Posted inForum Playback

How to build bankable AI infrastructure in Egypt

Nick Edwards [N.E]: We are now joined by experts in the building, the buying, the deploying and the financing of infrastructure and technology. And we're going to spend the next 25 minutes on these four themes. So between them, our panel might just have the blueprint for how we build the future of AI infrastructure here in Egypt.

Mohamed, let me start with you. It is your company that builds what we're talking about. What are the principal barriers you face as you do so?

Mohamed Magdy Allam [M.A]: To answer your question, allow me first to attempt to define what AI infrastructure is. 

Think of HSBC. It has a massive data set that needs to be analyzed. And this is where you use GPUs to train and run models, to provide useful information to reach the root cause of a certain problem and find solutions.

And theoretically, each institution can procure their own chips, build their own data centers. But practically, this is not very efficient. It burdens balance sheets. It does not work operationally because of the skill set required at scale. So a company such as ourselves, as a third-party infrastructure provider — this is where we step in.

We invest, we build, we operate data centers that accommodate these GPUs at scale, and we're able to provide this compute not just for one institution but for several. And this way, we democratize the compute and we avail it at very efficient and competitive capacities. If this becomes the definition of AI infrastructure as I see it, then back to your question. What are the barriers? 

First of all, we need to build it at scale for this to become meaningful and economic. What we're seeing [as] meaningful scale now is, let's say, a 20-megawatt IT load data center that houses these GPUs. We're speaking anywhere between USD 600 million and even more, depending on the technology used.But that's just as a starting point. 

And to have a product to actually offer customers, you need to have the land, you need to have the building, you need to have all the equipment there. To do so, you need the financing. To get the financing in place, you first need the customers — you need the agreements. To get the customers, you go back, and you need the building, the land and the product. 

And that's a circular situation where you need to sort of break it one way or the other. And that's one of the foremost barriers, not just in Egypt, but around the world.

In Egypt specifically, the market is very nascent. So then there's another layer: we have very limited precedent, not just for AI infrastructure as we discussed it, but [for] data centers. It's a very small market. How do you penetrate this market and how do you scale it? 

And I think this is where [we come in]. And just to clarify, we are a partnership between Hassan Allam and A15. And we believe we have that capability between infrastructure delivery and, specifically, digital infrastructure.

Through our partners, we have delivered 100 megawatts internationally between Europe and Asia. And we're very optimistic about the opportunity here in Egypt.

N.E: Well, great that you're optimistic despite the barriers that you've just outlined. Let me ask the rest of the panel. Does this sound familiar to you? Lamyaa, does what Mohamed has spoken about resonate with you?

Lamyaa Gadelhak [L.G]: Yes. Well, the complexity of a data center is… it stands out. There was a time when a renewable energy company or a power project was a project on its own, and it had its own magnitude. But now, this is just one very important component, but one component of a larger infrastructure and technology piece.

So I certainly agree.

N.E: So complexity, you recognize that as a problem. 

Ahmed, is that something that you're familiar with? Do you think those three things that have been outlined are the key barriers?

Ahmed Sobhy [A.S]: I think demand is probably the biggest barrier, to be transparent. I think that we do a lot of hyping of AI. It's important to realize that Egyptian customers haven't gotten there completely yet, for a number of reasons, which we can talk about. But we're not there in terms of significant demand for AI, at least in Egypt.

I agree with Mohamed. For sure, there's an opportunity outside of Egypt, and if you build the right data centers, that compute can be used. But in terms of significant compute in Egypt, I think demand is probably a big barrier.

N.E: Okay. So, customer demand. 

Maged, what do you think? Are those three things that you've heard there, about the buildings, the power, the chips, the complexity — is that something that you're coming across in the projects that you see?

Maged Ezzeldin [M.E]: Yes. I think in general, organizations come to advisors like ourselves with a business idea, right? They want to seek validation. Some of them think they got it right; they just want a rubber stamp for a bankable document. And then the advisor will look at the assumptions and challenge them one by one.

And the way I see it, it's kind of [a] formula, right? I think Mohamed Allam mentioned a lot of the components, but it's a formula where there is no one magic solution, or at least people did not discover the secret sauce in Egypt. As Mohamed said, the formula has a long list of factors. There is the land. There is the connectivity. There is the energy. There is the talent. There is water cooling, [which has] not been mentioned as yet. There is also the supply chain, the policy and regulation. And that equation equals data center and AI infrastructure.

In some cases, you can have the land, but not the power, right? You have the power, but not the demand. You have the demand, but not the finance. You need all of the above, and you need to look at it. And don't get me wrong, this is not a real estate project only, and not a technology project only. So it's like marrying both of them.

N.E: Okay, so getting the definitions right of what it is that you're financing adds another degree of complexity to what building an AI data center looks like, and AI infrastructure more broadly. 

You've articulated the problems. How do you begin to solve for them?

M.A: I think, as Maged said, it's a very complex situation where you have a number of ingredients that need to be sequenced in a certain way to eliminate uncertainty for the various stakeholders. So on the demand side, yes, I agree with Ahmed. Demand in Egypt has still not caught up with where the world is, but that's kind of also where the opportunity is for Egypt specifically, because if you look at the US or Europe today, to get access to infrastructure, it's a very constrained market.

Yet the amount of companies — and I think a robotics company was mentioned — the amount of startups that require compute is phenomenal. We are seeing different statistics [on] demand, but the last statistic I've seen, it's six to one. So today, with the situation in the world, Egypt is very well positioned. As Maged said, we have the connectivity, we have the power, we have the capabilities locally, albeit perhaps still limited, but we can do it, and we can build this infrastructure to serve the world. So that's one: you get the demand. 

Two, on the financing side, I think, given our experience — we've developed three gigawatts of renewables. We've developed logistics parks, water desalination assets. Our partners actually invested, built and operated data centers in Italy and Thailand. So we understand how to sequence each one of those, how to structure, whether on the debt or equity side, an opportunity that is bankable.

So the way we see it is that we're splitting the hard infrastructure from the GPUs, because with the GPUs, you have a lifetime of, let's say, five years; other assets, potentially ten. So those are two different sets of financing structures. And if you reverse engineer [from] the demand for the financing, and then sequence capital at risk to secure the land, building, equipment, construction, all the way to operation, then it's a complicated formula, but ultimately it works.

There's no one-size-fits-all, but we believe [we] have the right formula for Egypt.

N.E: Okay, well, you mentioned the financing. That's a nice segue into: how do we get this stuff financed? How do we make it bankable? 

Lamyaa, this is your area of specialty. You're global head of project finance at Baker McKenzie. Is what we're seeing in Egypt radically different to everything else?

L.G: Well, when it comes to project finance, this data center is not necessarily [financed] the way you do typical project finance. You would borrow a lot of knowledge from other types of financing. But I think the beauty of it, given the complexity, is that you can cut it down — on Maged’s point, he said it's an infrastructure project and an IT project marrying together.

But if that doesn't work, you can always deal with them separately and fund them separately. So rather than reinventing the wheel when it comes to financing, I think what we need to do for Egypt, if there is the appetite of course from the lenders, is to look at how they were funded elsewhere.

And I think it's worth taking a step back: why are we here having that discussion now? Data centers are not new, okay. Data centers have existed for some time. And the reason for that is that the hyperscalers — I'm talking about that specific model because this is the biggest model — have been developing their own data centers.

But with AI, with increasing demand and increasing investment cost — because also the chip technology, the more evolved it becomes, the more it consumes power. So far, we haven't reached a point where the technological advancement has reduced the need for power.

So the investment costs have increased incredibly, and the magnitude of a data center has been very heavy, even on the balance sheets of those hyperscalers. And the reality is they are not infrastructure companies. They are software companies, and they shouldn't be valued as infrastructure companies. They don't want the infrastructure on their balance sheets.

And then it started with them pushing the infrastructure piece to infrastructure companies, and that would be financed separately as real estate financing. And then they still take on the financing for the technology. 

But we're seeing, in the US, in Asia and Europe, different models of data centers exist. And so the financing piece, or the financial architecture of how you fund it, follows how you structure it.

N.E: Okay. So at the moment, there's no one model that stands out. Europe is not having more success at financing this, or doing it in a markedly different way to the US—

L.G: There's not even one model that is typical to a geography. It's the debt structure that they use for a model. So you have platform financing. This is the model when the sponsors have the credentials, and the banks would take their risk because they have the ability, through their platform, to develop multiple data centers across multiple geographies. So they fund on the platform.

In the US, for Meta, it was an SPV. It was bundled in an SPV between Meta and a private credit firm, and then this SPV issued corporate bonds. But [in] that SPV, the sponsors had investment-grade ratings.

N.E: So different approaches, different horses for different courses. Fundamentally, though, demand has got to be there. 

Ahmed, how do you see demand? I mean, you're potentially key to this, with your plans to build out.

A.S: I'm here representing, I think, the regulated industries. Todd [Wilcox] was talking about it, and I think Khalil [Elbawab] was talking about it earlier. What we do is government and payments, and both require sovereign infrastructure. So this has to be in Egypt. You have to be able to deploy it here. Two is very low latency, really high-quality connectivity. So that's very important, and great uptimes. And I think the third one that's very difficult — and that's why building the bankable side is a big question — is it needs to be in pounds.

Our business is in Egyptian pounds. Those financial institutions' businesses, [are] in Egyptian pounds. So you need to be able to find a way for that infrastructure to be in pounds. And I think those three things, if you can kind of put them together — I agree with Mohamed — then if you can build it in one place, you will have the demand for it.

But I think as long as regulated industries today in Egypt have to work on-prem [on-premises], so each of them builds their own, you're never going to get to the economies of scale where you can deploy AI actually at scale, and at prices that would help any corporate actually deploy AI within its business. And so I think one very important enabler for demand is government and regulated industries providing commitment for compute, and saying it's okay to be on the cloud and it's okay to use the data center, as long as it's in Egypt.

And obviously, the models need to be in Egypt as well. So just to be clear, it's not that data stays here but you go and train it abroad. It still needs to be here. And so for that, you need a significant amount of compute, and to share that compute is the only way this will make economic sense.

So I think that's a very important part of the demand question. We see it coming, and we're having discussions with different parties. But it is something that needs to be accelerated for you to be able to get significant bankable AI infrastructure in Egypt. And what's important about this — and it's important to think about — is that those industries will subsidize, in a way, the rest of the market, so you can get a lot more innovative companies. It has strong spillover effects. So it needs to be thought about that way, rather than, "We're the only ones on this thing." Without them, you're not going to get a lot of AI infrastructure for the rest of Egypt.

N.E: And are people willing to share? You talk about, "We need to be in this together." Are people willing to be in it together, do you think?

A.S: Yeah, I think so. I think it's in terms of how you build the environment. It can be your own separate environment and everything, but the GPU — you don't have to have your own GPU. It doesn't matter to you. What matters to you is that the process that takes place on that GPU happens fast enough and gives you the right results, but your model, your data, will be your own.

You can compartmentalize that. I've heard of certain financial institutions [that] have gone out and bought GPUs. So if every financial institution only buys ten GPUs, that's not the same as one data center having 5,000 GPUs, because those financial institutions can actually use 100 GPUs at a time when they need to, and then ramp it down when they don't.

And I think that's the key about sharing the infrastructure. It's not about sharing the data. It's about sharing the infrastructure.

N.E: Okay. Yes, Lamya?

L.G: Just to follow up, if I may, on that point, just to give you a bit of light: actually, in certain geographies now, GPU financing. So separate from the data center financing. It has different terms, a tenor of five years, approximately. Which just confirms that it could be dealt with as its own business within the shell.

N.E: So, thinking that we can look at best practice from elsewhere and then bring it here with the right deployment, that's an option.

Maged, you wanted to come in, I think. Go ahead.

M.E: I'll start with my friend Sobhy. I like what he said about localization. I think it's an important point to be on the national agenda, if you like. But it's important to be realistic about which part of the value chain can be localized, right? We're not going to do chips — maybe, I don't know, the GPUs. We need to know what we [can do]: maybe the renewable power that could power it. We can do solar and wind and stuff like that. So it's important to consider where Egypt has the right to win, and I will come to that.

On what Mohamed Allam was saying, I like the word "certainty" that you said. We do models all the time, and certainty, I would say, is the cheapest capital, if you like. Because when we do a model, it's all about cash flows and discount rate, right? And the more risk you have that you cannot mitigate — I'll just give you an example that was mentioned here: a lot of parts are imported, and [there is] foreign exchange exposure that you cannot naturally hedge by exporting services [as] a regional hub for data centers.

If you don't have this natural hedge and you put it in the risk, the more risk you have, it will affect the cost of capital of the project. And of course, it affects the internal rate of return, whether the project is economic or not. 

On Lamyaa, you touched on the financing. It's very important, the way I see it — and I think it was briefly mentioned — I see it as a local currency loan or financing component, and a foreign currency component. And the sources can be different, right? If you need Egyptian pounds, you can go to local banks and have it for civil works and so on. If you need dollars, there also you can consider [an] export credit agency, consider DFIs [development finance institutions], even the vendor. And it's important to match the cash flow, the sources and uses of funds. Don't get dollars [when] all your revenues are in EGP. That's classical.

Also, Lamyaa, as you mentioned the tenor, and you took the GPU financing as well — again, I fully agree. I think we're looking at three different clocks. We're looking at the long term, where it's the land and the building and the infrastructure. We're looking at the medium term: the UPS [uninterruptible power supply], the cooling and network equipment. And we're also looking at the short term, which is the GPUs that were mentioned. Accelerators and high-performance technology increase rapidly. So that's something. And mismatching between financing and using the funds is, again, classical.

So we have two, the currency and the tenor, that need to be done right.

N.E: So nothing new here, fundamentally, is what we're saying. So that brings us back neatly to the point that you made, Mohamed, right at the open. This is a cycle. It's a cycle that we're familiar with. We need to break the cycle. 

What's the catalyst that would break the cycle? What's the one thing that is most important to bring about the change, so that we open the door to get bankable AI infrastructure here? One thought from each of you, please.

M.A: So I think, as Ahmed mentioned, demand is key. We're seeing global demand. And that demand, once we secure it at scale for the global stage, we can scale our infrastructure to serve that demand, but also avail the infrastructure for local demand as well. And with that scale, it becomes economical for the local component of it. 

On the back of demand, everything else is solvable. We've done it before. We think we can do it again for that sector.

N.E: Okay. Demand, number one. Lamyaa, what do you think?

L.G: Well, of course I have to mention demand again, because this is the equivalent of the off-taker in a typical project finance, or a finance project generally. But I think we mentioned the currency issue. And if these data centers are going to be hosted here, your customer — in this case, what we were talking about — is someone who's also generating its revenue in local currency, while a good bunch of the procurement is going to be in foreign currency.

So I think there will be a need for multilateral agencies, ECAs and [the like] to finance. And they usually expect, if the country is keen on hosting these mega projects, that there should be — not government support — but a sort of streamlined process for them to get comfortable with the risks associated with the complexity. 

From a structural point of view — just to make a reference to another complex but successful project, for example, the LNG project — you just break them down, and as you scale, you take risks that are associated [with] the current demand only.

N.E: So, demand, currency and source of finance, and the ability to replicate. Ahmed?

A.S: I would have normally said demand, but Mohamed took it, so that's fine. I think the one important part in Egypt that's also holding [things] back a little bit, even [for] people building smaller versions and then getting bigger over time, is power. 

Power allocation to the right plots of land isn't happening as quickly as it should.

So I think that's the one unlock that's so important as well.

N.E: Maged, last thought from you.

M.E: The challenge always when you're speaking last is that everybody has said everything. So I'll try to take a helicopter view and see, for Egypt to win — I mean, it has, in my view, a lot of things. 

Of course, the strategic location; we [learn] it in school: [the] crossroads between three continents. Very important, the connectivity here, and then [it] has low latency. That's very valuable.

You have 20 active submarine cables, and I don't know how many in the room know that 90% of all Asia–Europe internet [traffic] passes by us. The traffic goes through here. You have [a] cost advantage, of course: the construction, the labor. It's important. And the green energy, which is one big bet: the solar and wind, and hopefully in the near future, we'll have nuclear power to power it as well.So that's something we consider.

So Egypt's right to win, really, is, in my mind, [a] green AI compute hub. If we're able to do that, that could change the game. 

The problem in a project like this is it falls between GAFI, MCIT, NTRA, ITIDA, land, water, electricity. You need a catalyst. You need a golden license to glue that together.

There is one way that could be used as a catalyst — and I was discussing it with Allam before the event — [which] is a concept called data or digital embassy. And this touches the data sovereignty that everybody [talked about]. Because data sovereignty is good — it can create demand in the country — but it doesn't help when you're trying to attract hyperscalers from abroad. So the way around this is: think of it as any embassy. The land is the land of another country, and the goods and the regulation. You can do the bilateral [agreement].

And this is not something fictitious, I think Estonia, in 2017; [they] did this in Luxembourg. So it's done. Other countries have done it. And it's important for resilience. Like when we do [a] data center, we do disaster recovery. That's why people are talking [about it] now; it's [a] hot topic: Egypt can host a data center. [There are] wars, there's also cyberattacks, there is the possibility of natural disaster, God forbid, or power failure. 

So you need your backup, right? And this digital embassy, whatever the name would come [to be], could be one of the catalysts, in my view.

N.E: Okay, so a digital embassy. We're going to have to wrap it up. If you've got a very, very quick thought, Mohamed?

M.A: Just one last, very quick thought, because it perhaps might be on the minds of the audience. 

Over the past years, there have been impediments from the regulatory side. But as someone who's been involved in this since 2021, we have really seen a very proactive situation from the government, led by the MCIT, being very practical and flexible in pushing away any barriers and impediments, especially on the regulatory side.

And I think this is where the digital embassy concept is coming in, aggregating demand. I see a lot of head shaking. So this is something that needs to be said, to ensure that really this is not an excuse anymore for us, for example, as a private sector investor and developer.

N.E: No more excuses. Time to get it done. Brilliant. Thank you, all four of you, for a wonderful discussion.