Europe in the Global AI Race

Live from Morgan Stanley’s European Tech, Media and Telecom conference in Barcelona, our roundtable of analysts discuss artificial intelligence in Europe, and how the region could enable the Agentic AI wave.

Read more insights from Morgan Stanley.

----- Transcript -----

Paul Walsh: Welcome to Thoughts on the Market. I'm Paul Walsh, Morgan Stanley's European head of research product. We are bringing you a special episode today live from Morgan Stanley's, 25th European TMT Conference, currently underway.

The central theme we're focused on: Can Europe keep up from a technology development perspective?

It's Wednesday, November the 12th at 8:00 AM in Barcelona.

Earlier this morning I was live on stage with my colleagues, Adam Wood, Head of European Technology and Payments, Emmet Kelly, Head of European Telco and Data Centers, and Lee Simpson, Head of European Technology Hardware. The larger context of our conversation was tech diffusion, one of our four key themes that we've identified at Morgan Stanley Research for 2025.

For the panel, we wanted to focus further on agentic AI in Europe, AI disruption as well as adoption, and data centers. We started off with my question to Adam. I asked him to frame our conversation around how Europe is enabling the Agentic AI wave.

Adam Wood: I mean, I think obviously the debate around GenAI, and particularly enterprise software, my space has changed quite a lot over the last three to four months. Maybe it's good if we do go back a little bit to the period before that – when everything was more positive in the world. And I think it is important to think about, you know, why we were excited, before we started to debate the outcomes.

And the reason we were excited was we've obviously done a lot of work with enterprise software to automate business processes. That's what; that's ultimately what software is about. It's about automating and standardizing business processes. They can be done more efficiently and more repeatably. We'd done work in the past on RPA vendors who tried to take the automation further. And we were getting numbers that, you know, 30 – 40 percent of enterprise processes have been automated in this way. But I think the feeling was it was still the minority. And the reason for that was it was quite difficult with traditional coding techniques to go a lot further. You know, if you take the call center as a classic example, it's very difficult to code what every response is going to be to human interaction with a call center worker. It's practically impossible.

And so, you know, what we did for a long time was more – where we got into those situations where it was difficult to code every outcome, we'd leave it with labor. And we'd do the labor arbitrage often, where we'd move from onshore workers to offshore workers, but we'd still leave it as a relatively manual process with human intervention in it.

I think the really exciting thing about GenAI is it completely transforms that equation because if the computers can understand natural human language, again to our call center example, we can train the models on every call center interaction. And then first of all, we can help the call center worker predict what the responses are going to be to incoming queries. And then maybe over time we can even automate that role.

I think it goes a lot further than, you know, call center workers. We can go into finance where a lot of work is still either manual data re-entry or a remediation of errors. And again, we can automate a lot more of those tasks. That's obviously where, where SAP's involved. But basically what I'm trying to say is if we expand massively the capabilities of what software can automate, surely that has to be good for the software sector that has to expand the addressable markets of what software companies are going to be able to do.

Now we can have a secondary debate around: Is it going to be the incumbents, is it going to be corporates that do more themselves? Is it going to be new entrants that that benefit from this? But I think it's very hard to argue that if you expand dramatically the capabilities of what software can do, you don't get a benefit from that in the sector.

Now we're a little bit more consumer today in terms of spending, and the enterprises are lagging a little bit. But I think for us, that's just a question of timing. And we think we'll see that come through.

I'll leave it there. But I think there's lots of opportunities in software. We're probably yet to see them come through in numbers, but that shouldn't mean we get, you know, kind of, we don't think they're going to happen.

Paul Walsh: Yeah. We’re going to talk separately about AI disruption as we go through this morning's discussion. But what's the pushback you get, Adam, to this notion of, you know, the addressable market expanding?

Adam Wood: It's one of a number of things. It's that… And we get onto the kind of the multiple bear cases that come up on enterprise software. It would be some combination of, well, if coding becomes dramatically cheaper and we can set up, you know, user interfaces on the fly in the morning, that can query data sets; and we can access those data sets almost in an automated way. Well, maybe companies just do this themselves and we move from a world where we've been outsourcing software to third party software vendors; we do more of it in-house. That would be one.

The other one would be the barriers to entry of software have just come down dramatically. It's so much easier to write the code, to build a software company and to get out into the market. That it's going to be new entrants that challenge the incumbents. And that will just bring price pressure on the whole market and bring… So, although what we automate gets bigger, the price we charge to do it comes down.

The third one would be the seat-based pricing issue that a lot of software vendors to date have expressed the value they deliver to customers through. How many seats of the software you have in house.

Well, if we take out 10 – 20 percent of your HR department because we make them 10, 20, 30 percent more efficient. Does that mean we pay the software vendor 10, 20, 30 percent less? And so again, we're delivering more value, we're automating more and making companies more efficient. But the value doesn't accrue to the software vendors. It's some combination of those themes I think that people would worry about.

Paul Walsh: And Lee, let’s bring you into the conversation here as well, because around this theme of enabling the agentic AI way, we sort of identified three main enabler sectors. Obviously, Adam’s with the software side. Cap goods being the other one that we mentioned in the work that we've done. But obviously semis is also an important piece of this puzzle. Walk us through your thoughts, please.

Lee Simpson: Sure. I think from a sort of a hardware perspective, and really we're talking about semiconductors here and possibly even just the equipment guys, specifically – when seeing things through a European lens. It's been a bonanza. We've seen quite a big build out obviously for GPUs. We've seen incredible new server architectures going into the cloud. And now we're at the point where we're changing things a little bit. Does the power architecture need to be changed? Does the nature of the compute need to change? And with that, the development and the supply needs to move with that as well.

So, we're now seeing the mantle being picked up by the AI guys at the very leading edge of logic. So, someone has to put the equipment in the ground, and the equipment guys are being leaned into. And you're starting to see that change in the order book now.

Now, I labor this point largely because, you know, we'd been seen as laggards frankly in the last couple of years. It'd been a U.S. story, a GPU heavy story. But I think for us now we're starting to see a flipping of that and it's like, hold on, these are beneficiaries. And I really think it's 'cause that bow wave has changed in logic.

Paul Walsh: And Lee, you talked there in your opening remarks about the extent to which obviously the focus has been predominantly on the U.S. ways to play, which is totally understandable for global investors. And obviously this has been an extraordinary year of ups and downs as it relates to the tech space.

What's your sense in terms of what you are getting back from clients? Is the focus shifts may be from some of those U.S. ways to play to Europe? Are you sensing that shift taking place? How are clients interacting with you as it relates to the focus between the opportunities in the U.S. and Asia, frankly, versus Europe?

Lee Simpson: Yeah. I mean, Europe's coming more into debate. It's more; people are willing to talk to some of the players. We've got other players in the analog space playing into that as well. But I think for me, if we take a step back and keep this at the global level, there's a huge debate now around what is the size of build out that we need for AI?

What is the nature of the compute? What is the power pool? What is the power budgets going to look like in data centers? And Emmet will talk to that as well. So, all of that… Some of that argument’s coming now and centering on Europe. How do they play into this? But for me, most of what we're finding people debate about – is a 20-25 gigawatt year feasible for [20]27? Is a 30-35 gigawatt for [20]28 feasible? And so, I think that's the debate line at this point – not so much as Europe in the debate. It's more what is that global pool going to look like?

Paul Walsh: Yeah. This whole infrastructure rollout's got significant implications for your coverage universe…

Lee Simpson: It does. Yeah.

Paul Walsh: Emmet, it may be a bit tangential for the telco space, but was there anything you wanted to add there as it relates to this sort of agentic wave piece from a telco's perspective?

Emmet Kelly: Yeah, there's a consensus view out there that telcos are not really that tuned into the AI wave at the moment – just from a stock market perspective. I think it's fair to say some telcos have been a source of funds for AI and we've seen that in a stock market context, especially in the U.S. telco space, versus U.S. tech over the last three to six months, has been a source of funds.

So, there are a lot of question marks about the telco exposure to AI. And I think the telcos have kind of struggled to put their case forward about how they can benefit from AI. They talked 18 months ago about using chatbots. They talked about smart networks, et cetera, but they haven't really advanced their case since then.

And we don't see telcos involved much in the data center space. And that's understandable because investing in data centers, as we've written, is extremely expensive. So, if I rewind the clock two years ago, a good size data center was 1 megawatt in size. And a year ago, that number was somewhere about 50 to 100 megawatts in size. And today a big data center is a gigawatt. Now if you want to roll out a 100 megawatt data center, which is a decent sized data center, but it's not huge – that will cost roughly 3 billion euros to roll out.

So, telcos, they've yet to really prove that they've got much positive exposure to AI.

Paul Walsh: That was an edited excerpt from my conversation with Adam, Emmet and Lee. Many thanks to them for taking the time out for that discussion and the live audience for hearing us out.

We will have a concluding episode tomorrow where we dig into tech disruption and data center investments. So please do come back for that very topical conversation.

As always, thanks for listening. Let us know what you think about this and other episodes by leaving us a review wherever you get your podcasts. And if you enjoy Thoughts on the Market, please tell a friend or colleague to tune in today.