With US software stocks down — checks notes — 30% in four months, it’s fair to ask if the “SaaSpocalypse” narrative’s overcooked.
I don’t doubt, at all, that agentic AI has the potential to disrupt software businesses. But the indiscriminate selling seen since Anthropic unveiled Cowork in January seems to overlook the potential for those businesses to incorporate AI rather than be antiquated by it.
Nothing’s free in this world. As the chart below reminds you, the AI buildout is the most expensive undertaking since the Louisiana Purchase when expressed as a share of contemporaneous US GDP.
This moonshot, I remarked earlier this month, is far and away more expensive than the actual moonshot.
Currently, AI services may be free, or next to free, for the average netizen. But for complex applications, it’ll come at a cost eventually. The cost of this technology is running into the trillions. The idea — and this is implicit in some of the SaaS selling — that AI’s going to be free, or something close to free, for any and all possible purposes assumes no one involved in financing and developing it will ever demand a return on their investment. That’s absurd.
Sure, a lot of agentic tools will be more or less free and for some service providers and software businesses, that’ll be a death knell. But formalized, comprehensive assistance on tasks that require precision under penalty of, for example, audit, in the event of a mistake, won’t be completely free.
It’s far more likely that Anthropic, OpenAI and all the rest, partner with existing software companies to incorporate their technology into paid services. TurboTax, for example, has an “AI assist” feature already. And for God’s sakes, folks, have we all forgotten? Microsoft, early AI adopter par excellence, is a software company.
I bring this up because the disparity between forward earnings expectations for software stocks and the indiscriminate selling mentioned above is quite something to behold. Have a look:
In two words: Somebody’s wrong. In three: Somebody’s badly wrong. The disconnect there’s as wide as it was during the early-2025 tariff scare, with the difference being that was a selloff catalyzed by an exogenous shock, whereas this is (purportedly) a fundamentals-based de-rating.
I say “purportedly” because with apologies to “efficient” markets, traders are almost surely assuming far too much for agentic AI and far too little for these businesses, particularly considering my point above about AI incorporation into existing products.
Markets are trading agentic AI success versus software viability as though the two are mutually exclusive. I don’t think that’s right. It might be wholly wrong, which is to say agentic AI might compliment a lot of software businesses.
“While the last few weeks have been characterized by investor concerns about AI disruption, the software stocks at the center of those fears have generally reported earnings results that exceeded consensus expectations and drove analysts to lift forward estimates,” Goldman remarked, adding that over the past three months, software stocks “have plunged by 24%, but two-year forward earnings estimates for the stocks have risen by 5%.”
Again: Somebody’s wrong.
Couldn’t agree more, for example, software company UiPath has been caught up in this rotation as a more traditional vendor, but it has crucially released agentic AI in its product line up with its Maestro product that has been received exceptionally well by customers. Its AI capability works alongside its deterministic robotic process automation capabilities and allows clients to make use of AI in a more seamless manner.
The shares are trading on a 2026 PE of 14.7x earnings and have fallen circa 45% since December. Bonkers!
Software professional here. I agree. It compliments these businesses.
These headlines purporting that AI is going to do literally everything are in the dog and pony show so that the IPO goes big. It’s always a big tech or prominent tech company executive who is going to get very very rich when these companies IPO.
Have we seen the test where we ask AI whether we should drive to the car wash when it’s 50 meters away. LLMS are going to hit their limit. God bless them. It’s truly a great time to be an employed software professional in a good culture.
I have not and don’t plan on betting against AI, especially when teamed up with the likes of Microsoft, Google, Nvidia, Broadcom, etc.
GLTA.
As for costs, I’m reading and hearing a bunch of comments recently about how much it’s costing in some companies. Those who are racing to build automations and really smart tools from this really smart tool, are running into expensive overruns when they need to buy more tokens (essentially processing power) to work on these models. And they’re not even paying for the extra energy load yet. That push seems to be on the horizon.
But clearly speed to market in this era may make the dot com boom seem pokey in comparison.
A lot of SaaS stocks are just abstract constructs to investors, who usually have never used the software themselves or even seen anyone use it. The companies report lots of metrics, some more made-up than others, but at best they tell you how the company is doing now and its visibility for a year out. The concern is further out than that; no-one thinks AI is seriously affecting SaaS sales “right now”. Valuations have come down, a lot in some cases, and may look cheap for 30% growth and 25% net margin, but suppose growth becomes 20% and that’s a non-GAAP margin. So most investors are trying to guess the threat to software that they barely know anything about, from a technology that they also don’t understand with a future potential they can’t know. The confidence is not very high. The FUD is high. The AI propaganda is relentless.
It is tempting to be a hero and catch some falling SaaS knives, but as AI fear cycles through different industries, there are alternative catchable knives that aren’t SaaS, are in the physical world, more concrete to investors and where you don’t have to pretend that stock comp isn’t real. I’ve recently done a bit of the former but a lot more of the latter.
” So most investors are trying to guess the threat to software that they barely know anything about, from a technology that they also don’t understand with a future potential they can’t know.”
+1
That said, the same holds for AI as well, no?
Yes. Thus, in a scenario of “no landing” and AI risk & uncertainty it makes sense to own more “real economy” names and less tech. Not that technology and AI are the same, but in market cap terms they may as well be. So much market cap in tech that even a moderate shift in this direction is a major tailwind for the recipients. Of course, if “no landing” is wrong then . . .
This is a classic plow and plow shares example. Do not invest in plow shares invest in plows, plow owners or suppliers to plow owners. Someone has to store and maintain all the data ai develops. Memory chips, computers, data warehouses and other long term data storage.