The software-as-a-service sector has spent much of the recent market recovery stuck in the red, and the reason can be summarized in two letters: AI. The pressing question for anyone watching the space is whether artificial intelligence is a friend or a foe to software companies. The honest answer is that it is unambiguously both — and understanding why requires moving past the headline panic to look at what AI actually changes.
Friend and Foe at Once
AI is a friend in the sense that it dramatically improves productivity. It is a foe in the sense that it disrupts current markets and the business models built around them. The crucial insight, however, is that AI is not replacing software itself. What it is replacing — or rather, reshaping — are the workflows that people use software to accomplish. AI changes the tasks a person does, and it does so in a way that is genuinely disruptive. The downstream consequence is simple but profound: a person assisted by AI may no longer need as many software tools to get the job done. The threat to software vendors is not that code becomes obsolete, but that the number of discrete tools required to complete a workflow shrinks.
This is why the impact is so uneven across the sector. Some businesses are highly disruptable. Website development, for instance, is a fundamentally replaceable skill — exactly the kind of generic, commoditizable task that AI handles well, which is why companies built around it have seen their valuations punished. On the other end of the spectrum sit firms whose products are far harder for AI to dislodge. For the right company, AI is not a threat at all; it is the entire point. It takes an already good tool and makes it better.
The Vulnerability Test: How Close Are You to the Customer?
If AI separates winners from losers, what distinguishes them? The clearest dividing line is how deeply a company understands and participates in its customer's final output. The most vulnerable players are those that don't understand the background of what their end customer is actually doing. A generic software company, offering generic functionality, with no real grasp of the business process it serves, can simply be replaced by a generic AI that does the same thing.
The strategic imperative that follows is to get as close as possible to the customer's business process. The relevant questions are: How much do you understand about what your customer is trying to accomplish? How much do you genuinely matter to that outcome? How connected are you to the way your end customers actually make money? Companies that can answer those questions with depth and specificity have a moat. Those that cannot are exposed, because their value was always interchangeable, and interchangeable is precisely what AI commoditizes first.
Picks, Shovels, and the Next Leg
For investors, a useful historical frame is the "picks and shovels" strategy. During the technology, media, and telecom boom, the durable money was made not on the speculative end but on the infrastructure that everyone needed regardless of which application won — the Cisco-style plays. In the current cycle, semiconductors are that pick-and-shovel layer, the thing everyone has rallied around because that is where the most significant interest sits.
The natural next question is what comes after semiconductors. The answer points back toward software names — but with a twist that traditional investors find uncomfortable. Software has long been prized as a capital-light business: minimal heavy investment, high margins, predictable returns. AI breaks that template. Companies are now pouring capital into AI infrastructure, and investors accustomed to the old model are scared to death of these capital spending plans. Yet some of these firms have been investing in AI for a long time; it has only recently become a perceived problem. That reframing matters. Spending money on something that is going to end up making money is not a betrayal of the software model — it is simply ordinary business investment that happens to look unfamiliar on a software balance sheet. Viewed through that lens, several of these heavy spenders look less like cautionary tales and more like long-term AI winners.
Two Ends of the Same Trade
This produces a barbell view of where value lives. In the immediate term, owning the infrastructure — the pick-and-shovel layer — is the most interesting play, because that is where the concentrated interest and momentum reside. Over the longer term, though, the more important variable is not which specific stock or ETF you pick, but how well a company is connected to genuine customer value. The infrastructure play is the near-term answer; customer attachment is the durable one.
A practical concept ties these together: attachment rates. Some software embeds itself so deeply into a customer's operations that it cannot be pried loose. A customer relationship management system, for example, is extremely hard to separate from a sales organization — its attachment rate is high, which makes it resilient even as AI churns through the rest of the stack. This is what makes certain traditional software names attractive on the opposite end of the AI trade from the pure infrastructure plays: not because they are flashy AI stories, but because their stickiness protects them in a world where flimsier tools get absorbed.
The companies occupying the most interesting middle ground are those that combine heavy AI investment with this kind of durable customer relationship — firms that are simultaneously potential AI winners and entrenched incumbents. They unsettle investors precisely because their spending defies the old capital-light expectation, but that spending is the price of staying relevant. Buying into that story right before an earnings report can be nerve-wracking, yet the underlying logic is sound: the firms doing well in this new world are the ones investing money today to make money tomorrow, while staying close enough to their customers that no generic AI can simply take their place.
Conclusion
The so-called SaaS-pocalypse is less an extinction event than a sorting mechanism. AI is not killing software; it is exposing which software ever mattered. Tools that automated generic tasks for customers whose businesses they never understood were always living on borrowed time. The survivors — and the long-term winners — will be those embedded so deeply in their customers' value creation that they are inseparable from it, and those willing to spend real capital to own the infrastructure and intelligence that the next era runs on. The capital-light fantasy is ending. What replaces it is a harder, more honest test: prove that you matter to the people who pay you, or be replaced by something that costs a fraction as much.