Spending on AI infrastructure continues at a breakneck pace. But is AI also a good ongoing return on investment? Coming up, we have some expert predictions.

From Thrivent Asset Management, welcome to Advisor’s Market360™, a podcast for you, the driven financial advisor.

It’s estimated that by the end of 2025, spending on AI infrastructure will approach $1.5 trillion globally. That’s trillion — with a T. With that level of spending, will AI show an ongoing return on investment?

Long-time listeners of this podcast may recall several past episodes about AI and broad approaches to investing in it. We again asked our AI experts to weigh in with their thoughts on this technology. Returning to the podcast is Senior Equity Research Analyst Peter Karazeris. Also joining us is Senior Portfolio Manager Jaimin Soni, who manages Thrivent Mid Cap Growth Fund.

Let's get into it…

The most logical place to start is with a broad overview of the current state of investing in AI. Here’s Soni:

Soni: We are coming up to three years post the launch of ChatGPT. It's not been that long of a time, but the technology has evolved rapidly. And ChatGPT right now has over 700 million weekly active users. A lot of adoption. But from our seat as investors, we continue to evaluate how to position ourselves as this overall landscape continues to change. Last year when we met, we talked about how whenever there's a new technology that comes to market, there is a spending cycle that comes after it.

We asked Soni to elaborate on the spending cycle:

Soni: First there is development of the infrastructure to build out the core components, which triggers spending for hardware and semiconductor companies. And then software applications get built on top of that, services get deployed and a whole different set of companies benefit.

Now, this cycle has played out multiple times in the past. We saw this during the build out of the internet in the late '90s, and then the public cloud build-out, and even for 5G in the last decade. When we translate that to what's happening with AI right now, we have been in this data center development cycle for the past couple of years, and that helps us to put the first building block. We are still in the thick of it. Even year to date, when we look at what large public company earnings have reported, they are continuing to spend to build out data centers. And the stocks which have been the most rewarded are the Arms providers like NVIDIA and Broadcom. From a portfolio management perspective, while it might feel that we are getting long in the tooth for investing in AI, there is still a lot of runway left, and we continue to be excited and invested in the companies we think stand to benefit.

When Soni references Arms providers, he’s referencing companies that supply critical components to fuel the growth of AI. Next, we wanted to get a better understanding of the size of the opportunity. We asked Karazeris to give us some numbers…

Karazeris: If we look at the big five cloud service providers, Amazon, Meta, Google, Microsoft, Oracle. If we look at those five at the end of last year, coming into this year, expectations were for capital spending to increase about 15%. And the out year, being 2026, it was expected to be up about 4%. And there were some hand wringing about, well, what if 2026 doesn't grow? Isn't that what we're really playing for? Well, if we look here nine months later, that 15% growth at this point, it looks like 2025 we're going to have 56% growth in capital spending. So, we dramatically underestimated how much would be spent on AI infrastructure.

We wanted to know what Wall Street is expecting down the road. Karazeris explains:

Karazeris: As I said, at the end of last year, if you looked at the out year, which was 2026, the growth expectation was about 4%, so mid-single digits. And 2025 was mid-teens. Well, at this point, the next year, 2026, has grown to be mid-teens, 16%. The Street seems to anchor to that number, some mid-teens growth for capital spend. Now, it was a baseline at the end of last year, but we blew right by it. And then the out year, 2027, is mid-single digits plus 4% just like it was last time. The setup for consensus is pretty similar.

However, Karazeris has his own projections, based on deep research and field visits to companies across the globe. We’ll let him explain:

Karazeris: I was recently in Asia meeting with the supply chain for these AI infrastructure companies. If I pulled it all together, AI semiconductor units, these semiconductor processors from NVIDIA and Broadcom, the ASICs that Broadcom is making, the expectation for growth is 55 to 60%. I would suggest that the capital spending plans are probably low relative to what we see in the supply chain. Also, I met with the biggest server manufacturer in the world, Hon Hai. And they talked about NVIDIA’s server racks, the racks that hold the servers, these AI servers, those builds doubling from 2025 to 2026. So again, if we think about where the Street is, there's hand wringing about the out year. We think it's going to be mid-teens growth for next year. I suspect based on what those customers are telling the infrastructure players and they're telling their supply chain to build — we're going to blow by that again. So again, good setup for stocks.

The overview of AI provided some good context. But the question, which is probably top of mind for financial advisors, is about the potential for the AI industry to continue providing positive returns on investments. Here’s Soni:

Soni: The key thing to understand here, I think, is that it's not about immediate financial payback. AI needs to be viewed as a foundational technology, which is like electricity or the internet, where costs precede exponential gains that can come in the future. In the near term, and by near term, I mean 1-3 years, the ROI may appear underwhelming. If we zoom into companies, the big spenders on AI are some of the Mag Seven companies that we mentioned, like Microsoft, Amazon, Google and Meta, and they are powering their investments and really driving revenues of the other hardware and semiconductor companies. But for them, we are also starting to see some returns show up. Microsoft is estimated to be at a $20 billion annual revenue run rate from AI directly, and that's growing at a very rapid pace. For other companies like Google and Meta, their investments in AI are helping them improve their core products like search and advertising in the social media area. I think there's a long tail of monetization in front of us, and we as investors can benefit by identifying which companies can be leaders as this ecosystem develops. So that's where we want to be positioned.

As a reminder, when Soni mentions Mag Seven, that’s shorthand for the Magnificent Seven, a group of mega-cap stocks tied to AI and tech. We asked Karazeris for specific names that he believes will continue to be good investments.

Karazeris: If I think about AI from an infrastructure perspective, we've always said, first, it's a compute problem, next it's a networking problem, and least of all, it's a storage problem. Those being the three main elements of infrastructure. On the compute side, we continue to like NVIDIA and Broadcom, which have been the big drivers, the big innovators on AI acceleration. And then we also like Taiwan Semiconductor because that's where the chips are made. Part of the compute infrastructure is system memory. And we've had investments laid out in Micron and Samsung. We continue to like those. Pricing is very robust in system memory, and there's actually a secular tale to it. So, we're not just riding a cycle trade, but we actually are riding a secular investment, which is also hitting cyclical peak at some point next year. And then finally, we like networking. And networking is an important part of clustering together all these different compute nodes so you can get maximize your performance. And in that space, we still like Arista. We like names like Amphenol and Celestica. So that's somewhat where we've been and also where we are looking for. We still like it for next year, at least.

Soni weighed in with some names he likes in the software and services areas of AI:

Soni: When we think about software, it's a big space where we can bucket it into a couple of different ways. One is infrastructure. Infrastructure software, and the other is application software. Infrastructure software are companies that work closely with hyperscalers that are software products that sit in the data center. And that's where we think there is more immediate opportunity because they're closer to the AI workloads that are getting ran. And that's where companies like Datadog is in the mid-cap space is there, in addition to the bigger companies like Microsoft and Oracle, which also have infrastructure software offerings.

Now, when we come to application software, that's where there is a little bit more debate. This is where companies like Salesforce, ServiceNow, HubSpot are present, and there is some disruption risk from AI itself. And that's where the debate is right now in terms of what the next two or three years is going to look like for application software companies. We have a more guarded opinion on what those companies can do in the near term.

We do prefer vertical software companies. These are companies that target specific industries and mission critical systems. These are very, very sticky products, and the customers have built their entire businesses around what this product is, which allows them to minimize disruption from any new way of doing things like AI. That’s where we like companies like Tyler Software or Guidewire. These are mid-cap companies which target government or insurance verticals, and we think they continue to benefit as they will start incorporating AI in their products.

The point solutions here—which are just solving a single problem—they probably are at more risk from things like agentic AI, which are getting developed. Now, as we move along to other areas like social media and services, that's where Meta and Google have large revenue bases, really big customer engagement, is strong. And they continue to improve their core products with AI, and that has continued to drive revenues. And we believe that will be the case in the future as well.

We've talked about generations of AI—the infrastructure generation, the cloud generation. What's the next generation? Our experts have some thoughts. Let’s start with Soni:

Soni: As AI is evolving, there are more use cases developing on how we can use it. A

classic example here is agentic AI. I'll explain the term. That is essentially the next step in the evolution of generative AI. Now, our interactions with generative AI thus far have been in this request-response type mechanism where we are interacting with a chatbot, we ask a question, and we get a very well-thought-out answer in response. But there is no action beyond that. We are left to do what we think is best with that information.

Now, what if we can enable AI with better reasoning capabilities and ability to act on our behalf? Then we get into the realm of agentic AI. That’s the next step here where agents can complete tasks on our behalf in the internet or in the office environment as well. A simple example here, just to make this real, could be an agent who can book a trip for you. Now, booking a trip is a multi-step process. It has to incorporate your preferences in terms of the hotels you want to like to stay in, obviously where you want to go, the airline companies, your mileage information and all of that, and your preferences overall. And of course, it's all got to fit in a budget. There are several steps here, but wouldn't it be great if an AI agent can just do that all for you? Now, even in the office, if I let my workplace agent know about my upcoming vacation plan, maybe it can block out my calendar, reschedule my meetings, update that information in Workday, and also communicate that to my boss. That would be great for me. These are just some examples on how agentic AI is developing. We are going to start seeing this in the next year.

Karazeris offered some interesting examples of physical AI which is the next big step.

Karazeris: There’s another wave after agentic. The future vision of that is really the humanoid robots that you can watch videos of them doing different types of tasks, playing a sport. What's fascinating about them is that they would train by watching a video and then learn from that and then be able to go do the task. That's down the road. The more immediate implementation of automation and robotics is really autonomous vehicles. And we can see those today on the roads. Waymo has taxis out in the road. Tesla is next to come to market with robotaxis. I really think we're starting to see the tip of the iceberg, and there's much more to come on robotics and automation.

Another development is governments building their own AI data centers, which is referred to by our experts as sovereign AI. Here’s Karazeris:

Karazeris: The latest example was the UK committing to spend $30 billion to build out their AI infrastructure. The announcement was made with NVIDIA as well as with the Trump administration there. And it's going to be used for government organizations, defense, security, as well as startup companies and academic institutions like Imperial College of London and University of Cambridge. That’s probably the next customer set to come up.

We asked Soni to frame these advances in terms of their investment potential:

Soni: This development of agentic AI, physical AI, these are all mechanisms that will help to build the ROI case for the investments that have been happening thus far. These are pieces that are continuing to build up and over time, should help companies get the return on this massive investment that's happening.

Discussing AI also means discussing its impact on employment. Here’s Karazeris’ take on the subject.

Karazeris: The risk to employment, it's a very broad topic, and we can analyze it several ways. I think one of the best things to do is go back to kind of a first principles basis. Every time a new technology comes along, there are jobs that are lost and there's legacy technologies that are diminished or go away, but there's new jobs that are created and there's new opportunities that are created. Ultimately creating more productivity for the workers. And productivity gains are an important part of growth in the economy. And ultimately, that growth in the economy does generate more jobs. I think that's the first principle basis that we really have to look at this through, as opposed to an individual job loss.

We asked for an example of this principle…

Karazeris: When the email came around, there was job loss of staff in the mailroom. Now that was real loss. Those were people who had those jobs, but they could be retrained into something, a different type of job. Quite often that was an entry level, you know, people would say, oh, I started in the mailroom and, you know, worked my way up into management. Okay. Well, they started in a different type of entry level job, but they could still work their way up. But what were the benefits, right? Email became a tremendous productivity tool which made every employee at the company more productive and made the company more productive overall. I think that's the right way to think about AI. Yes, there's going to be some job loss. Yes, there's going to be legacy technologies or processes that that go away. But it's the productivity gains and it's the new opportunities.

We covered a lot of ground in this episode. So, in keeping with the subject of this episode, we asked ChatGPT for a summary. Here’s what it came up with:

CHATGPT: ChatGPT’s rapid adoption has fueled a massive AI investment cycle, now centered on data center and semiconductor build-out. Soni and Karazeris note that spending by major cloud providers like Amazon, Microsoft and Google is far exceeding earlier forecasts, with 2025 capex now expected to rise over 50%. Key beneficiaries include NVIDIA, Broadcom, Micron, Samsung and Arista.

While near-term ROI is limited, AI is viewed as a foundational technology akin to the Internet or electricity, with exponential long-term gains. Microsoft already generates about $20 billion annually from AI, and others are improving core products through it.

Future waves include agentic AI (autonomous digital assistants), physical AI (robotics and automation), and sovereign AI (government-led infrastructure). These advances are expected to strengthen the investment case over time.

On employment, AI will displace some roles but boost productivity and create new ones—echoing past tech shifts.

Thanks, that's a nice summary, ChatGPT.

Investing in AI is an important discussion happening at Thrivent Asset Management. We are closely following the trend as it shifts through different generations. And we will keep you updated on what we learn.

We hope you enjoyed this look at the ongoing potential for ROI from AI. Once again, we would like to thank Jaimin Soni and Peter Karazeris for their insights. What did you think of this episode? Email us at podcast@thriventfunds.com with your feedback or questions for our experts. Want more episodes of Advisors Market360™ and other market and investing insights? Visit us at thriventfunds.com, where you can learn how we can partner with you, the driven financial advisor. Bye for now.

All information and representations herein are as of 10/2/2025, unless otherwise noted.

Past performance is not necessarily indicative of future results.

Investing involves risks, including the possible loss of principal.

The views expressed are as of the date given, may change as market or other conditions change, and may differ from views expressed by other Thrivent Asset Management, LLC associates. Actual investment decisions made by Thrivent Asset Management, LLC will not necessarily reflect the views expressed. This information should not be considered investment advice or a recommendation of any particular security, strategy or product. Investment decisions should always be made based on an investor's specific financial needs, objectives, goals, time horizon, and risk tolerance.

This podcast refers to specific securities which Thrivent Mutual Funds may own. A complete listing of the holdings for each of the Thrivent Mutual Funds is available on thriventfunds.com.

Asset management services provided by Thrivent Asset Management, LLC, an SEC-registered investment adviser. Thrivent Distributors, LLC and Thrivent Asset Management, LLC are subsidiaries of Thrivent, the marketing name for Thrivent Financial for Lutherans.