Over the past year, no industry has come under greater scrutiny than software. The once unstoppable cloud business model is facing new threats from artificial intelligence. Former highfliers like Salesforce and Adobe are being forced to prove their relevance. Their stocks get hammered with every rumor and announcement from AI labs. Despite a spring rebound, the closely followed iShares Expanded Tech-Software Sector exchange-traded funds is still down 15% over the past year.
But software isn't going away, and there are companies that will survive and thrive through the AI transformation. After a year of mostly indiscriminate selling, investors are becoming more selective, recognizing there is nuance within software when it comes to AI.
So far, despite the shock of software's selloff, the pattern that's playing out is consistent with prior technology inflections, in which hardware wins first. We see that today in AI, with chips and cloud stocks soaring; the PHLX Semiconductor Index is up 108% in the past year.
This wave of spending also forces a trade-off: Higher capital expenditures on hardware means there's less money available in IT budgets for software purchases. That reality came into sharp focus this past week when IBM released disappointing results that seemed to stun investors.
"In the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases," CEO Arvind Krishna said in a letter to shareholders.
Customers bought fewer IBM mainframe computers than anticipated, and less associated software. The stock fell 25% on the announcement. The good news is that other software stocks held tight -- evidence of software's next, more discriminating phase.
Eventually infrastructure -- even the AI models themselves -- will become commoditized and the so-called application layer, i.e., software, will take over in importance and profit. The process will take years, but some existing software players will be among the winners.
As investors make their peace with software's future, they'll need to rethink what software means in an AI-first world. Lots of AI-native start-ups have gotten a head start on doing exactly that. They're unencumbered by legacy baggage, from existing sources of revenue to bloated costs and code. But technology disruption doesn't entail wholesale replacement -- see Walmart's e-commerce success, Microsoft's cloud prowess, and JPMorgan Chase's fintech bona fides.
The Software Apocalypse
The cloud software model that emerged over the past two decades has been wildly successful. Its user-based subscriptions were a good deal for customers thanks to continuous software updates and lower internal IT expenses. It was an even better deal for the cloud providers themselves, which saw recurring revenue and big profits. The marginal cost to the vendor for each new user is low. Consequently, gross profit margin is very high -- somewhere between 70% and 90%. For comparison, Apple, one of the best-run companies in the world, has gross margins below 50%.
In 2011, venture capitalist Marc Andreessen observed that "software is eating the world," and he was right.
Through the 2010s, VC and private-credit investments flowed into the sector, eclipsing other industries. These software start-ups -- many of which are now giant public companies -- typically saw high sales growth, coupled with mounting free cash flow.
But AI has thrown a wrench into the machine, putting the outsize growth and cash generation at risk. Now it's AI that's eating software.
In the enterprise, AI's earliest use case has been writing software, which is leading customers to make their own custom applications. In its latest earnings call, software company Palantir Technologies claimed it had replaced its customer-relations management vendor with an internally developed application. This may be a particularly appealing path for smaller companies that see large software packages as overkill.
But the bigger threat may be agents, which can use AI models to accomplish a complex series of tasks from a simple conversational prompt. Bots already outnumber people on the internet, and if trends continue, there will soon be far more agents on enterprise networks than humans. Machines replacing humans is a worst-case scenario for companies that charge by the person.
Moreover, as these agents continue to develop, they will begin to supplant more software functions, and make customers wonder why they are paying for subscriptions at all. The good news for tech investors is that software companies are already transitioning to hybrid sales models that include revenue based on AI consumption. The bad news is that consumption models carry a lower gross margin, depriving software of its profit-generating superpower.
Already Winning
Investors frequently talk about the "software industry" as a monolith tracked by a broad index like the iShares Expanded Tech-Software Sector ETF, which holds stocks in over 100 companies. But underneath the "software" rubric lies a lot of variability. For some software providers, the AI inflection is a boost for their business.
There are mountains of data that go into making AI models, but the key to effective implementation is the proprietary data that lives on enterprise networks. It's what fuels agent work, and it can be used to refine AI models for bespoke purposes. Software companies that help AI work with data look to be winners.
Oracle and its industry-leading database software would be the primary example, but the company is busy transforming itself into a cloud giant, and that has overtaken the core software story. The cleaner story is Snowflake, which provides a platform for enterprises to warehouse their cloud data, and combines everything into "data lakes" that are ready for agents to use. Moreover, Snowflake already has a consumption-based revenue model and the lower gross margin that comes with it -- 66% for the latest one-year period.
Snowflake stock is up 88% in the past three months, as investors begin the process of differentiating software vendors.
Then there's Palantir Technologies, perhaps the longest-running AI winner in software. The company makes structured data lakes for its customers, and builds custom applications on top of them. Despite some recent weakness for the stock, Palantir is bringing value to the data estates of enterprises. It's offering the increasingly important operating layer that links people, AI models, and enterprise data. Its 84% gross margin shows software profits and AI aren't incompatible.
Datadog is another example. The firm's software allows customers to automate the monitoring of their information technology, which includes agents and other AI infrastructure. Like Palantir, Datadog avoids user-based pricing, and still gets an 80% gross margin.
AI also creates a favorable environment for cybersecurity companies because the technology enables traditional attacks to be carried out at a scale and speed that humans can't match.
Enterprises are beginning to recognize the need for beefed-up protection. Last week, Microsoft used AI to patch more than 500 vulnerabilities to its own software, smashing the previous month's record.
Most companies, though, will need help. CrowdStrike Holdings and Palo Alto Networks offer the broadest security platform and continue to make acquisitions to fill holes in their AI security portfolios. Their stocks are up 95% and 112%, respectively, over the past three months. They'll continue to thrive as the AI threat grows.
Meanwhile, agents themselves have become an attack surface, with entirely new sorts of exploits, like prompt injections. It means enterprises will have to authenticate who -- and what -- they want running on their networks -- and issue permissions and restrictions. That fits in with Okta's wheelhouse. Its shares have risen 105% over the past three months.
The AI success for data and cyber companies comes with a trade-off for investors. The stocks are all pricey. Adjusting for share-based compensation, they trade at multiples of expected earnings ranging from 37 to 148.
Investors can find better deals elsewhere in software.
There's Still Hope
At the center of the software apocalypse are the full-featured cloud applications that still rely on the seat-based subscription model. This includes offerings from Microsoft, SAP, Salesforce, and other cloud natives like Atlassian, Workday, and ServiceNow. The stocks in this group are where investors can find the best value.
SAP trades at 18 estimated earnings for the next 12 months, well below the S&P 500's multiple. At 20 times, Microsoft trades roughly in line with the large-cap index, after years of fetching a substantial premium. Salesforce has a forward P/E of just 12, after starting the year at 20. In June, the stock was down 14 trading days in a row.
Executives at these firms will have to reread Clayton Christensen's The Innovators' Dilemma and Andy Grove's Only the Paranoid Survive and live the lessons therein.
"Business success contains the seeds of its own destruction," Grove wrote in his 1996 book.
AI is what Grove would have referred to as a "strategic inflection point, " which are "full-scale changes in the way business is conducted, so that simply adopting new technology or fighting the competition as you used to may be insufficient," he said. "A strategic inflection point can be deadly when unattended to. Companies that begin a decline as a result of its changes rarely recover their previous greatness."
The CEOs of these adapters have to rethink their value propositions and pricing models for this new world and be willing to sacrifice software's golden gross margin. The offset will come from cutting operating expenses, especially in sales and marketing, typically the largest expense. Winners will be determined by strategy and flexibility, which in the end comes down to leadership. Business as usual won't do.
The companies that have the best chance of success are the ones deeply embedded in their customers' business processes, namely Microsoft and SAP. Much of the unique data produced by enterprises live inside Microsoft and SAP software, which remain crucial links between people and machines -- not something AI will change. Their applications will be hard to dislodge, especially in government and highly regulated industries with strict requirements.
Microsoft also has AI servers for rent in the cloud and owns about a quarter of OpenAI, so it has other ways of offsetting losses in the company's business software. SAP is less dependent on per-user licenses than other vendors, accounting for less than 40% of its sales, according to the company.
Salesforce, one of the first cloud software firms, is AI's greatest wild card. Once the disrupter, it is now on the cutting table.
Salesforce seems willing to try anything, and disrupt everything. It has its own agents for sale under the Agentforce banner that deliberately cannibalize its own offerings. Annual recurring revenue for these agents remains relatively small, $1.2 billion at last count, but that has tripled from the year before. Salesforce had $43 billion in total sales in the past 12 months.
Salesforce has implemented more flexible pricing, combining subscriptions with consumption-based plans. It has been willing to put its apps into OpenAI's ChatGPT, helping the start-up to get between Salesforce and its customers. It made an $8 billion acquisition of Informatica last year to bring itself closer to the data infrastructure group described above. Informatica provides $1.1 billion in annual recurring revenue at last count.
Maybe most dramatically, Salesforce has released a "headless" version of its software, with no human interface. It is meant for agents from any vendor to operate, cutting people out the picture entirely.
Salesforce is showing the kind of hustle that every software firm will need to survive in the AI age. It's no guarantee of success, though -- it's the bare minimum. When Microsoft began preparing for the mobile revolution in the mid-1990s, it seemed to have every advantage -- cash flow, technology, and strong brand recognition. It released mobile operating systems for the next two decades, and they were all failures. Windows was once the dominant operating system in the world; now it's in third place behind Alphabet's Android and Apple's iOS.
"Coping with the relentless onslaught of technology change was akin to trying to climb a mudslide raging down a hill," Christensen wrote in the Innovator's Dilemma. "You have to scramble with everything you've got to stay on top of it, and if you ever once stop to catch your breath, you get buried."
Some software companies are destined to get buried. Picking the ones that can adapt will be profitable for investors.
Permanently Challenged
For some software makers, even hustle might not be enough. For companies in the creative field, AI represents unstoppable disruption. In the same way that software once revolutionized publishing and design, AI is changing the tools used by creative industries to make everything from Hollywood films to the flyers in bodega windows.
Unlike at Microsoft and SAP, Adobe's customers aren't locked in. They have other choices that are less expensive. One seat of Adobe Creative Suite costs $70 a month, much of which is now replaceable by a $20 ChatGPT account that can produce art at great scale and speed.
Adobe's profitability demonstrates its many years of success and impressive pricing power. Its gross margin was 89% over the past 12 months. But AI changes the equation. On a five-designer team, four can now use AI. If only one buys Adobe's software, Adobe's take-home pay falls precipitously.
AI is making life difficult for other more focused software makers, as well. So-called point solutions like Asana, Monday.com, and DocuSign make tailored software that solve a small set of problems. AI offers credible alternatives in each area. Homework help and educational site Chegg may be the best example. The firm identified the AI threat in 2023, shortly after ChatGPT's initial launch. With a stock that now trades at 78 cents, Chegg may have been the market's first AI casualty. It won't be the last.
New Kids on the Block
Software investors have focused on the frontier AI labs -- primarily OpenAI and Anthropic -- as the source of the threat, and with good reason. Though their cloud computing expenses are astronomical, sales are growing rapidly, and they're likely to be public before long.
But we are already seeing a c onvergence in AI model capabilities, and there seems to be a new best product every few weeks. The top models from six different labs, including Chinese Zhipu AI, are clustered near the top of benchmark rankings. There is a range of pricing, with Zhipu and Elon Musk's SpaceX charging much less than OpenAI and Anthropic. For the AI model makers, it isn't a great sign to see pricing battles this early in their development.
Intelligence is destined to be commoditized, and that's when software built on top of the models will begin to rise. Today's AI is mostly about chatbots and coding agents, but it won't stop there. A host of well-funded AI start-ups are rethinking what software means in the AI age, and in the end, they will be the biggest challenge to today's incumbents.
Write to Adam Levine at adam.levine@barrons.com
This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
July 17, 2026 11:43 ET (15:43 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments