Cheaper, smarter and ever broader uses: AI offers a long growth runway

Talking Heads podcast with Derek Glynn and Pamela Hegarty

Daniel Morris: Hello, and welcome to the BNP Paribas Asset Management Talking Heads podcast. Every week, Talking Heads will bring you in-depth insights and analysis on the topics that really matter to investors. In this episode, we’ll be discussing the latest developments in AI (artificial intelligence). I’m Daniel Morris, Chief Market Strategist, and I’m joined today by Pam Hegarty and Derek Glynn, Co-portfolio Managers for our disruptive technology strategy. Welcome, Pam, Derek. Thanks for joining me.

Pamela Hegarty: Thank you so much, Daniel. It’s great to be here.

Derek Glynn: Thanks, Daniel. Excited to be here.

DM: I have to be honest; I don’t envy you and your job. There is just so much going on with AI, how you possibly keep up with all of it, assess all of it, given the wildly divergent opinions about what all this means. Is it going to usher in a new wave of growth and productivity and disinflation alternatively?

Is it going to take all our jobs? Your job is to try to figure this out, and more importantly, figure out how to invest in it. Over the last year or so since AI really leapt to the forefront of investors’ minds, there’s been a huge array of developments, seemingly almost by the day. Could you give us the latest, most exciting developments that you see? Derek, maybe you can start off.

DG: Sure. The technology itself continues to rapidly improve. Each new model generation demonstrates this persistent trend of improving model intelligence and declining cost per unit of intelligence. So, in essence, the models continue to get smarter and cheaper, which is unlocking new use cases and spurring adoption.

More specifically, AI is evolving from simple tools to complex autonomous systems. Agentic AI systems are increasingly capable of reasoning, planning and executing across multi-step workflows. Enterprises are beginning to adopt these models to automate tasks and accelerate research processes. A prominent emerging use case is vibe coding, where developers guide and coordinate systems through prompts rather than writing code line by line, and this is redefining how software is built.

At the same time, breakthroughs in robotics and embodied AI are allowing machines to sense, interpret and act in the physical world. We expect use cases to broaden as AI capabilities are embedded into hardware that have ‘eyes and ears’, enabling richer real-world interaction. At the edge, intelligence is moving onto devices like smartphones and wearables. On-device AI supports real-time personalised experiences without depending on any centralised cloud infrastructure.

So collectively, these developments all point toward a future of seamlessly integrated intelligence spanning both digital and physical worlds.

DM: Now all of this is requiring a lot of investment. We’ve all seen the figures about big increases in anticipated capital expenditure. As capex on AI infrastructure continues to grow at a quite rapid pace, Derek, maybe you can pick up on what the implications of all this are for cloud service providers as well as for the hardware and semiconductor companies.

DG: The cloud service providers are building out the AI data centre infrastructure to provide the compute for model training and inferencing.

Capital expenditures are already high, but we expect this to grow more than 70% year over year in 2026, and to continue to grow in 2027, surpassing $750 billion in total spend for the collective group, which would include the four major US hyperscalers and a leading social media platform in the US.

It is still [too] early to gauge the future return on investment for today’s capex dollars, but there are encouraging signs. Cloud revenue growth is accelerating, backlogs are surging, and margins remain healthy. In calendar first quarter of 2026, for the three largest US hyperscalers, cloud revenue growth accelerated six percentage points to 39% year over year growth.

Backlogs surged close to $1.5 trillion, and margins remained healthy in the mid 30% range. We’re optimistic the group can earn returns above their cost of capital. Pam, what do you think is the read-through to the hardware and semiconductor companies?

PH: You know, all of this increased outlook for AI capital spending has been the primary catalyst for both revenue growth acceleration and positive earnings revisions for a broadening group of hardware and semi(conductor) names.

The leadership has diversified as investors are looking for the next area of investment to support both training and inferencing applications. For example, with agentic AI, there’s a growing role for the central processing unit or a CPU, and not just the graphics processing units or GPUs.

And these CPUs are being used for tasks including workflow orchestration data movement in and out of the models, and to enable the agents to call and run software tools on their behalf. One company in particular in the CPU business recently raised their total addressable market forecast for 2030 by two times, although they had just given that forecast only six months ago.

There are a lot of risks, and this includes the potential for a digestion period at some time during the AI infrastructure build-out. But for now, we are seeing a lot of momentum given the emergence of agentic AI. Another area of potential risk is in memory chips and hard disk drives.

The current economics that companies are achieving may not be sustainable because when supply finally does catch up with demand, prices will fall. So, we continue to be selective in our investment approach.

DM: You started to list some of the risks that you see in the market, and Derek, earlier you talked about vibe coding as one of the more recent developments.

But to speak of risks, we think about what the implications are for software companies, and I imagine we haven’t forgotten what happened at the beginning of the year when that threat was reflected in the prices of software stocks. Derek, what do you see as the key risks and opportunities, particularly in the software sector?

And is AI ‘eating software’?

DG: At a very high level, it is important to keep in mind that AI is fundamentally software. It is a distinct and new type of software compared to traditional programs, but it is nevertheless exciting that the driving force behind a lot of the innovation in the economy right now is fundamentally software.

The essence of the question, though, is really what AI means for the traditional incumbent software companies. The answer is complicated. We think some companies will be disrupted or disintermediated by AI, but others could thrive. In terms of the AI-related bear cases, investors fear that AI agentic workspaces will be the new place where we do work, routing around the traditional user interfaces and software applications that many of us interact with day to day. This could lead to incumbent software being essentially reduced to a dumb data repository instead of the place where we do work.

A second fear is that competitive intensity in software will increase because the private AI labs will develop their own software applications. There are several other bear cases.

On the other hand, we think some incumbents could thrive. Those that have scale with large install bases could serve as the natural delivery mechanism for AI, and we also think AI expands the total addressable market, or TAM. Software companies that increase productivity for their customers through AI should be able to earn incremental revenue for that capability.

Finally, companies with unique domain knowledge, system of record status, robust data security protections, and mission-critical capabilities should prove more resilient. Pam also covers a couple industries within software. Pam, how are you thinking about the risks and opportunities in those industries?

PH: Indeed, we are finding some resilient software businesses both in cybersecurity and infrastructure. Leading cybersecurity companies have been leveraging machine learning for years to help identify and combat threats, and to do so in real time. Those that are able to adopt generative AI are in a strong position to help protect against agentic AI-enabled attacks as we go forward.

Some of the interesting areas in infrastructure software include modern database systems that can handle both structured and unstructured data. That means it’s not just data that’s presented in files and rows or in columns and rows, but it’s also unstructured data like different types of media, emails and so forth.

One of our positions in the fund is a software tool that is used to store code during the software development process. As you mentioned, since AI is software, AI models are creating much more demand and usage for this particular application.

DM: Of course, we can’t talk about AI without addressing the bubble question. You see more and more references to what happened in 1999 and then in 2000. That said, if we look at valuations, there is certainly a question whether or not we are in any kind of similar territory today.

Pam, are you concerned that we’re in an AI bubble?

PH: That is a very important question, and we started to study this issue more deeply in the fall of last year, where Derek and I looked at some of the similarities and differences between today and the dotcom bubble of the late 1990s.

Our conclusion was that AI was not yet in a bubble given that the infrastructure spending was in line with demand and primarily funded from the operating cash flows of these very large cloud service providers that have strong balance sheets. In addition, AI products and services were –and continue to be – monetized rapidly, given the infrastructure exists to reach the end consumer, whether it’s through broadband or mobile phones and so forth.

Today we see that demand remains stronger than supply, and agentic AI is catalysing additional spending to support AI models in production. However, we are very watchful of any signs that AI exuberance is becoming irrational due to these recently heightened investor expectations, and we are definitely monitoring the increasing use of debt and off-balance sheet financing, as well as some of the more recent equity raises.

Just today I read that Morgan Stanley predicts global AI-related debt issuance will be $570 billion this year, more than double last year’s level. So, we are very watchful of some of these developments. And Derek, I think it would be great if you could discuss some additional specific areas of concern.

DG: Sure. We have observed some pockets of froth or irrational exuberance, for instance, in areas like quantum computing. Some of these companies have very little revenue, rich valuations, and an uncertain path ahead to reach commercial viability. We stay true to our investment process, are generally disciplined when it comes to valuation, and have tried to avoid areas such as this.

We are always mindful of valuations broadly speaking, and try to avoid areas of extreme optimism, but we continue to find attractive stock-specific opportunities. We think there’s still a long growth runway ahead for companies that could benefit from the broader theme of AI.

DM: If I could summarize what I think are some of the key messages you’d like to leave with people listening to the podcast. You pointed out very simply [that] the AI models are getting better, but importantly, also cheaper, which is opening up more opportunities to use them by businesses beyond, of course, how individuals use them.

Importantly, on [the] one hand we see this very significant increase in capex, you’re also seeing AI revenues and profits growing which hopefully will justify that spend. Ultimately, around the very critical bubble question, I guess the answer is ‘it depends’. Generally, not where we were, say, at the end of the ’90s.

That said, you remain very watchful, and do see some pockets of what we might term irrational exuberance.

Well, Pam, Derek, thank you very much for joining me.

PH: Thank you so much, Daniel. We appreciate your questions and your time.

DG: Thanks, Daniel. Appreciate the opportunity.

DM: That’s it for this week’s episode of Talking Heads. If you would like more information about our capabilities in investing in the areas around AI, please reach out to your asset management contact, or check out Viewpoint, our website for investment insights at viewpoint.bnpparibas-am.com.

We recommend subscribing to Talking Heads on your favourite podcast channel, such as YouTube or Spotify. You’ll receive your podcast episodes every week. If you like Talking Heads, leave us a positive review and a nice rating.

You’ve been listening to the BNP Paribas Asset Management Talking Heads podcast with me, Daniel Morris, Pam Hegarty and Derrick Glenn, Co-portfolio Managers for our disruptive technology strategy.

Please do join me next week. Until then, take care.