INVESTOR INSIGHTS

2026

Software Stocks Aren’t Facing One AI Threat. They’re Facing Three.

Fewer seats, higher running costs and cheaper competition. Any one of them would worry us. All three at once is why Frontwater owns almost no software as a service (SaaS) companies today.

Image generated with AI

INVESTOR INSIGHTS

2026

At Frontwater, we usually write about the companies we own. Sometimes the more interesting story is the companies we deliberately choose not to own, because a portfolio is defined as much by what it leaves out as by what it holds.

Software as a service (SaaS) companies rent their software by subscription rather than selling it outright. For now, we are watching them from the sidelines. We are doing so even though the sector’s valuations, measured by the price investors pay for each dollar of sales, have fallen to levels last seen in the mid 2010s.

Outside of Microsoft and two cybersecurity companies, Frontwater client portfolios hold no individual SaaS companies. That is not an accident, and it is not a call that software is finished. It is a judgment that the risk of owning the sector has changed, and that in a world of AI these companies have far less pricing power than they used to.

Consider last week. On Friday, October 23, ServiceNow closed at $158.96, slightly above where it started the year. One earnings report later, the stock fell about 27% on October 29 and finished the week down about 34%. The company had grown subscription revenue 21%. By most traditional measures, it was a good quarter. But as Adobe shareholders have learned since the start of 2025, the market prices the future, not the past, and it is the future that has investors worried.

A good quarter is no longer protection, because the threat is not coming from one direction. It is coming from three at once. Here is how we see each of them.

Reason One: Fewer Seats

Most business software is sold by the seat: a monthly fee for every employee who logs in. For two decades that was a beautiful model. Customers hired more people, and the software company grew with them, without lifting a finger.

AI agents, software that can carry out a task on its own rather than help a person do it, put a question mark over the whole arrangement. If an agent can handle the work of ten sales reps, a customer may need ten fewer seats.

INVESTOR INSIGHTS

2026

Salesforce is the clearest example, because it sits on both sides of the trade. It sells seats to salespeople, and it now sells the agents that could reduce how many salespeople a customer needs. Its agent product, Agentforce, has been growing quickly: annual recurring revenue, the yearly value of subscriptions in force, crossed $1.2 billion, up about 205% from a year earlier. That is impressive growth. It is also, by our arithmetic, roughly 3% of the company’s last full year revenue of about $41.5 billion.

What it means for us: the new business is growing fast, but it is small, and nobody yet knows whether it grows faster than the old business shrinks. We are not willing to pay for that answer before it arrives.

Reason Two: Running Costs Are Rising

The second threat is quieter and, in our view, more important. It is not about whether customers keep paying. It is about what it costs to serve them.

A traditional software seat costs almost nothing to deliver. Once the code is written, the thousandth customer is nearly free. That is why gross margins at leading software companies, the share of each sales dollar left after the cost of delivering the product, have tended to sit above 80%.

An AI agent is different. Every time it works, it uses computing power, and someone pays for it.

ServiceNow’s quarter shows what that looks like. According to the company’s October 28 release, gross margin on subscriptions slipped to 80% from 83% a year ago, and operating margin on a GAAP basis, the standard accounting rules, fell to 9.4% from 16.8%. Part of that squeeze comes from integrating Armis, a security company ServiceNow agreed to buy for about $7.75 billion.

In our view, the rest comes from a change in how ServiceNow makes money. It used to charge a fixed fee for each user, and serving one more user cost it almost nothing. Increasingly, it charges customers for how much they use its AI, more like a utility bill than a subscription. Chief executive Bill McDermott told analysts the company has spent many quarters building this pay for what you use model into the business. The catch is that usage based revenue comes with usage based costs. Every time a customer’s AI agent does a piece of work, ServiceNow pays for the computing power behind it.

What it means for us: the cloud giants are spending hundreds of billions of dollars on data centres, and somebody has to pay them back. Increasingly, that somebody is the software industry, one AI query at a time. We would rather own the company collecting that bill than the one paying it.

INVESTOR INSIGHTS

2026

Reason Three: Competition Is Getting Cheaper

The third threat is the most existential. If AI makes software dramatically cheaper to write, and the work it supports dramatically easier to do, what stops a customer, or a startup, from replacing it?

Not everything, and not everywhere. Some software is simply too important to get wrong. We do not expect companies to start writing their own cybersecurity, because the cost of a mistake is a major breach, and that is why cybersecurity is the one corner of software we continue to own. Nor are many companies about to build their own version of ServiceNow. But even where customers stay, the economics have changed. A company that once set its own prices now has to defend them against cheaper AI alternatives, while paying more to deliver its product. In the language we have used for some time, it is moving from price setter toward price taker.

Adobe shows the concern from the customer’s side. For three decades, professional creative software was protected by a steep learning curve. Mastering Photoshop, Illustrator or Premiere took years, and once a designer had invested that time, they paid for the seat. Today, a growing number of people can produce a marketing image, an illustration or a short video clip by typing a sentence into tools from companies such as Midjourney, Runway and OpenAI.

Adobe’s software remains far more capable than most of those tools. That is not the debate. The debate is whether AI reduces the expertise needed to produce professional work and, with it, the moat around the software itself.

The business has kept performing. Revenue in the first quarter of its 2026 fiscal year was a record of about $6.4 billion, up 12%, and the company reported that its AI first annual recurring revenue had more than tripled. Adobe is also fighting back on the second front, building its own Firefly AI into its products and charging for heavy use through generative credits, a monthly allowance of AI generations that premium features draw down. That is a sensible response, but it also means Adobe now carries the second threat too: it pays for the computing power behind every one of those generations.

The shares tell a different story. By late April they were down about 31% for the year, and in April Mizuho downgraded the stock, saying it expected organic growth no higher than the high single digits over the next two to three years. In late October the shares still traded around $264, roughly 25% below where they started the year, by our arithmetic.

INVESTOR INSIGHTS

2026

The market is not questioning whether Adobe is a good business. It is questioning whether the economics of creative software will look the same five years from now.

Constellation Software is the Canadian version of the same debate, from a very different business model. Constellation owns hundreds of small vertical market software businesses: specialist programs that run one kind of organization, such as a transit authority, a medical clinic or a municipal tax office. It buys them, keeps them and lets them compound.

The market sold it anyway. The shares fell from about C$5,060 in June 2025 to a low near C$2,196 in January 2026, the deepest drawdown in the company’s history. Two things drove it: AI fears, and the departure of its founder and long time leader, Mark Leonard. In early October the stock closed around C$2,950, still roughly 40% below its peak by our arithmetic.

The case for resilience is switching costs. A town does not rip out the system that runs its property tax rolls because a cheaper one exists. The data, the training, the integrations and the risk of a botched migration all make customers stay. Management has said it has seen no material impact from AI on the business to date, and its businesses kept generating cash through the first half of 2026.

The case against is that switching costs erode slowly, then quickly. AI lowers the cost of building a replacement and of migrating the data, and some investors have also questioned whether capital allocation will change under new leadership.

What it means for us: in both cases the debate is not whether the company is good. It is whether the moat is changing. When even the most defensible software models in North America are a genuine debate, the risk is hard to price, and we prefer not to pay for risk we cannot price.

The Exceptions: Microsoft and Cybersecurity

Microsoft is the one large software company we hold, and the reason matters. We own it for Azure, its cloud infrastructure business, which sits on the other side of this story: it is one of the companies that gets paid every time an AI agent runs.

Microsoft also sells a great deal of software by the seat, through products like Microsoft 365, and it is not immune to the first threat. We hold it for the infrastructure and accept the software exposure as part of the price of owning it.

INVESTOR INSIGHTS

2026

Our other exceptions are CrowdStrike and Palo Alto Networks, both sold by subscription. We hold them because cybersecurity sits on the right side of every threat in this article. Customers are unlikely to cut protection when they adopt AI agents, since every new agent is another door an attacker can try. They are unlikely to build their own, for the reason above. And when a breach can cost far more than the software, price matters less than trust. That does not make these companies immune to the risks we describe, and we hold them knowing that.

What Would Bring Us Back

Stepping aside is not the same as walking away. Plenty of software businesses are growing, profitable and adapting, and the market may well end up overpricing the threat for some of them. We will look again when companies can give clear answers to three questions:

1. How does it charge? Per seat, per use, or per outcome? A company moving from seats to usage is not failing, but its revenue becomes less predictable and its costs become more visible. We want to see the new model working, not promised.
2. Is gross margin holding? This is the early warning. Revenue can keep growing for years while the cost of delivering it quietly rises underneath.
3. What would it take to leave? Net revenue retention, the revenue a company keeps and grows from last year’s customers, tells you whether switching costs are real or just assumed.

None of these questions requires a forecast. They require evidence, and today too few software companies can supply it.

The Bottom Line

Software companies are being challenged from three directions at once: the seat model is being questioned, the cost of delivering AI products is rising, and the cost of building a competitor is falling.

ServiceNow, Salesforce, Adobe and Constellation Software have different products, different customers and different business models. Yet the market has questioned all four for essentially the same reason. Investors are no longer asking whether these companies are growing. They are asking whether the economics that made them exceptional will hold. Any one of these concerns would justify caution. Together, they change the risk of owning the sector.

INVESTOR INSIGHTS

2026

That is why Frontwater owns almost no software today, and why the few software companies we do own sit on the right side of those threats. Not because we think software is dead, but because we do not yet know which companies will come through it, and we would rather wait for the evidence than pay for the hope.

So here is the question worth asking about the software companies in your own portfolio: do you know which of the three threats each one faces, and whether you are being paid to carry it?

If you would like a second set of eyes on that question, book a complimentary 60-minute introductory meeting, by Zoom or in person, at www.frontwater.ca.