Investors keep having the same argument about AI, but it’s the wrong one. One camp says it is a bubble, the other says it’s the future. But both are trying to win a debate that does not need a winner.
Two things can be true at the same time. There can be a genuine, society-altering technology and a very expensive, very crowded trade sitting on top of it. Short-term returns and long-term potential do not always line up. In fact, they usually don’t.
As I like to point out, railways were very much a civilisational game-changer. So were electricity, aviation, telecommunications, and a load of other things besides. Each of them produced an investment boom that we later called a bubble. And a lot of people who believed in the technology, for very sensible reasons that were mostly validated in time, were taken to the cleaners.
They weren’t wrong about trains. They were wrong about where the value unlocked by that technology would accrue. The graveyard of railway stocks in the early twentieth century, or of internet stocks a century later, was not evidence of a failed technology that didn’t live up to its promise. Indeed, all the major technological breakthroughs produced great fortunes for those who were smart or lucky enough to get in at the right time, and who backed the right players. Most of which, by the way, were downstream of the core technology itself.
So, as usual, there’s a lot more subtlety to the AI debate than at first seems apparent. Instead of asking whether AI stocks are cheap or expensive, it’s more interesting to consider what the broader economy looks like in a world where high-quality intelligence is cheap, abundant, and sitting in everyone’s pocket, and then work backwards from there.
When I try to reason it all through, I keep ending up at the same place. Value accrues to where true scarcity lies. Which, if you think about it, has always been the case. It’s just that the future we are hurtling towards is one in which there is much less scarcity left in expertise.
And when the sort of thing that used to live in expensive heads working at expensive firms becomes a utility, well, the civilisational bottlenecks become the stuff that intelligence, even extreme intelligence, still has to push against. Energy. Commodities. Proprietary data. And, for lack of a better word, judgement.
The first two only insofar as demand can continue to outstrip the inevitable supply-side response. Frankly, demand may be strong enough, for long enough, that it proves an interesting thematic. But proprietary data and judgement are where the real moats will be found. Proprietary data because it is the one input the model cannot hallucinate into existence. Judgement because it decides what to do with the answer. One is a dataset. The other is a person. Neither gets cheaper just because the brain on tap did.
The public internet is already in the models. What isn’t is the data that never left the building: claims files, sensor feeds, patient records, the ugly operational stuff that tells you how a business actually works. Intelligence without that is just a very fast reader of what everyone else can already see.
Judgement sounds a bit wishy-washy, but I think it is the one that matters most. These models are already frighteningly good. They have depth. They do not get tired. They will grind through the tax act, the case law, the 400-page contract, without a coffee or a complaint. What they do not have is true agency. They do not wake up and decide where capital should go, or which problem is worth solving, or what would actually be valuable to another human being. They wait for a prompt, then they do their best. Someone still has to steer. And that requires judgement.
Two different people using the same model can produce wildly different outcomes. One person produces a mountain of useless AI slop, the other leverages it to cure cancer. The model is the same. The difference is taste, priorities, knowing which question to ask, and when to ignore the answer. That’s not a feature you can download. It’s accumulated, and it’s unevenly distributed.
I am leaving generalised superintelligence out of this. The god-like takeoff scenario is fascinating, but it’s a different conversation. Deciding how to invest in that world is like asking a chicken what it thinks of the latest developments in semiconductors. But if we stay on the path we’re actually on — models getting better and cheaper, still lacking true agency, still unable, on their own, to want anything — a few things follow.
The first is that a lot of the old moats start to look pretty flimsy. In the past, only the already wealthy could access the best lawyers, the best accountants, the best advisers. The labyrinth of documentation, regulation and legislation was a huge barrier if you couldn’t afford the right helpers. Quality advice was scarce, and therefore it was valuable. But that’s rapidly changing.
Case in point: Greg Baker, a computing academic at Macquarie University, recently represented himself at the Fair Work Commission with, by his own account, close to zero knowledge of workplace law. He used AI agents to build the case and ultimately won against the university’s top-tier legal team. In the old world, he wouldn’t have stood a chance without (very expensive) professional legal help.
Businesses whose service is navigating complex rules, and that need highly trained humans to do it, are facing something like what various blue-collar roles faced at the start of the industrial revolution. A lot of SaaS companies are in trouble too. Their value proposition was, in the main, about giving us a nicer interface, easier access, and automations that made pen, paper, ledgers and filing cabinets obsolete. When anyone can spin up their own software by just having a chat to a coding agent, that value proposition is no longer as obvious.
Software built around proprietary data, or that enjoys network effects (two-sided marketplaces and the like), is probably fine. Anything else will, I suspect, struggle to maintain pricing power.
Xero (ASX:XRO) is a potential example. I’m not calling time on it this year, or next. But we are getting to the point where a few more generations of models will make doing the taxes and keeping the books about as difficult as forwarding an invoice to an agent and saying “deal with it.” Who wants to spend hours clicking through screens and lodging forms when an agent can do it all for you?
Remember, Xero does not write the tax code or the accounting rules. Those are open and available to anyone. Yes, it has a network of accountants, and that’s its strongest moat today. But that extended network is being directly challenged too. What’s the moat when I have a tireless expert accountant and tax law specialist on hand 24/7? I could be wrong (and probably am), but when most of the value of these kinds of companies is based on a long tail of high-margin future cash flows, it just seems a little tenuous.
Even the leading AI labs themselves may not have much of a moat. The tenth-best system in the world today isn’t nearly as good as the leading frontier models. But they are as good as the bleeding edge was a few months ago, and are accessible for a fraction of the price. That’s a super narrow lead to defend. Maybe this is a winner-take-most market. But if capability keeps diffusing through the industry, it’s hard to spot the durability of any current competitive advantage.
None of this means there is no money to be made. It means the money is probably not where the crowd is pointing. If what seems “obvious” is right, then it’s largely priced in. If it’s “obvious”, but wrong, well, it’ll end up as it usually does. It’s the unobvious thing that’s right that’ll mint the fortunes of the next era.
Strawman is Australia’s premier online investment club.
Members share research & recommendations on ASX-listed stocks by managing Virtual Portfolios and building Company Reports. By ranking content according to performance and community endorsement, Strawman provides accountable and peer-reviewed investment insights.
Disclaimer– Strawman is not a broker and you cannot purchase shares through the platform. All trades on Strawman use play money and are intended only as a tool to gain experience and have fun. No content on Strawman should be considered an inducement to buy or sell real world financial securities, and you should seek professional advice before making any investment decisions.
© 2026 Strawman Pty Ltd. All rights reserved.
