Global Crossing: How Ireland Invested in the Internet
"Their eyes were open when they took this"
Produced by Luke Fehily and Tom O’Connor, Memo to File examines how Irish public projects were executed in practice, focusing on implementation, lessons learned and the institutional memory often lost over time.
Ireland’s position as a global hub for technology companies is usually explained by corporation tax, an English-speaking workforce, and EU membership. Less often noted is that, but for a decision the Irish State made in June 1998 to spend €77 million as anchor customer on a transatlantic fibre-optic cable, almost none of those companies would be here. The Global Crossing deal — committing the State to a twenty-five-year indefeasible right of use on two submarine cables between Wexford and the United States — gave Ireland direct, internationally competitive bandwidth at a moment when the alternative was routing every bit through the UK at multiples of the price.
Brendan Tuohy was Secretary General of the Department of Communications, Marine and Natural Resources from 2000 to 2007, having served as Assistant Secretary in its predecessor from 1992. He held the telecoms brief through the Global Crossing process. He is now Chairperson of EirGrid, and also chairs MaREI, the Science Foundation Ireland centre for climate, energy and marine, and TILDA, the Irish Longitudinal Study on Ageing.
In this conversation — the third episode of Memo to File — Brendan walks us through how the deal came together against a 12-month deadline, why an Irish state acting as anchor purchaser was the only way the cable was going to get built, what was lost when Global Crossing itself entered Chapter 11 a year later, what the Public Accounts Committee made of the eventual write-down, and what the same Irish state apparatus would and wouldn’t be able to do today. We close on judgement, risk appetite, and a Seamus Heaney anecdote.
After the interview, we sat down with Dr Conor Igoe, an AI researcher at Edison Scientific in San Francisco who recently completed his PhD in the Machine Learning Department at Carnegie Mellon University, and who has written for The Fitzwilliam on Ireland’s AI research gap, to digest what the Global Crossing playbook would look like applied to compute infrastructure today.
This transcript was edited for readability and accessibility with the assistance of AI tools.
Set the scene. What was the telecoms landscape when Global Crossing came about in the 1990s?
We were looking at the whole telecoms picture and the Internet companies that were beginning to start in the US, and wondering with the IDA and the Department of Enterprise, Trade and Employment how we could attract those companies to Ireland. From a number of discussions over in the States with the IDA, and conversations with the companies, the message was clear: number one, you don’t have direct connectivity to the US or out of Ireland — you go to the UK; secondly, the cost of your international bandwidth is far too expensive. We took those as the two challenges.
In June 1998 the Minister set up the Advisory Committee on Telecoms, built around a number of people from the US who were involved in Ireland — Brian Thompson in particular, who became chair, and who had been on the Taoiseach’s North American Economic Advisory Board. We had Irish people from Forfás and the Department of Enterprise, and I was in the Department of Public Enterprise with the telecoms and communications brief. We had Vint Cerf, one of the fathers of the Internet. They started in June and reported in November.
The report had ten recommendations. One of the key ones was the need to develop international connectivity in a very cost-effective way — internationally competitive, number one or two in the world — and to provide choice of supplier, which we did not have. At the time we mainly had two telecom providers in Ireland, Telecom Éireann and Esat, and they were hugely expensive. The challenge was to set something up and do it very quickly because the industry was evolving very quickly.
Was the Advisory Committee following an existing model, or was it unique?
It was a unique model. We realised we didn’t have the experience in Ireland at that scale. The report itself is 24 pages — it’s not a magnum opus, but it’s very clear. There was background material on what was happening at the time. They encouraged us to adopt a readiness chart that was available in the States, to see where we were going and what we would have to do to make the country ready. It wasn’t overly prescriptive — these are the types of things you need to do.
In fairness to the government, even before the report was finished, from the conversations it was clear some of these things were going to come out in it, and we had the authority from the Minister and from government to begin work on them. We didn’t need to wait. The other point worth making is that not one of those people took a single penny for their contribution. They did this pro bono. They were delighted to see a country like Ireland willing to put its hand up. Ireland had a tremendous record in the States and very strong linkages, but these were people who wanted to do this for the good of the Irish State. I don’t think you would have got any consultants to write a report as good as they did, because the committee had credibility.
Were there international exemplars at the time, or was Ireland an early mover?
The whole industry was starting out. This was a fast-moving time. There was huge money at stake and Ireland needed to be in a position to be part of this future. We didn’t know what the outcome would be — we certainly didn’t expect the downturn — but we realised something was happening quickly.
The report was published in November 1998, and by June a joint memo to government came from the Minister for Public Enterprise and the Minister for Enterprise, Trade and Employment seeking approval for the Global Crossing deal. It was approved knowing the scale: within 12 months we wanted a fibre cable into Ireland, a fibre cable to the States, and connections to a number of European countries. Two submarine cables would come into two different locations in Wexford, and two Telehouse facilities in Dublin with a backup, then back into the States to seven cities, and subsequently over forty cities in Europe and Asia. Ireland would position itself as a hub.
Our hypothesis at the time was that demand for bandwidth was going to grow exponentially and the price per bit was going to head towards zero. It was a hypothesis — we had no way of proving it — but that came from discussions with the advisory committee and others in the sector. To put this in context, in the late 1990s if you had access to the Internet you were likely on a dial-up line at 28 kilobits per second. WWW stood for the World Wide Wait. You could load a website, go and have a cup of tea, and come back and it would be loaded. Part of the problem in Ireland was that you were paying per bit. The committee’s recommendation was to follow the American model and go for a flat rate, and ultimately the regulator was instructed to introduce a flat rate charge for the Internet.
Dozens applied to the Global Crossing tender; it was narrowed to four companies and Global Crossing won. The contract was awarded at the end of June 1998 and the submarine part of the work, including the Telehouse, was done by June the following year. There was a slight glitch and it was August before it was fully operational, in 1999.
The Bertie Ahern government had come in in 1997 and been handed a derogation from telecoms liberalisation — the right to delay full market opening from 1 January 1998 by five years. The government had originally said it would avail of that only to 1 January 2001, but industry and the Department were strongly saying we couldn’t afford to delay further; the companies here and those about to come here would not come unless they had choice of supplier. So in October 1998 the government agreed to bring liberalisation forward to 1 December 1998.
If you look at the telecom value chain — international global connectivity, national backbone, metropolitan area networks, and the local loop — there was real pressure to unbundle the local loop so that people had competition. That’s what you have today: even though Telecom Éireann, now Eircom, provides much of the physical infrastructure, other players have the right to access it and resell. That’s what the multinationals were saying they needed, and that’s what the government delivered in a very short space of time. Telecom Éireann floated in July 1999, and by March 2000 the Internet bubble had started to collapse. If we hadn’t built the cable at that stage, there was no way it was going to be built for the foreseeable future. All the capital had been pulled. Most of the telcos went into receivership or Chapter 11, as did Global Crossing itself. But Global Crossing was of strategic importance to the US — a lot of US government traffic ran on it — so they were not going to allow it go into liquidation.
The financial and technological logic is one thing. Politically, this was complex infrastructure with significant upfront cost and benefits years out. How was the political backing secured?
The context was that there was a whole move to liberalise the markets — telecoms first, then energy. That model of liberalisation did not apply to the international connectivity side. The government realised very quickly, and Irish ministers travel a lot and a lot of the multinationals based here talk to ministers, that if we wanted to stay relevant — to retain the companies here and attract new ones — we had to respond to their needs. That meant international connectivity at the right price, which subsequently became between 10 and 15 per cent of the prevailing market rate. We became number one or two in the OECD on the cost of international connectivity. It was cheaper to connect from Dublin to New York than from Dublin to Galway. That was a policy choice by government.
The investment was €77 million. It sounds like a lot — it was a lot — but it was a 25-year indefeasible right of use, effectively buying the capacity for 25 years, with options on top if we needed more. You can’t buy part of a cable; somebody had to build it. So Global Crossing offered to build, and as part of the deal they built it. The economics of bandwidth aren’t linear at the start — three STM-1s isn’t a quarter the cost of twelve — you build a critical mass and after that it gets more linear. So the upfront cost is large. Ministers were tuned in to what was happening at the technology and telecoms level, but in particular they recognised the importance of what was going on. Irish ministers travel abroad and the advisory committee met them more than once, and the companies in Ireland would meet ministers fairly frequently. So the message came through very strongly.
Lots of big moves taken in such a short period — the liberalisation, Global Crossing, the regulatory architecture. Was there a particular nimbleness in the Irish system at the time that allowed that, and how did it differ from today?
There were very good relationships across the system. It was a smaller, less complex government. Ministers knew each other. My predecessor as Secretary was John Loughrey, a very well-respected and supportive figure with a good business head. On the Enterprise side I dealt with Paul Haran, who was Secretary General of Enterprise, Trade and Employment, and his Minister was the Tánaiste, Mary Harney. Charlie McCreevy was Minister for Finance; David Doyle and others were very supportive too. Officials met frequently, and we had an interdepartmental committee from 1998. There was a lot of trust there. We were in new territory, but it sort of worked. The proof was in the pudding — Google and Amazon and others came to Ireland in 2003 to 2005, and the data centre sector developed off the back of it, building expertise that today contributes between €3 billion and €4 billion a year in Irish engineering services exports.
Going back further, there were very irregular meetings of Secretaries General before that period. But certainly by my time there was a regular Wednesday meeting where the Secretaries were briefed by the Secretary to the Government after the Tuesday cabinet. That allowed for follow-up — it wasn’t an alternative government, it was implementing government decisions — but even people meeting like that on a regular basis was really important. Out of that you got Assistant Secretary groups, ministerial committees, senior officials groups, and so on. Today they’re more structured and more formal and you can have a lot of people at those meetings. It was probably less formal then, but there were meetings and committees, and the great thing was the ability to put the key decision-makers around a table and make decisions quickly. There’s no point getting them around a table if you don’t make the decisions.
Pivoting to challenges. The main risk that was raised at the time was insufficient demand. How did the structures handle that?
We brought in legal and financial and other expertise from the start. As Global Crossing went into Chapter 11 we needed American expertise on the case as well, so those fees went up. The Comptroller and Auditor General criticised that as having exceeded the original amount; we could not have planned for the collapse of the global economy on this.
We had two issues here. There was the strategic issue, which even the C&AG said was achieved — providing the cable to the States, at the right price, putting us at number one or two in the OECD. That allowed the other things to happen. You can’t do this incrementally; we couldn’t buy a few STM-1s, somebody had to commit to enough capacity to make the build of the two submarine cables economic.
There was a company called Hibernia that pulled out of the process because they wanted to do their own cable — which they did subsequently between the US and Ireland — but they also wanted to play in the market here. We would not allow Global Crossing be a participant in the Irish market. They provided the infrastructure, and that infrastructure was then shared out competitively to others. There are consequences to those decisions, but they were the right calls. We also put a very strict twelve-month timeline on Global Crossing, and they effectively delivered on it — with the slight glitch by a month that wasn’t their fault — and in addition they brought in the Telehouse facility.
If we had sold everything in advance of the deal — we had commitments from different companies, who put in €50,000 each to be part of it — we had more than the 160 STM-1s we were buying. Then the economy collapsed from March and particularly October 2000. Investment was sucked out of telecoms. In our case, even though both Global Crossing and the Irish subsidiary went into Chapter 11, we got the cable built. It was up and running and it stayed going. Just about everything that could go wrong externally did go wrong, and the system still delivered.
Chapter 11 is reorganisation rather than receivership in the US sense. On the economic downturn, the IDA wrote down the value of the unsold capacity from €23 million to €2 million. What was the deliberative process around that?
That was an accounting process. If you’re sitting on assets and you do a market valuation at a time when the whole global market has collapsed, you write down. But you still had the assets at the end of it. You still had the cables. The C&AG line was drawn in 2002, with a subsequent report in 2005, but if you look at the IDA accounts much later than that — right up to the 2012 accounts — they continued to get revenue from the cables. They were still being sold and used.
The way we structured it: we purchased them. We never wanted to make money — we wanted to sell them on at the price we’d paid. But the government took 16 of the STM-1s out for the use of the Higher Education Authority, and that had a huge positive impact on third- and fourth-level research, because Irish researchers were then directly connected into the core of the Internet in the States. If you look at the level of publications coming out of Ireland at the time, it was disproportionately higher than peers in Europe. Subsequently HEAnet, the network for higher education institutions, was asked to run the schools network too. Today Ireland has every school connected to a national system run by HEAnet — from 10 megabits per second to secondary schools at the start, to over a gigabit to the bigger schools now.
The role of government in a liberalised market — particularly a network market — was something we were thinking through at the time. If you want to build a piece of infrastructure to a town or a metropolitan area, you have to build it upfront. You can’t ask the first customer to pay for it. Somebody has to take that risk. With Global Crossing, the Irish government took the risk, with commitments from companies to buy it. A lot of those companies couldn’t ultimately buy because they themselves went to the wall.
The Department’s own commissioned review at the time concluded the project’s objectives were achieved, but flagged that risk-sharing with companies, rather than the State holding the risk, might have been preferable. Do you think a risk-sharing structure like that would be the right call now?
Hindsight is 20/20. At the time we were trading off the speed of getting this done. We had discussions with the companies and got commitments from them; we got 160 STM-1s of commitment when you take out the 16 reserved for the Higher Education Authority. But when it came to going after them, the companies went to the wall themselves. It’s always better if you can load the risk to others, but to do this there was a risk required and the government was willing to take it.
Their eyes were open when they took this. They knew there was a risk. It was all laid out. When the memos go to government, these issues are outlined very clearly, and that’s when leadership is called for — you decide, look, we’re still going to do this in the context of what we think is really significant and important. The counterfactual, if we hadn’t done it, is that you would not have seen a lot of the technology companies here today. Once they came they were established, and we and the IDA could work with them and build their presence here. If they had gone to some other country in Europe, we would not have had that.
So you make a call. You bring on advisors and you get advice. There are extraneous factors around timeline. None of the companies would have had the capital to do this themselves, and if they had they wouldn’t want to sell it on — they’d want to be a monopoly. The companies on the other side, that we were trying to attract in, did not want monopolies. So the government acted as a large-scale bulk purchaser, got a really good deal, and said: we’re not going to make money on this, we’ll cover our costs, take out the strategic element, and resell on that basis.
The review process — you appeared at the Public Accounts Committee multiple times; one transcript ran to 92 pages. Was that review process beneficial in practical terms?
In anything, you’ve got to be accountable. When you start out on a project like this, you know there’s going to be accountability. What you have no idea of is what the future is going to look like. We did not expect a global downturn that wiped out 80 per cent of the NASDAQ in 18 months and wiped out the telecom sector. Nobody did.
The system stood up. There was no criticism of the process. The C&AG report said it could have been better if you had managed to outsource the payment — if you had got the private sector to pay for the cables in advance. Of course it would have been better, but despite the efforts we made to do that, we had a very strict timeline, and the companies we were dealing with didn’t see the need for it because it wasn’t in their strategic interest. The State saw it as in its strategic interest, and the Public Accounts Committee were very clear that the strategic interest was met. They almost said this was a risk worth taking, and that’s probably as good as it gets from the Public Accounts Committee.
As Secretary General you do PAC every year. Your accounts are published and you appear during the year. Each department is zoned in. You do a lot of homework. Every Secretary General takes it seriously. The accounting officer concept predates the State — it goes back to British legislation — and the whole idea is a split between the Minister as the political entity, the corporation sole, and the Secretary General who is accountable for how the money is spent. There’s a whole set of rules around this, the public accounting procedures. At the Public Accounts Committee, the Secretary General is not allowed to comment on policy — that’s a matter for the Minister. The Secretary General comments on how the money was spent and whether procedure was followed.
In more recent times, EU procurement rules have added cost. In Ireland, where you have a strong economy and a huge demand for building, a lot of builders had been doing contracts in the private sector because they were less onerous. Many didn’t want public sector business and put in very high tenders specifically to not be selected — and sometimes those tenders got accepted because they were the only ones left. The procurement rules themselves, even though they were set up for the right reasons, can result in quite additional costs. And in Ireland we use a cash-based accounting system rather than accruals, so pension costs and the like sit outside the departmental accounts in central accounting, which raises long-term questions about how you fund the public service as the dependency ratio changes.
How does that climate affect decision-making in practical terms today?
Budgets are allocated earlier in the year now — by October you know what the budget is for the following year. There was a period when I was there of multi-annual budgeting — your administrative budget covered three years, which gave you certainty and meant everything didn’t have to be spent this year. With a cash-based system you get drawn into spending in the year in which you get the approvals, because your budget for the following year is often built on what you spent last year. That doesn’t always lead to good behaviour, because you expedite things to spend the money.
There have been major moves more recently to link the tendering process to the final payments and make it transparent. Deputy Albert Dolan has been driving an initiative to make it clearer what is being paid for and what is being received. With the data now available, in the right formats, the public themselves can do the analysis — it doesn’t have to be civil servants or the C&AG. There’s more interest now at all levels in how money is spent. It used to be more of a black box.
There’s no reason much of this isn’t available continuously — we see dashboards now from companies and public entities, right down to the buses where you can see if they’re going to arrive or not. EirGrid has a dashboard for energy use and sources in real time. Moving towards data being available and transparent — transparency is probably the greatest antidote to corruption. I’m not saying there’s corruption going on. It helps people know that the system is transparent.
Are there elements of the Global Crossing model that you could see working for other infrastructure projects today?
It’s horses for courses. You have to decide what you’re trying to do and what your criteria are. If you have plenty of time, you’ll do everything well, you’ll always do everything above board. Under a timeline things may have to take shortcuts and get approvals to do that. Global Crossing didn’t take shortcuts, but it determined that speed of delivery was a key tender criterion, and that was specified up front.
You saw the same dynamic across Europe during the Ukraine crisis, particularly in energy, where countries had to replace Russian oil and gas very quickly. You saw it during Covid. Governments have to be nimble, and in my view they have done very well in those moments. Government can have an ability to pivot, and sometimes performs better under pressure than in ordinary time, because people get in behind it and want it to work.
Government always has to balance conflicting advice and conflicting pressures. Most memos to government, when they’re circulated to departments before they go up, get an input and a view from each department. So when government goes to take a decision, it has the conflicting views and a response to each of them, and it makes the call with the best available information. You may not agree with the decision, but it has been made knowing generally what is knowable.
Can the decision process be expedited more widely, or is it inherent to the system that it can’t?
Juncker, the former head of the Commission, said: we know what to do, we just don’t know how to get elected after it. Part of the challenge is that people know what the issues are, but sometimes the electorate or the public don’t particularly want the solutions. So government weighs the balances all the time. From the outside it might look like, why did they take so long? Why did they make that decision? They probably were aware of what you’re aware of, but they were also aware of something else.
The further up an organisation you go, the more important the skill of judgement becomes. That’s what you get judged on. It’s not your detailed knowledge — it’s the judgement call you make.
Has the climate in which those judgement calls have to be made evolved significantly since you were in that position?
I’m nearly 20 years out of it now. Social media and the more open and detailed knowledge of stuff across the media has changed things. FOI didn’t exist at the start of my time but by the end it was very much part of the system. Today there’s even more — President Trump recently sharing text messages from other presidents, which would have been unheard of years ago. So you don’t know what you can say in public and not say in public. At senior level between ministers and governments, they’ve got to be able to talk. And you see now, with the decision not to renew the nuclear agreements between the US and Russia, that it appears there are no back conversations going on anymore. That’s not good for society. Even at the worst of times — go back to the Cuban missile crisis — there were back channels. There must always be back channels, even with your enemies.
Diplomacy is probably changing. People are more nervous of saying things now. Recently the Slovak prime minister, Fico, came back to the European camp after a meeting with President Trump and said Trump didn’t look great. He’s now denying he said it, but according to the media he did say it. You have to be able to have honest and open conversations. So it’s not just about the structures, it’s about the informal structures, but it’s about trust. And the trust element has probably gone down dramatically. A lot of countries now have a challenge of trust in the US administration at the moment, and that’s not good for the world.
Trust is needed at the micro level too. If it fails at the top, it becomes more challenging at local level — if there’s no trust at the top, it’s really difficult for it to filter down, because trust is no longer a core value. Whereas if there is trust at the top, officials know it and that filters down too.
Has the nature of coalition government changed during your career?
Coalitions came in when Dessie O’Malley and the Progressive Democrats joined with Fianna Fáil under Haughey, and that was followed by the Fine Gael, Labour and Democratic Left government. What that brought with it was professional programme managers, and that brought another system where programme managers dealt with a lot of the issues before they went to government. At the start there was concern that some ministers or civil servants might not have liked them, but a modus vivendi operated very quickly. People realised the programme managers were not going to take over — they had no role in managing the civil service. They had a role to advise the Minister and were accountable to their Minister.
In my experience that worked really well. Some of the advisors were very good, and it helped civil servants because ministers could tease out ideas through the advisors and have them talking to civil servants, and civil servants could float ideas to ministers through those channels. But you never lost the relationship with the Minister. The advisor was not the decision-maker. The Minister was the corporation sole. The Minister, as the entity, was who you had the relationship with.
What impact did the cable have on Ireland’s tech sector and FDI capacity?
Global Crossing was the catalyst for subsequent cables, but more importantly it gave confidence to the industry to come here. The Irish government had a good reputation: if you dealt with the Irish government, it stuck to its word. Traditionally, across all parties, that has never been an issue — on things like the corporation tax and the commitments generally, this is not party political. The foreign direct investment community trusted the government.
There was a time when we talked about Ireland as a politically stable country and everyone took that for granted. They don’t take that for granted anymore. So stability has become a key element. There’s a stability in Ireland that is consistent across all governments — policy doesn’t shift overnight the way it has in the States or other countries. That’s important for certainty. What businesses like is certainty: they can price in a policy if they have one. They cannot price in uncertainty, and the markets hate uncertainty. Most of our governments have been five-year governments, and increasingly they have been coalitions. Coalition is probably going to be the model for the future, given the breakdown of party support numbers.
A question we ask every guest — do you think there is an institutional memory problem in the Irish state?
There’s an institutional memory problem in most organisations, not just the Irish state. Part of it is the pace at which things are happening. Part of it is that one of the downsides of FOI is that less is being written today than would have been written in the past. If you look back at older files released under the 20-year rule, the file notes and notes to file are amazing documents — beautifully written, comprehensive, detailed. That probably changed with FOI, because people were afraid material could be embarrassing later.
The second thing is there’s been a policy of moving people every three years or so. The upside is fresh ideas. The downside is that in technical or detailed areas that need expertise, you won’t build that in three years. There’s been a traditional split in government between the administrative side and the professional and technical side, and that has broken down — people can move into senior positions in the civil and public service — but skills still need to be rewarded. In companies, people can spend their lives being technical experts and still be rewarded without having to be promoted, but that really doesn’t apply in the public service. The only way to move up is into management and into a more generalist area.
Henry Mintzberg, the management thinker, wrote a famous piece called What the Private Sector Can Learn from the Public Sector about the complexity of managing the public sector. This is a more complex environment. It’s not always about short-termism. Government is in the long-term business — the intergenerational business — doing things for one hundred years from now, not today.
What advice would you give someone earlier in their career hoping to work in the public service, or at a senior level today?
Stay open. Keep your mind open — you’re not going to learn everything in any one job. Young people being born today are going to live, on average, beyond a hundred. They are going to have multi-skilling, multi-occupations. You might spend a number of years in the public service, then go out to something else and come back.
You’re in a marathon, not a sprint. Don’t be afraid to take risks and move outside your comfort zone. There’s a real danger of getting afraid of taking risks, and if we do that as a country we will suffer.
Build relationships and networks across the system. One of the downsides of remote working in the public service is the loss of connection, which has a more general impact — the rise in loneliness among young people, in part because they’re not engaging, not socialising, losing some of those skills. We saw a 20 per cent drop in alcohol use in Ireland in the last 12 months, which lots of people would think is great, but there may also be a drop in socialisation skills.
A friend of mine was filming Seamus Heaney once, and he wasn’t in good form. Heaney looked at him and said: when I look down each morning when I get up, I see there’s nothing beneath me. And then I look up and I know there’s nothing beyond me. And in between those two places I find the motive force to live each day. The young people in Ireland today believe there’s nothing beyond them. I don’t think that was the way 30, 40, 50 years ago among most young people. We live in a good country and we will continue to live in a good country because we have the people. Our job is to provide the challenge but at the same time provide the support.
Episode Digest with Conor Igoe
After speaking with Brendan, we sat down with Dr Conor Igoe, an AI Research Scientist at Edison Scientific in San Francisco. Conor graduated with a Bachelor of Engineering from University College Dublin before completing a PhD at the Machine Learning Department at Carnegie Mellon University. He recently wrote Ireland’s AI Research Gap for The Fitzwilliam. We wanted his read on what the Global Crossing story tells us about Irish infrastructure decisions at the frontier today.
Your immediate takeaway was that Global Crossing is a good example of strategic foresight delivering a compelling piece of infrastructure — but that today the equivalent decision looks different.
It’s an interesting example of how the Irish government can provide strategic foresight and actually deliver a really compelling infrastructure project that then serves the Irish community for subsequent decades. In today’s world that looks a little different. We’re no longer bottlenecked by the communication layer. It’s now the compute layer that is the strategic resource. So it makes me think a lot more about what the coming decades would look like for Irish compute infrastructure as opposed to communication infrastructure.
I was impressed and also frustrated when hearing Brendan’s own account of how things have shifted to be more risk averse. That tracks with what I’ve seen in the commitments Ireland is making for the type of compute infrastructure it’s looking to build over the coming years. It seems like it’s not really drafted with the same headspace that you’d have had when thinking about Global Crossing.
That headspace in the late 1990s came partly from the dot-com boom. Does anything analogous exist today, even in an altered form?
Things are moving hugely fast in the AI world. In that sense there’s even more urgency. The type of work you need to do today to be relevant at the frontier of AI research is totally different to what it looked like five years ago. A lot of that is just governed by resources — access to compute and to data. The speed at which that is evolving is striking.
What would sovereign compute actually unlock for Irish researchers?
The main thing is that it facilitates engagement with the types of models being developed around the world — open-weight large language models, the things touching a lot of different software ecosystems for different startups. These models have their own weirdness that needs scrutiny, and to do that scrutiny you need to have the GPUs, you need to have the compute infrastructure to be able to run them and start inspecting them. Ireland at the moment doesn’t have nearly enough compute to do that.
There are really important research questions there: what implicit biases are these models coming along with, how do they make specific decisions in this way or that way. Those questions are hard to study unless you have the compute to load these models and tinker with them. Without an infrastructure base to do that, you incentivise researchers to think about mathematical analysis, or more reflective pieces on ethics or regulation — all of which is important. But you have a gap of actually engaging with the technology and being deeply involved in co-creating it.
One of the most compelling applications I’ve seen recently is a project done in Kentucky in 2025 with Google. They used language models to facilitate a really large town hall — thousands of participants discussing their concerns and hopes for the town, with the models consolidating it into a co-created document of concerns. That would be impossible to do with a traditional town hall: getting everyone there is already hard, getting everyone to say what they want simultaneously is chaotic. But there is a valid concern that Google, or a private tech company, may not be the most appropriate entity to be orchestrating that process. It’s a great proof of concept but the type of thing best suited to a research institution — publicly funded, open, transparent. That’s the sort of thing Irish research institutions would be challenged to engage with at the moment, because they would pay a bigger premium for the infrastructure to do it.
How big would the investment need to be, compared to Global Crossing?
The basic unit of compute in AI research these days is the H100, an Nvidia GPU. For interesting frontier research you need to be on the order of hundreds of thousands of these chips. If you look at what’s available to Irish researcher at the moment, it’s a handful — maybe thirty or forty domestically, plus on the order of a hundred rented through international arrangements. There’s nothing at home that we really have.
Ireland does have a plan for more national high-performance compute. It’s called CASPIR. At the moment it looks like about €25 million of investment, walked back from something floated before at around €60 million. The details are a bit frustrating in that when you look at what €25 million would actually give us, it puts us at the lower end of where we stand with other peer EU nations. We are doing something — but not really at a level that unlocks the kind of value I’d be excited to see Ireland embracing. In that sense it’s a little different from Global Crossing, which was a real strong commitment to making it work and having it be lasting infrastructure.
You set the comparative funding in your Fitzwilliam piece — Ireland’s commitment is small not just versus other small EU nations, but globally. Why does it matter?
I think this research-output gap will only get worse if we stay on that path. It’s hard to imagine a really ambitious Irish research startup emerging from the current research ecosystem within Irish universities, in AI research, because they’re just not exposed to the types of infrastructure you need to make those projects happen. When you go to research institutions with an abundance of compute, the types of questions people start asking, the types of things people start wanting to build, the types of concerns people start operationalising are totally different.
When you look at how the big frontier research labs source their talent, they overwhelmingly source from environments that have access to this sort of compute. So their talent base has familiarity with how to train and explore these models at a deeper technical level. Insofar as those positions matter for Irish representation in AI research, we’re not setting up our talent pool to be able to engage with the frontier. Long-term it’s not obvious what the Irish representation would look like in these labs in five or ten years, and those labs are extraordinarily powerful — they are shaping a lot of things in the world.
Beyond getting the commitment, are there significant practical challenges to standing up this infrastructure in Ireland?
The most salient one is the energy usage associated with compute infrastructure. I don’t think what I’m talking about is on the scale of commercial-level high-performance compute centres. But there are challenges in Ireland with excess grid strain and the price of energy is already high. Those two make it a more difficult endeavour.
Final question. If a policymaker was listening to this, what do you want them to do differently on Monday morning?
Take seriously the idea that compute infrastructure is not just a nice-to-have. It is rapidly becoming a strategic resource — and that is not really well acknowledged at home, or discussed in a deeply engaged way with what the technicalities look like. Engage with that discourse. Think about what Ireland would look like in ten or twenty years with a much more vibrant compute research ecosystem. And think more carefully about what this compute research infrastructure really is. It’s not just a budget line item.
This is the third episode of Memo to File, an Irish policy podcast produced by Luke Fehily and Tom O’Connor. The next episode, with economist Colm McCarthy — former chair of An Bord Snip and An Bord Snip Nua — drops next Thursday. Theme music is from Epidemic Sound. Archive audio is from RTÉ Archives. Recorded at Ballyogan Library.