large language models. The reality is that data centres are part of our lives. It is going to be impossible for us to divorce ourselves from them. Even if there were none in Ireland - from a data security point of view, it is critical that we have them - we need some system whereby we can store our data. I totally accept that there are questions regarding energy usage, but it is dishonest to paint data centres as just systems that are guzzling loads of energy and are not connected to what we, as citizens and consumers, do on a daily basis. One of the main reasons that data centres grow is because is a demand for their services. What is the alternative?

There is no alternative. The Deputy is right that data centres are part of the infrastructure of the modern world in the context of all those things he mentioned. AI is one of the uses that we have in data centres. All those very everyday boring uses of the data plumbing that underpins our lives are key. It is a question of balance and asking whether we are doing the right things.

Somebody mentioned the idea of an AI arms race. We should not get tangled up in that. At the same time, AI and data sovereignty are key. People may have noticed in the news that the Claude models, which are US-based and are provided by the company called Anthropic, have been shut down outside of the US. If we are not at the forefront when it comes to some of these technologies, we run the risk of other countries deciding to stop the use of them. As the Deputy has said, it is key that we do not separate data centres and that we remember that they are a piece of the modern infrastructure and plumbing in a modern environment.

I welcome the announcement at the weekend of investment in Cork. That investment will help address some of our data sovereignty questions. What does the State need to do from a sovereignty perspective in order to ensure that we can continue store data in centres here?

We need careful regulation regarding what is important to have within the country - these points may have been made previously and people may want to pick up on them - and what the data centres that we support here are being used for. We must ask, for example, whether these centres are being used for things that have a benefit for the country. That may be something that we could prioritise.

We are talking about the holding of citizens' data, university records, health records-----

Yes.

-----and banking data. In other words, any of our interactions with Government institutions should be stored on data centres here.

Having sovereignty over those things is really important.

Can somebody suggest an alternative to using data centres in a modern economy?

On the inevitability narrative, these is an issue because these companies are cultivating demand and dependency.

I appreciate that, but I am talking about all of the records to which I just referred. There is the Government and there is the State. If people watch these proceedings on Oireachtas TV afterwards, they will be doing so by means of accessing a data centre. What is our alternative? Where should we move all of our health data, etc., to if we do not use data centres?

It would be a longer transition than something that would happen overnight. It would be something that would be developed more by and for public interest-----

As of right now, however, we cannot say that we are closing all the data centres and transferring the data somewhere else.

I do not think anyone is suggesting that. With respect, we agree that data centres are a part of the modern infrastructure. While other countries have 3% of their electricity demand consumed by data centres, we do not have the balance that we have been mentioning. We are not necessarily talking about closing the data centres that are here, but, maybe, about not allowing the next 19 to come in so quickly. That is more than Ireland needs.

Last January, an interesting discussion took place at the World Economic Forum in Davos. All of the CEOs of the major tech companies talked about the lack of demand.

If we think about AI, it is a stack and there are chips and videos as part of that stack. There is the infrastructure of the cloud service providers, such as AWS, Google and Microsoft. There are also the language models and the applications at the top. There is not enough demand. The money is going into the infrastructure but they are finding that there is not the demand. There is not the diffusion of their models and the applications to the extent that they want. It is the reverse of what Deputy Byrne is saying. There is a huge multitrillion dollar investment in these infrastructures without a clear social benefit or even commercial benefit having caught up with it. This tells us a lot about the possibilities of it being a bubble. There is also the question of whether we want to keep going down this trajectory.

I will start with the pricing issue. I thank the witnesses for their very compelling statements and a very interesting discussion so far. One of the difficulties I have, which I know that many people at home have, is that those households struggling to pay their electricity costs at present pay far more for the energy they consume compared with data centres. Essentially they are subsidising some of the wealthiest companies on the planet. It was particularly jarring when listening to Professor Ryan's contribution. The constant demand of data centres seems to contribute significantly to gas-fired generation. Will Professor Ryan drill down into how this is happening in real terms? Why is it that data centres pay less per kilowatt hour than an average citizen? Why do they benefit from lower network charges and taxation on their electricity consumption? Is all of this a policy choice of the Government?

I thank Deputy Gibney for the question. Industrial policy in general in every country usually gives lower per unit prices to industry. In Ireland we do not have heavy industry so data centres as a sector represent our industry. It is a services sector and that is our industry. It is quite normal that they pay a lower energy price per unit.

In terms of why it is driving up prices, in the electricity market we have a merit order, so the price is set first by the cheapest units which are produced by wind. Wind has a marginal cost of zero. There is no fuel cost. It comes in first. As demand increases and we do not have enough wind to satisfy all of the demand, then we put on a gas-fired power plant and this has much higher cost because it uses natural gas. If people did not need that last plant, then the energy cost would just be based on wind, assuming there is wind on that day. As demand goes up, we put on more gas-fired power plants, and at peak times, when we have high demand, the least efficient gas-fired power plants are the last to be brought onto the supply system. This sets the marginal price. This is why it drives up prices. This is only one third of our overall prices. Then we have standing charges and other costs associated with renewables. It is not a simple picture actually.

Professor Ryan does not necessarily have an issue with the structure; it just has to do with the volume of data centres.

That is the most efficient way to run an electricity market and I do not have an issue with it. It was designed around thermal plants and there is a lot of consideration as to how we need to restructure the market a little bit because now we have demand revenue streams. Overall it can be changed but the main problem is that as demand jumps up a level and there is the space demand from data centres, it means there will be a higher price charged.

I thank Professor Ryan. I am interested in hearing more about how the energy demands impact grid capacity in certain areas. I want to ask Professor Keane about the impact of data centres on our ability to provide critical infrastructure in the same areas as they are located or to complete transport projects. What issues are we presented with regarding the electricity demand of data centres when it comes to trying to build a new train line, EV charging or a new hospital in a particular location?

I thank Deputy Gibney for the question. Grid capacity is a local issue. We can imagine an area or a neighbourhood with a transformer and a substation with a certain amount of power that can flow in and out. If a large energy user comes in and connects, it absorbs a lot of the available capacity which then triggers network upgrades and infrastructure upgrades. It takes up the available space for anything else that wants to come in. As I noted in my statement, it is the job of EirGrid and ESB Networks to connect people who want to connect to the grid. They are not mandated to prioritise who gets in. This is something that I think needs to be looked at. There is physical limitation of the equipment. What we have seen, as I have mentioned previously, is that we have not upgraded our grid fast enough.

I am repeating myself perhaps, but the real solution here is not really private wires. It is that high-power, high-voltage infrastructure - the motorways of the grid - which just frees up capacity and provides the ability to move power across our country in much more efficient manners. To go back to the Deputy's specific question, that local capacity is very much a physical sizing of the equipment that is there.

Could there be a weighting on particular projects to adjust where it goes?

That is a policy choice, for sure.

Is it something that could be implemented into the system?

Bizarrely, I can speak to the British case perhaps more clearly. They have a strict queuing system, so if a data centre applies to connect at 12 p.m. and something else applies to connect at 12.05 p.m., it is behind the data centre in the queue. It is completely agnostic, and that served us well in a way.

Is ours agnostic?

As in it serves all electricity customers. It is there to connect whoever wants to connect. It is a further policy question as to whether or how we prioritise.

I thank the witnesses for their presentations. I will start on the issue of jobs because we have heard very overblown figures about how many jobs are reliant on data centres in Ireland. There was a particular KPMG report which said that data centres underpin 876,000 jobs in Ireland, which is a high percentage of our jobs. Do any of the witnesses have a view on how many jobs are actually provided in our data centres in Ireland?

I do not have any figures, but even in the KPMG report, the majority are in construction, and construction is short term. The interesting thing about those construction jobs and, again, it was also in the KPMG report, is that Ireland has developed an indigenous industry around the construction of data centres, which is exporting around the world. There are Irish companies building data centres around the world. That is of economic benefit, but the point I was making in my statement is that there is also an opportunity cost because a lot of construction workers are working abroad building data centres rather than building housing and critical infrastructure here. It would be interesting to look more into those figures and to do more research into that, specifically around the question of labour. We hear a lot about energy, and maybe less so about water, but the question of labour and construction labour is also a part of that opportunity cost. I do not have any figures.

There was a report in The Irish Times earlier this year which had a quote from the spokesperson for the lobby group for data centres about the number of jobs directly created by data centres, and it was 1,800. We used to get figures from the Government about this. We used to ask it, and we used to get them, then for the past four or five years, it has just point-blank refused to answer the direct question about how many people actually work in the data centres, as opposed to how many use cloud computing or are involved in building them. There is a data centre around the corner for me. It is massive. It is located on the site of the former Jacob's factory that used to employ huge numbers of people making biscuits. Very few people seem to work there or come in and out of it. They are very large facilities with a relatively small number of people working there. Does that seem about right?

I have been researching data centres in Ireland for quite a long time, and this is something that the industry admits. There are not very many people employed full time in a data centre at any given point. I have visited data centres where there are six to seven people on site, but, again, I do not know how generalisable that is because, of course, there are people in and out, and it depends on if it is co-location or hyperscale, but the consensus is that there are very few people employed in a full-time sense in data centres.

I want to add one thing on the construction sector. The KPMG report brought up the development of a domestic construction sector which has developed €2.1 billion in construction exports in the specific model of data centre construction. That is obviously important, but, as Dr. Bresnihan said, there is a way in which that might be a misallocation of construction resources if there are very clear shortages in other sectors. At the same time, foreign-owned technology companies exported €262 billion in computer services during the same study period, so the scale of the domestic industry developed around this was at less than 1% of the value in exports developed by these five foreign-owned technology companies.

There is something important there in terms of that scale, and how those resources are being allocated and distributed in the context of domestic benefits in Ireland. That was something important from that report.

I want to explore one of the points made about these major tech companies operating as de facto energy utilities, and the public and climate needs of energy transition becoming secondary to the imperatives of growth in this one sector. Effectively this is like a drive towards the privatisation of our electricity and tying it to the needs of big data centres. If we continue on the kind of policy trajectory that we are on, what percentage of our power supply is likely to end up being privatised or privately owned, and what are the future implications of that?

The figures are the ones EirGrid have. It is currently 22%, as Professor Hand outlined, and I think that the forecast is to 30% by end of the decade. There was then the EirGrid market forecast that predicted that it would be up to 5.4 GW by 2040. I do not know what that is in terms of the percentage but it is quite a huge increase.

Going back to a point I made earlier about uncharted waters and the question of privatisation, it is not exactly that there is a public grid that is being sold off to a private entity. It is more that a sector - the tech multinationals - are being enabled to build and provide their own private infrastructures, and that is why it has less scrutiny around it. If it was the selling off of public infrastructure, there may be more controversy and questions but, as I said earlier, the idea of them providing their own solutions seems like a win-win. It is, however, uncharted. The Private Wires Bill is a reversal of the legislation that was introduced in 1927, so for pretty much the history of the State there has been this public mandate and a public grid has managed the energy system. We do not know what this will look like, particularly with offshore energy, and clearly there is an interest with this expanding demand around AI but that is an abundant resource that we have not developed. The tech companies do have financial resources. They do have increasing energy expertise because energy is as important part of their business model as data. That is what they are dealing with. If we are going down the line of what will that look like, we do not know. Going back to this idea of an opportunity cost, the various alternative transition pathways seem to be narrowing towards the ones that suit or towards the needs and growth of AI as opposed to decarbonisation.

Can I just clarify that the figure of 22% right now does relate to private ownership-----

It is usage.

It is a portion of that 22%, and a portion of the 31% in 2024.

I thank Professor Ryan for that clarification.

I thank the witnesses for coming. Between them all, they have a lot of knowledge, and I want to draw on some of that knowledge. What is the most significant change that the witnesses have seen in their fields regarding AI and data centres? What can be done to make things better for the future of our country and earth?

Speaking from my domain in the electricity grid, when we think of data centres accounting for 22% of usage and maybe getting to 30%, they should be incentivised or mandated to become active participants in support of the grid. It is about flexibility. They have a lot of resources behind them. Even if they have a private wire connection, they will probably still have a connection back into the public grid, so we should incentivise or mandate them to support the variability of renewables, so as renewables vary, their demand flexes with that. If we take that 22%, it may be an opportunity to uniquely balance renewables in a way that nowhere else has. That is my perspective on it.

From an AI point of view, the jump in performance in AI has just been massive. I have been doing this for about 20 years. For the first ten years of that, it was a very niche thing with a very narrow range of applications.

About ten years ago, we shifted to a new approach of building AI models using large neural networks - very data driven - and that has increased the capacity of things we can do with AI now beyond what we could have imagined at all before. Applications of that are things like medical imaging, the automated captioning of images and the automated translation of languages. It just jumped massively in terms of what is capable. The moment that we are in now is where we are trying to figure out where the really valuable uses of that technology are and where slightly more spurious uses of it might be, and to pick the valuable path.

In terms of the research around data centres as it has developed over the last decade or so, since I have been doing research on the industry, what is interesting is that when you follow an industry or a topic for quite a long time you see things as they develop. About ten years ago, you would have started to see companies get very interested in the energy sector. Data centre companies especially started developing certain interests in particular technological solutions in the energy sector. A lot of that was gesturing towards the kinds of privatisation and integration in the energy systems that we are seeing today. With the AI boom we have seen a step-change in the ability of tech companies to get involved in energy for all of these various reasons, specifically being accelerated by the AI arms race. That is what I would say, at least in my sector of research. Where I have seen things change really rapidly is in terms of the involvement in the energy system.

Dr. Brodie spoke of necessary trade-offs. Who is benefiting the most and what are his worries about the industrialisation by privatisation?

My worries are that we are seeing climate objectives and environmental objectives fall down the list of imperatives. There has been a lot of research in terms of a lot of EU policy, especially around the twin transition. Climate imperatives are starting to become secondary to the growth imperatives around AI. Something that is acknowledged by a lot of policymakers is that this is a race and we are trying to be competitive and accelerate. To me, the danger is getting caught up in an arms race, especially in the context of the wider social and environmental harms of AI. If you get caught up in this race, the basic environmental and climate implications become secondary to the imperative of growing these industries. Especially when they are being driven by some of the biggest companies that have ever existed on the planet, there is a real danger even just in the ability to rein in that growth in the context of the environment. That is my biggest worry in that area.

This is my last one. Does Dr. Brodie think a moratorium should be introduced on data centres, given the number we have in this country at the moment?

I personally think that is a minimum to pump the brakes. The growth at the minute is very clearly coming up against very practical constraints. Even from a pragmatic grid perspective a moratorium would be advisable.

I thank the witnesses very much. I apologise for missing the opening statements at the beginning of the meeting but I have read them. I am attending another committee that is on at the same time.

In his opening statement Professor Keane stated, “data centres should be treated as active participants in the electricity system, [capable] where they invest appropriately of providing flexibility and system support services that enhance grid stability and security but this will not happen without a clear policy framework.” The large energy user policy has been published, as has the private wires legislation. Is he saying that those two pieces do not go far enough or give enough clarity on what the policy position is and what it should be?

In a word, yes. As I stated, the large energy user connection policy is a step forward but there is more that needs to be implemented. They need to be mandated or incentivised to be much more active participants in that demand-side flexibility and that what they have behind the meter is in sympathy and in line with our 80% renewables target. That is there to come after six years. There is a question of timing that also needs to be considered here, but they should be there actively on the grid providing support services, as other people do. Some of these issues are quite technical. Members may have seen in the media recently about fault ride-through being a particular issue that EirGrid has talked about that data centres need to provide. That is one very good example of the kind of technical service that when you have demand that represents 22% of the grid, they do need to perform in real time in a way that ensures there is no disruption to our supply.

I am not sure that we have all those things laid out in policy yet.

Naturally we are looking at it from an Irish perspective but this is an issue, in particular, with data centres and energy capacity that we are experiencing across the world. If we look at it from a European perspective, particularly in the Nordic countries, have there been any good examples among our colleagues of how to do it right? I am looking at this from the perspective that maybe we should be developing a European policy around it, as well as member state policy on energy provision and data centres. Does anybody care to comment on that?

Again, as noted earlier, we are leading the way in terms of the level here being at 22%, if one wants to consider that leading the way. There are no European or global comparators to where Ireland is at in terms of data centre concentration. I do not know if there are good examples. In general, what I see across the EU, though, is a move towards flexible grid connections. This is where if they have 300 MW, or whatever the size is, maybe they only need 150 MW because they are going to operate flexibly. There are technical things that can be done that can be supportive of the grid.

The data centre developer community has come on a bit of a journey. Not knowing a lot about energy and certainly not knowing a lot about the grid, it is having to learn quite quickly and realise that all the ambition and capital it wants to deploy are hitting the public grid infrastructure. It is having difficulties with that. The answer may be to mandate data centres to operate more supportively towards the grid.

I cannot think of examples involving data centres because we are the only country with as many data centres. However, other countries in Europe have a lot of heavy industry. In Germany, with its heavy industry, companies are mandated to provide flexibility services. There are precedents in other sectors.

In that case, we should definitely be examining flexibility services in more detail.

Definitely. In the example I gave to Deputy Gibney about how our prices go up when our demand goes up, I did not mention that one way around that is if we could have more flexibility at that peak time. If use is reduced, it would bring prices down. That is exactly what we are trying to achieve.

I have a question on a separate topic for Professor Mac Namee about lean AI and how we could compute more carefully curated data and how Ireland has the opportunity to become a leader in lean AI. Dr. Brodie and Dr. Bresnihan spoke about the AI arms race. At the end of the day, we are a very small country in comparison to where a lot of this is being developed. We look at what Anthropic has done with its recent models. It just shut them down because it could not locate them in the US. How can Ireland play a role in driving lean AI, primarily academically, but if we could influence these large tech companies to maybe shift towards a leaner type of AI?

I can think of a couple of ways. One is through research in this area. It is great that we have enabled a new research centre to do that and to continue to do that. That is one thing, and to take that perspective to say because of our scale, let us look for a different way to do things. One of the interesting things that we have in Ireland is that we often have surprisingly large data sets compared to other areas. Medical imaging is a good example in Ireland. Because we tend to have a single version of things, we need to have some of the most coherent medical imaging data sets in the world. That gives a great resource to build on top of this and to say, let us not look at how we build AI models where we just hoover up everything we can get our hands on; let us look at building AI models where we take a much more focused, curated look at the data we are going to use to put into those models. We can take advantage of the fact that we are a little smaller and have a slightly different perspective.

The other thing in terms of impact is to take advantage of the fact that we are in a unique position where so many of those large tech companies are based here. That gives a pathway to try to make that influence and bring those ideas out to that audience, whereas in other places there is not that aspect to doing it.

It falls to me now to ask a few questions. I might continue in that same vein with Professor Mac Namee about lean AI. It is a phrase I had not heard but it is a very useful one to conceptualise. He is speaking about how Ireland can avail of models like Mistral, which he identified, which unfortunately has a lower energy usage in the model to which he refers. What is happening globally in terms of the AI arms race others are referring to?

There is this race to get multiple data centres built all around the world because of the huge energy demands that AI is bringing, but is there a huge drive for energy efficiencies taking place at the same time?

You definitely see it more recently. There are two parallel drives. There is the drive to make the ever bigger, ever more capable, model and you see the acceleration in model releases. If we think about OpenAI and its models, there was great expectation around ChatGPT-4. There was a huge build-up when it was coming whereas now we are in a place where there are new iterations of those models coming and coming. That has really accelerated. It is a race, and the push there is among a small group of organisations that have the capability to build the models at scale and to build the most capable model yet. That most capable model tends to mean the biggest, most computationally expensive model.

The flip side of that is people are waking up and taking a more critical view of the cost involved. Often, that is a financial incentive. The Leas-Chathaoirleach may have come across lots of stories about the ideas of cost per token. There is a phrase, "token-maxxing", that became very popular there for a little while where there was a drive to be the organisation that was using AI the most. There has been a real backlash to that, with people all of a sudden saying that they were just spending a ferocious amount of money on doing that. There is a slight maturing in the use of those techniques as we see a little more of the real cost. A big thing at the moment is we are not paying the cost of using these models. What we pay as consumers for those is massively reduced in price. As we are shifting to a lot of these organisations starting to charge closer to the real cost, people are waking up to how much they are using. That is really pushing a move towards the smaller and more efficient models, which is, I guess, one of the motivations for that.

The main driver towards efficiency is the market. To draw an analogy from the housing sector, if you build a new house in Ireland, it has to have a very high energy rating. That is an excepted norm but it is also in law that that is what has to be done. In terms of the regulatory space for data centres, is there a regulatory framework dealing with a drive for energy efficiency in addition to the market forces that Professor Mac Namee says are already driving it? Is that a space that is being looked at, either in Europe or elsewhere?

My colleagues here are probably better informed on that. Overall, there is a huge lack of transparency in the amount of energy being used in data centres and how that ties directly to specific uses by consumers of AI models. Probably, people here would be better informed.

To briefly jump in, there is the European energy efficiency directive, which is coming after the somewhat successful attempts of the German energy efficiency directive. There was some push-back in Germany from the data centre sector around the limitations of that and their ability to scale.

One of the key things about these directives has been transparency. There have to be mandatory reporting mechanisms in place, at the very least in the pushes - in the principles - for these directives. So far, these have been very heavily contested by the data centre sector. I do not know how much the AI companies specifically are getting involved but obviously these things are pretty interlinked.

For anyone who wishes to reply to this, we had Professor Hannah Daly before us last week who made a point about the large energy user action plan, LEAP, that the State had designed whereby 80% of the energy that a data centre used had to be from renewable sources. Then there was the 20% and a six-year lag period. I asked Professor Daly whether there was a split in the environmentalist movement. I remember Eamon Ryan, when Minister for the environment, acknowledging that large energy customers such as data centres could be a draw for the type of renewable energy and offshore investment that we needed. Professor Daly acknowledged and accepted that, but her point was that the six-year lag period between when the data centre came and when the renewable energy had to fulfil its obligation meant that, even though right now data centres were not causing Ireland to not meet its climate target - that was her position - they might do so in future were the full capacity to be arrived at. Perhaps Professor Ryan wants to comment on that. What is her view?

The question is whether they are causing us not to meet our targets now or in the near future and whether there are some benefits associated with having renewables too. It is true that we know that we need to develop offshore wind. It should be coming a lot faster than it is, but, at the moment, we do not have offshore wind. When we develop our offshore wind sector, it will be good to have demand. If the demand, as Professor Keane outlined, is inflexible and stable all the time, it will not be a great match for the offshore wind resource that we will have. It is good that some of it will be used, but unless it is flexible and there is storage, it will not be a good match for that offshore wind. We can think of other ways of using the electricity that we will have in the offshore wind sector.

I thank all the witnesses for their contributions. I was listening to the discussion online and my overall feeling was one of concern when I heard what was said to the effect that we are an outlier, globally and within Europe, in this area. There are 500,000 households in arrears in terms of energy bills in Ireland. On Wednesday, my colleagues in the Labour Party will table a motion in Dáil in relation to data centres calling for a moratorium, a levy to reduce costs for households and an independent review into the amount of energy these centres use. I want to get the witnesses' views on that initially. In response to a previous question, Dr. Brodie stated that a moratorium should be the minimum. How can we go further than a moratorium? If anyone else wants to respond to those questions, I would be grateful.

I will not spend too long answering. In the research that Dr. Bresnihan and I have done, we have thought a lot about the economic model which has led to these contradictions being so heavily concentrated. It just requires a serious conversation about foreign direct investment as the primary driver of economic activity. That is the longer term conversation that needs to be had around data centres as something which represents the concentration of this model that is leading to such significant and outsized infrastructural demands and constraints. All of the precarity and the environmental implications that come with this can be traced to a conversation that should probably be had about the reliance that we have on tech companies, specifically those from the US. That would kind of be my provocation, but it is something that might be viewed in light of the longer term conversation that I was gesturing towards in terms of dependancy.

I am not sure I would support a moratorium. The problem is that we do not have the infrastructure. We have not designed our system to work with data centres. If we do not do that, then maybe there could be a moratorium. The two go together. There are a few things going on here. One is the location of data centres. Ninety-six per cent of them are in the greater Dublin area. As Professor Keane alluded to, the grid cannot take it. Why not put them somewhere else? Near Moneypoint, for example, could be one place because we have a big high-voltage cable there. You could have a green park there and you could have other things.

The grid around Dublin is essentially full. It is a question of moving them, requiring them to be flexible and building the necessary infrastructure. It should be conditional. We can have a sector, but we can only have it if we put the system in place. It is like building a housing estate with no facilities - you just cannot do it. It is not that we say, "Do not build any housing estates"; it is just that you have to design your system to facilitate that.

Right now, we do not have the infrastructure for the centres that we have, so why not focus on catching up maybe with the system we have, pausing and making sure that we have the necessary infrastructure if we want to continue? However, it is a very deliberate policy choice to continue. If we really want to go for 31% of our electricity demand being taken up by data centres, we need to design an extraordinary policy and infrastructure framework to facilitate that.

Do any of the witnesses have a view in terms of how we can better engage with citizens in relation to this? Should there be better consultation locally in terms of where data centres are placed? Would the witnesses have a view on the matter of education? It seems a lot of this has come into the public consciousness recently.

In many senses we have sleepwalked into this without proper public debate on the issue. Does Professor Keane agree with that? Does he a view on that?

The word "transparency" comes to mind as a starting point. We need really good transparency on the energy usage of data centres and the carbon intensity. Whether they end up on a private wire network or hybrid connection, we need clarity on the carbon intensity and resources they have and the impact relative to the community in which they are situated. I am not sure we have that transparency yet. That can be a basis on which to proceed with other actions.

As Senator Higgins brought up, it goes back to the question of costs and benefits. People see there is a lot of talk about the economic benefits of this data centre boom and the investments but that is not manifesting in their everyday lives in cheaper bills or the benefits of decarbonisation and retrofitting and other things. The education is not going to work if the material conditions do not change. With the direction we are going in, I do not see how that will happen. To go back to the question of alternatives, the discourse is generally that we cannot have an energy transition without AI - financial resources, technological nous and flexibility. We need to step back from that and think about what is the energy transition we would like. That is happening in other countries that are not so dependent or reliant on AI growth and data centre growth. We could think about what that might look like - offshore energy, retrofitting, transport etc. That is a positive space to have these discussions.

We have heard a lot about data sovereignty. A moratorium is a pause not a ban forever. It is a pause, letting the systems catch up because it seems that they have not. We have heard a lot about the energy system not being caught up but also if we proceed with this expansion under the current energy system or even under an energy system that plans for 20% fossil fuel, which has been talked about, the danger of locking in. We are effectively having an expansionary piece that is locking in, be it 20% or far more for the six years proposed. The time window we are operating is why the arms race language was useful. It is not just about whether to do these things but when and how. This is in the decade when we need climate action to be front-loaded. It is about what prioritising climate action would look like; it does not mean closing the door forever. It is avoiding lock-in which will be difficult. Another part of the lock-in and vulnerability I am concerned about is the witnesses mentioned examples of public data which are talked about. We know that is a fraction; a huge amount is the dark web and trivial usage. This has been documented. The are real gaps. The energy gap was spoken about but there is a regulatory gap in terms of what is actually happening in these centres. In the arms area, dual use is not part of the EU AI regulation; it is separate. Do we need much stricter regulation around the kinds of uses and perhaps from a sovereignty perspective to move public data into public data centres as has been done in other countries? We would then not be in a situation where we have an energy vulnerability behind the meter but if there is a crash, we have important public data sitting with these companies. Microsoft has shown itself somewhat unreliable for public data and public usages in other contexts. Dr. Brodie commented on the pollution piece.

We have focused so much on energy but other environmental impacts, including pollution, have not really been touched on. There have been a lot of reports from other countries on the impacts.

Working backwards from pollution, this also goes along the supply chain. To bring it back to the first question, the models we are using for this development enable companies to dictate the terms of our environmental and energy regulations and the broader industrial pathways we follow. The key conversation is how these companies organise their own supply chains in a way that we facilitate to try to participate in the arms race to build these things. Starting with public interest and moving beyond market-based mechanisms, a moratorium would allow that conversation on how climate action can happen without heavy constraints by this one sector.

I also had questions about public data and the lock-in.

I agree with the Senator. There is a strong conversation to be had. Other European countries - I think the Senator was referring to France - have been doing this. What is key is they are moving on to software. A report came out today from the AI Now Institute saying a lot of mechanisms for moving to software for sovereignty and things like that still operate on the backbone of US big tech infrastructure. For example, Mistral has entered into contracts with Google Cloud.

I think Lithuania has a public data centre.

Exactly. That is the direction of travel that would be necessary if we are thinking about data sovereignty.

I thank Dr. Brodie.

Perhaps I can get a comment on the lock-in later.

Sorry, Senator.

Perhaps the witnesses will get a chance, when answering later, to address the lock-in question.

If you gave them more time to answer, they would have more time. I call Deputy Malcolm Byrne.

I think Professor Ryan summed it up quite well. Our challenge, as our economy and society continue to grow, is that we will have greater energy use. In growing our digital infrastructure, we also need to grow our renewable energy infrastructure. I agree with Professor Ryan. One of the problems is we have been too far behind in investing in offshore in particular. We need to address the opposition to onshore wind and solar as well. Everyone talks about the AI race but the US and China are racing ahead. For middle powers, including the European Union, it is tied to sovereignty, energy security and data security. The Mistral point has been made a number of times. It is part of the approach France is taking. I want the committee to move on both of these agendas together. How can we manage the necessary growth of our digital infrastructure at the same time as the growth of our investment in renewables? Both are critical for Ireland's future development.

The thing that couples data centres and the energy transition and renewables is the grid and grid infrastructure. We cannot have more data centres at any significant scale unless we really invest in grid infrastructure. They are decoupled somewhat. Digitalisation and data centres are not the same thing as the energy transition; they are separate things but they interact hugely. We are experiencing a second great electrification with the electrification of transport, heat and so on. It is an opportunity, in that of all the sectors of energy, we can decarbonise electricity. There are other sectors of our energy emissions and greenhouse gas emissions that we just cannot decarbonise. We have the technology and resources available to decarbonise electricity. If we get it right and expand our grid infrastructure on an accelerated timeframe, we can get to a largely decarbonised electricity supply that can support things that include digital infrastructure but also other things.

I agree with Professor Keane. The multibillion-euro investment in the grid is crucial but it needs to happen yesterday. How does Professor Keane see AI benefiting his work and research?

To answer the Deputy's earlier question, of course we use it. It involves very powerful computational methods. Regarding my area of the energy transition, we know what the solutions are to the energy transition and how to achieve them. As I have just mentioned, we have the technologies available that we can deploy now in short order to decarbonise our electricity production. That is really good news. AI will, of course, have a huge impact, but we do not need AI to decarbonise our electricity.

It could be useful in respect of the flexibility issue. We can have all the electricity coming on but at night-time we often have excess wind. That is when we need someone to use the electricity. We do not need it at peak time when we are using gas to meet demand. Algorithms and smart meters are where you use AI in the electricity system. It is making the system much smarter, even to the point of having intelligent grids. The whole thing has become smarter because of AI. If we do not require something of the AI sector and the data centres to make them more flexible, and if we do not make sure we use all the algorithms beneficially, it is just putting a heavy load on the system without any of the flexibility.

I thank the witnesses.

I thank the witnesses. This theme has been running through both of the sessions we have had.

We touched earlier on the hype aspect and the positive aspects versus the negative ones. This is one of the things I find most frustrating because it feels like every time we talk about the positive applications, they are tied to or conflated with the negative ones. For example, if we want strong weather prediction technology, we need to have energy-intensive data centre LLM solutions to provide it, even though we heard last week, from Professor Parnell, that it is not resource-intensive and that we do not necessarily need it. What we are seeing, and what I certainly see, is that those who are trying to promote LLMs are piggybacking on, say, the scientific or social usefulness of other technologies. I am keen to hear Dr. Brodie's and Dr. Bresnihan's input on the kind of regulation that might be supported or developed, or where we see the regulatory gaps that could help us come to a more thoughtful approach to AI. If the witnesses have anything to add on Senator Higgins's question on lock-in, they should please do so.

I completely understand the Deputy's frustration because I have the same frustration in the context. Frequently when I am bringing up this research, one of the critiques I hear is that there are useful applications of AI. I am talking about multinational hyperscale data centres that are contracted to huge AI companies, mostly from the US. Those are what I am talking about when I talk about data centres in this context, and those are the ones in Ireland that are having the most significant impacts. This is key. I acknowledge, as I usually have to do, that there is, for example, Professor Keane's research on weather forecasting and the use of machine learning to model weather forecasting, which is very important regarding the climate. There are very important applications in terms of monitoring carbon levels and the like, but that is not what we are talking about when talking about generative AI models, which are about demand control by a few very large companies that are pushing the agenda. Again, it is not even that they are just pushing this agenda. It is that the adoption agenda is very strong in our policy because there is the sense that we need to keep up. It is key to push against that, especially because of the downstream environmental effects. That is crucial.

Does Dr. Bresnihan have anything to add?

We use the term "big tech" sometimes, or some people do. I am not a big fan of it, but it comes from big oil and big pharma and previous waves where there was a strong consolidation of power within a few companies in a very important sector. People who are critical of big oil are not against the use of oil, and people who are critical of big pharma are not against medicine and the use of pharmaceuticals.

It is about the concentration of power.

It is about the concentration of power. It is also a matter of trying to understand that when we talk about AI, it is not just about its applications and what it can do. It is a case of digging into the layers of the stack. What we are talking about mostly today is the infrastructural layer. There are also the chips. Nvidia, for example, has a very strong monopoly over the chips used in language models.

What is driving things is investment in the infrastructure and the chips. These are the only profitable parts of the stack so far. The money is pouring into infrastructure and into the chips without demand necessarily keeping up, so the companies are looking for new markets. With big oil, plastics are a downstream component of that industry. You drill more and you have more hydrocarbons. What do you do with them? You have to find new applications.

Plastic is effectively an application of petroleum. There has long been research and an understanding of how this works in oil and in pharma. Tech has not quite caught up. When we talk about AI, we need to be talking about it in these sorts of terms and understanding what is driving things. It is not about being for or against a technology.

Exactly. That is really helpful. I thank our guests.

Let me talk about that idea. Technology has a very fluid and fuzzy spectrum. It is the same technologies and techniques that we are using for the various applications. There is no crisp line that we can draw allowing us to say the LLMs are over here doing one thing and AI is over there doing another. It is exactly the same technologies that we use to do-----

We can distinguish between uses, though.

That is the key thing. It is the uses of the technologies, not the technologies themselves, that we can distinguish between.

Maybe I will build on that. As it stands, we as a society do not get a say in this stuff. We do not get to say that we think the application of the technology in medicine is socially useful and that we should use some of our energy on that, but that we think running all the algorithms to target people with ads and push stuff on them that they did not previously know they wanted, creating artificial wants, is not useful. The reason we do not have control of these things is because this is all happening inside big black boxes owned by private corporations that are dominated by a drive for profit. Is that a fair summation of our lack of control over this whole conversation? We can have a view of what is useful or not but, at the moment, the public has very little control over how its energy should be used.

One of the key points is the closer and closer alignment of US national interests with the interests of the companies. That is a layer of vulnerability, especially in the context. Anthropic was brought up already, and the Trump Administration has openly acknowledged that Europe is going to be a junior partner in the AI race. The demand being manufactured by the companies that is driving the growth in Ireland is tying us to vulnerabilities that are quite precarious. Maybe they have always been so, but this is getting amplified by the current US Administration’s geopolitical interest in the investment in these technologies.

I thank Dr. Brodie. Turning to Professors Ryan and Keane, I am also a member of the Committee of Public Accounts, which recently received correspondence from the National Transport Authority telling it that Dublin Bus has been unable to plug in new electric buses at its Phibsborough depot. It has bought the buses but it is not even able to charge them because the grid is being maxed out. The point the witnesses are making is that 97% of the data centres are in Dublin and that we have basically maxed out the grid. There have also been reports regarding the impact on both DART+ and the development of housing in west Dublin. Can the witnesses speak to those examples or others of where the grid being maxed out, with data centres contributing, is having real impacts on our ability to provide other infrastructure?

I do not have examples that I can give the Deputy but, as outlined to Deputy Gibney, there is a physical constraint. In a local area where a big energy user is connecting to the grid, it is going to be competing with other initiatives, which may be electric buses. It may be anything. I do not know the details of what the Deputy has referred to, but it does ring true.

It is important to distinguish between private goods and public goods. Even in his previous question, the Deputy referred to markets. We cannot control markets. One thing that is frustrating for people is that the data centre issue relates to the private sector. At some point, we need to decide whether the infrastructure is needed for society. Electricity infrastructure started off as private and then the grid was understood to be something that should be a public good, something everybody needs for daily life. If AI is needed for daily life, maybe this should become public.

The example the Deputy gave is a separate point on the same thing. It is a matter of public versus private and, as Professor Keane said, whether we should we be prioritising the public good over the private? This seems to be a clear example of where we would prioritise the public interest and the public good, and it should not be that we cannot connect public infrastructure to the system because of private considerations. Then again, households are also private. It is an important discussion to think about.

I thank our guests.

The immense pressure on the grid has already been discussed. If we continue on the same trajectory of building new data centres, what will be the outcome in the near future? Is it sustainable? What are the energy solutions? The social contract has been mentioned. I presume that there is a possibility of the social contract breaking down. What do the witnesses think of the twin transition model? Is the model working for our citizens and climate change? Is there anything that should be changed?

As we have outlined, there is pressure on the grid. If we continue on this track without upgrading then that pressure will continue. As Professor Ryan outlined, we have seen an increased utilisation of inefficient gas plant, which is having an impact on the price of electricity. That would continue and it would hamper our decarbonisation efforts, without the parallel actions that we have outlined.

There are cost implications. Obviously it is a decision we want to make. If we want to invest in the grid infrastructure, this will be a cost that is borne across all of society. Who should pay for it is the big question. We have not really got to the big question of who should pay for the grid or the infrastructure for data centres. Again, if it is a public good that we need the data centres then maybe it is okay for us all to pay for it. If we consider it is an industrial sector like any other, that is maybe coming from outside the country, then maybe they should pay for the grid. These are questions again in terms of public versus private and the social contract.

I think it relates to that same point. There has been a lot of talk not just about AI but more generally about infrastructural deficits whether that is housing, water or transport. You cannot have this discussion about data centres and energy infrastructure for data centres without putting it within that context. Until people feel that there is an effort to provide infrastructure for public or social good, that they can see that and that it is having a tangible benefit, then there will be growing resentment and grievances about the fact that there are infrastructural investments and planning for this particular sector, and that will put more strain on the social contract.

In 2019, I attended a meeting of an Oireachtas committee to talk about data centre growth. I remember making a similar point about there being a contradiction between climate action that was for general societal benefit and it being a burden on people as opposed to what was happening with data centres. The situation has probably accelerated since then in seven years.

As I said in my initial statement, it is a real issue that is beyond the remit of this committee exactly but it is also important in the context of discussing twin transition to at least be aware of that wider context.

I think it would be fair to say that AI has already become quite intrinsic not just in terms of academia but in people's personal lives, in business, etc. Dr. Brodie spoke about the geopolitical vulnerabilities, particularly of the likes of Anthropic, OpenAI, etc. In line with the conversation around lean AI, what are Dr. Brodie's thoughts on developing? Obviously we cannot compete. No sovereign AI can compete with the large Anthropics, OpenAIs and Microsoft Copilots of the world. Let us say there were a European lean AI-type of sovereign model that we could ensure had a more ethical perspective around energy usage, data usage and how it is developed to ensure that it is not biased towards socioeconomic backgrounds or countries. Does. Dr. Brodie or the other witnesses think that we, as policymakers, should examine in Ireland and, indeed, across the European Union?

That is a goal that I would agree with but underlying that goal would have to be a sense of a global justice perspective, around also resourcing requirements for this.

There has been some really interesting stuff written on the green industrial policies, which are about developing sovereignty in places in the global north while also respecting and allocating development funding for countries in the global south, for example, who are also trying to develop in the same way in terms of these recognised imbalances, especially in the context of countries in the global south providing minerals, manufacturing resources and things like this.

The principle of sovereign AI for Europe is something to aim towards but bearing in mind, one, what are the social uses and public benefits of this, and how that can be developed to not only respect but also develop sovereignty and environmental justice elsewhere and not just in Europe.

I am conscious about how achievable that is. If it was a public-private partnership between the European Union and, not an OpenAI or Anthropic but European companies that would be supported, would that still lend to the same thing or should it be done by the European Union hiring its own software developers, etc? From a realism point of view, should it be done in partnership or would that go against the idea of it being a fair and open system?

Yes, there are efforts in that direction of having those purely driven by the Horizon funding from Europe to do that. The other thing that is interesting and relevant to hear is that the frontier is moving fast, but what becomes quite accessible is catching up with that quickly. These smaller and more accessible models can perform at a level comparable to what was the frontier only a couple of years ago. There is a reasonable view that some of that acceleration will top out and we will catch up, and the gap between what is at the very frontier and what is much more accessible will narrow quite a bit. There are good opportunities there. Both of those directions are certainly technically very plausible that we could have those sovereign approaches.

On this talk about strategic autonomy in the EU, a point that Dr. Brodie already made, and it goes back to this idea of the stack, is that it is becoming apparent that you can develop different types of applications and even different types of models but they still sit on that underlying infrastructure, which, I think, 70%-----

Yes, 70% by Anthropic.

-----is provided by the three hyperscalers of AWS, Google and Microsoft. Delinking from that will be very difficult but is obviously necessary. That requires long-term planning and investment which cannot happen at the same time while there is industrial policy which is about encouraging increasing dependency on those same providers. There are hard decisions to be made. I do not think you can talk about strategic autonomy but still deepen dependency on those providers through their infrastructure. We need to get beyond tokenistic talk.

I find that there is a rhythm to these meetings that when we get towards the end, no matter who the witnesses are, we get into philosophy: good AI, bad AI, AI will be the fountain of all good and AI will destroy the world. I say that because Professor Keane, I think, made the very obvious remark that we do not need AI to decarbonise our country and yet it does feel like AI is driving the energy debate right now, as it is doing in so many other fields.

On that good data centre versus bad data centre vein of thought, does Professor Keane think that it is viable or feasible, in the absence of future data centre growth, for the five offshore wind energy planning applications that are currently live, two of which apparently they hope will be approved this year and three next year, to proceed in Ireland as distinct from just get planning? Do they need large energy customer users like data centres in order to get the same type of capital for the construction to complete?

They are absolutely viable without data centres. They are viable in and of themselves. They are separate issues. The country needs renewable energy to hit our 2030 climate targets. We will not hit 80% renewable electricity by 2030, but I see them as separate factors. We have enough energy demand in the country without further data centres. We do not need them to make those projects viable, in my view.

I would agree as well. We are still building interconnectors too. Interconnection just means we are importing electricity. We are building lots of infrastructure and we do not need the data centres. We have that demand there. We should be focusing more on how we can integrate that renewable electricity into our system so that we use less gas rather than whether we need more demand there to take up that renewable electricity.

Professor Mac Namee made a very salient observation about how we cannot separate the underlying technology, which for better or worse came from private innovation from which the public can benefit. I am sure there is a mixture of public and private in there somewhere. Of course, there always is an element of this. There has been a lot of discussion, particularly from Dr. Bresnihan, about our industrial policy and our dependency. When I hear those as I sit here, I think about the people who are working in the tech sector, the benefit it has been to Irish society and to families the length and breadth of this country who have roofs over their heads because they are working in these sectors, and the indirect effects of it. How do we separate the data energy dependency on the tech sector from Ireland's industrial policy as it relates to data centres, for example? Can we separate the two between the level of investment we are potentially going to attract from our data centre policy and the indirect benefits from our overall industrial policy for the tech sector? Can we delineate or is it like AI models where we cannot really delineate between the two? Is that a fair analogy?

In terms of the data centre demand and use for that, I am not sure if I could speak to that specifically. However, what is relevant in that is if we look in the AI world more broadly, when we talk to our colleagues in universities around Europe, we are the envy of a lot of those colleagues because of our access to the cutting edge of these techniques. The fact is we can bring that access and experience to our students and graduates within our universities. We have access to the very cutting edge of what is going on here. That is a real benefit we get from that. Like all these things, the balance then is how we take the advantage from that to push our indigenous local industry. There are good examples of that. People probably know Intercom, which is a local company that has been in the news this week. That is a direct benefit of that ecosystem our tech graduates and tech sector are working in that can then turn to building those really strong local companies as well. It is a real resource that we should be taking advantage of.

I thank Professor Mac Namee very much. I thank all the witnesses for a really useful, interesting, constructive and engaging debate, which I hope will assist us as we develop our next interim report from this hearing, the previous hearing and the subsequent ones to come. I thank everyone for giving their time and expertise to all of us.

I propose that we suspend the meeting for three minutes after which we will go into private session.