The Capital Cycle Podcast - AI Eats Software?

Episode Date: February 27, 2026

Rising investment appears to be leading to increasing capacity and competition. Edward Chancellor talks to Tom Wharram, a North American Equities Analyst. For more information, or to access select articles from Marathon’s Global Investment Review publications which accompany this podcast series, please visit www.thecapitalcycle.co.uk Hosted on Acast. See acast.com/privacy for more information.

Welcome to another episode of the Capital Cycle Podcast. This is Edward Chancellor. And I've got with me Tom Warram, who's an analyst on the North American portfolios for Marathon Asset Management. And I see you again, Tom. Very nice to see you. Thank you for having me on again. We're going to talk about the hot topic of the day, which is the, I don't even know how to say it, the SARS apocalypse, of the problems that are facing software as a service companies

that have been in the news recently, the threat has wiped billions of dollars off the market caps of these companies. It's hard to remember that up until recently, these companies will market darlings. Mark Andriesen, the Silicon Valley investor, referred to them as eating the world back in 2011. So that's what, 15 years ago. I sent you a piece that I wrote back in 2021, late 2021, showing how the cloud index was up 300% to mid-21.

And these companies at the time were trading at 30 times forward earnings. And some of the stocks, such as Adobe, was trading on 20 times sales. And at the time I pointed out that the investors were in cloud kuku land. and that it was extremely unlikely that investors were going to get a decent return by buying at that inflated level. But now market sentiment has turned against software services companies and they're being spurned by investors. Yeah, that's right. It's been a very tough time for software investors. The S&P 500 Software Index has underperformed the S&P 500 by about 35 points since the end of July last year. And it's a very stark reminder that apparently unassailable new business models are often those that are yet to encounter a sufficient shock.

So for us, the falls in the share price are certainly reasons to look at the sector. But what we ask ourselves is how should we as capital cycle investors approach this setup? And to give the rest of the story away, the answer is with caution. And the reason software services companies are out of favour is the market's beginning. to worry that AI is actually going to eat their breakfast. In his book, Engines at Move Markets, the investor Sandy Nann writes about new technologies and how the markets receive them. And he says when a groundbreaking new technology emerges, it's easier to spot the losers than it is to identify the winners, at least at the initial stage. But that's not necessarily true in this case,

because there seems a certain amount of disagreement as to whether AI is such a threat to software services. And even Jensen Huang of Nvidia says that it's absurd to think that software services, businesses are going out. And yet one reads almost on a daily basis and the market responds almost on a daily basis to this notion that AI models would soon be able to code so proficiently on their own that companies will be able to handle their own software needs. But before going on to that, you tell an interesting story, which I hadn't heard about before, of the Jacquard Loom. Yeah, so this is a nice little bit of history because it brings together both the capital cycle and also software and has some parallels to the current situation. So in the early 19th century, Joseph Marie Jacquard, and sorry for my terrible pronunciation, created an automated silk weaving machine. And

And he was actually awarded, interestingly, a patent by Napoleon in 1804 for this invention. What the machine did is it used punched cards, which had encoded in them complex weaving patterns. And those punched cards are then fed into an automated mechanism which sits on the top of the loom, and that controlled the threads during weaving. This meant that intricate fabrics could be produced far more quickly and reliably than before, dramatically increasing the productivity of weavers. The capital cycle element of this story is that Leon Silk Weavers invested in new looms, often using debt.

And secondly, there was a productivity improvement because the original mechanism wasn't actually that efficient, and over about 10 years they perfected the mechanism. And because of those two things, industry output ballooned. And then when this capacity increase was met by the weak economy of 1831, One, price has plummeted. That then was a contributing factor in the worker uprisings known as the canoe revolts. This story reminds me of one told by Warren Buffett, also about the textile industry, when Buffett was running Berkshire during its early years as a textile company in New England.

And he said the firm made large capital expenditures in a new technology that allowed, in principle, for Berkshire to reduce costs and therefore grow profits. This is what Buffett writes. But the promised benefits from these textile investments were illusory. Many of our competitors, both domestic and foreign, were stepping up to the same kind of expenditures. And once enough companies did so, their reduced costs became the baseline for reduced prices industry-wide. Viewed individually, each company's capital investment decision paid cost-effective and rational, viewed collective, objectively, the decisions neutralised each other and were irrational, just as happens when each person watching a parade decides he can see a little better if he stands on tiptoes. After each round of investment, all the players had more money in the game and returns remained anemic.

And that's a classic capital cycle tale. The Jackard story in Buffett's subsequent experience shows that new technology can boost productivity, while actually destroying an industry's underlying profitability. But what exactly, Tom, is the relation to the software apocalypse today? There's sort of two links to software. The first is the historic link, which is that the punch cards used in these machines were the inspiration for Charles Babbage's analytical engine. That was sort of designed but never constructed.

And then Ada Lovelace, who is considered by some to be the first computer programmer made some algorithms, which were then very influential in the creation of modern computing, and the remnants of this could still be seen into the 1970s when punch cards were still used to write software. The second way it's relevant to us is how it's relevant to us today. Like the French weavers in the early 19th century, the role of the software developer is changing and productivity is increasing. At the most bullish end of the spectrum, at Davils this year, the CEO of Anthropic, Dario Amodi, said that he had engineers who didn't write code anymore. I'll quote him, he said, we might be six to 12 months away from when the model is doing most,

maybe all of what software engineers do end to end. Was that just the type of AI hype that we have come to expect on a daily basis? Yes, well, he is not exactly a neutral source on the matter. You know, Anthropic IPO is probably in the pipeline. But even more pessimistic commentators believe that there will be a, meaningful productivity improvement in software development. And I think we can think about this in more classical capital cycle terms, the capacity to produce software is increasing. I'll say, well, there's already some evidence of this. So data from app figures, which is a mobile

analytics company, shows a 24% increase in the number of iOS apps released in 2025. And that takes it to the highest level since 2017. And we've got to be a bit careful because there is this new thing, which is AI and of course that generates more apps. Correlation isn't causation, but I think when you look at the acceleration in monthly app production, that coincided pretty much exactly with the launch of agentic coding tools at the start of 2025. The other thing is that many software companies have long list of products and features, which they never get around to making, and more of these can now be rolled out faster to customers. And you believe that capacity increases are unlikely to be an entirely benign,

development? Yes, I think as software companies add features, they will start encroaching on each other's territories, and also you'll have start-ups which can roll out copycat products faster than ever. As the capital cycle tells us, more capacity means more competition and weaker pricing power. And another note of caution you advise before rushing into this underperforming sector is that while valuations have fallen from the nosebleed sector valuations of 2021 during the everything bubble, they're still pretty high. Absolutely. So take ServiceNow, for example. And can you just remind me what ServiceNow does? They started off as cloud native IT help desk software, logging tickets and managing the help desk workflow.

And that's still the lion's share of their business today. but they've expanded the offer into other areas which have workflows that look a bit like IT, so HR delivery, finance, and then also customer service workflows. And this has been a market-darling software stock. Before the share price fall and last year, if we take the peak, it had compounded at over 30% since the 2012 IPO. Now, the share price has fallen 50% in the past year, but coming back to valuation, it still trades at eight times the 2025 revenue and 62 times the GAP PE. And I use Gap because unlike ServiceNow's management, Marathon consider share-based compensation to be an expense.

Then looking more broadly, the S&P 500 Software Index trades at 10 times the 2025 sales estimates. And it's a little bit contentious how you do this. But roughly speaking, once you properly expense share-based compensation, it's trading it over 37 times. So even with the share price declines, the valuations across the sector are still quite expensive. And the reason the market is negative on software is not because of the capital cycle story you were telling all the high valuations, but because of the concern about vibe coding. Can you tell me a bit of that is concerns? Yes.

So that's probably the concern you see most. And that's that non-programmers will vibe code, bespoke application. Can I ask you, why do they call it vibe coding? The word vibe coding comes from a tweet. You know, what it means is that non-programmers can create bespoke applications rather than buying off-the-shelf products. We have some experience, it's pretty limited, but we have some experience at Marathon of using vibe coding,

and it is very impressive. As a non-specialist, you can now easily produce simple applications. Many years ago, I learned JavaScript and HTML and CSS, and that's the programming languages you need to make. web-based applications. And there are tools now, and the one that I've used is called Lovable, which allows users to build web-based applications, including, by the way, integrating a sort of off-the-shelf back-end solution with a prompt of just a few sentences to an LLM. And if you've used it, you see that it is extremely powerful. That said, it seems unlikely that large companies, or for that matter,

medium-sized companies are going to vibe code up their own versions of mission-critical payroll or engineering design software. Things like brand, reputation, migration costs, technical expertise, cost of failure, all of these things still function as moats. However, I think coming back to that capital cycle point, those moats are probably narrower in a more intense competitive environment. Another concern of the market is that AI will redefine how users interact with software. Yes. Much of the value in software is packaging data to humans with a good user interface. If humans are interacting with AI agents who in turn interact with the underlying data,

that disintermediates or at the very least massively changes the software front end. And software companies have got to adapt to that. or they'll lose out to newer entrance. And you think these changes will have implications for software companies that charge clients on a per seat basis? Yes. If AI generates the promised efficiency savings, that means fewer users of the software, and that is a decline in volume on which to charge fees.

So the concern about the per seat charging model is that volume reduction, and those efficiency savings might end up accruing to the customer, or to the foundation model provider rather than the software provider. The solution is to charge by some other metric, and that's most likely going to be consumption. But the problem with that is what we've seen is quite dramatic deflation in costs per unit of intelligence. And so charging by consumption in a competitive environment could lead to topline deflation as the technology and compute costs fall. So despite this negativity, Marathon's statement.

still owns some application software companies in its client portfolios. Yes, that's right. So take Autodesk, for example, so they produce design software for engineers. And architects as well, don't they? Yes, it's a relatively broad business. But the sort of core is that kind of design software. It's highly technical. It's mission critical for the engineers that are using it.

There cannot be a mistake. The cost of failure in a mistake is huge. So new competitors, they don't have the reputation. They probably don't have the domain expertise. The engineers are trained on auto-desk solutions, and there's an enormous cost and hassle of migrating large teams of engineers to a competitor or new solution. And that creates a pretty high switching cost.

So we think those kind of software companies are better protected against rising capacity and competition. So to go back to my comments at the start, I think this is a good parable on the dangers of buying tech companies at high valuations because they're priced as if nothing can ever go wrong. And then, lo and behold, something happens, particularly in the tech world where things are changing constantly. So when investors, they keep on having to learn the same lesson with tech investing is not to pay too inflated a price in case the scenarios of current profitability. has suddenly changed. Anyhow, what's your final conclusion thinking about these software services businesses? Yeah, so it's been a very weak nine months for software stocks. There is some commentary from brokers and in the media suggesting that maybe in some areas sentiment is improving. I think some of the concerns, as we've discussed, are overblown and the companies continue to

report good growth. However, our capital cycle investment approach tells us to be cautious about entering an industry where competition and capacity is expanding. Great. Well, thank you very much for that, Tom, and look forward to speaking to you again soon. Great. Thank you very much. Thank you for your time today. I hope you will listen to the next edition of the Capital Cycle. This communication is provided for information purposes only. Please refer to Marathon's website and the Global Investment Reviews for further information, including important disclosures.