As often happens in markets, however, a reasonable question rapidly became an extreme conclusion. History is littered with industries that were supposedly facing extinction, only to emerge larger, albeit substantially changed. The PC was expected to eliminate mainframes. The internet was supposed to kill physical retail. Digital distribution was supposed to destroy traditional media. Cloud computing was subsequently expected to marginalize incumbents like Microsoft, Oracle and SAP. In virtually every case, disruption was real, individual companies disappeared, and enormous value migrated toward new winners. But the ultimate market became larger rather than smaller. That distinction between an industry dying and an industry changing seems particularly relevant today.
There were (and are) several reasons to question the software-is-dead thesis. The most obvious is almost tautological: AI is software. Making software dramatically cheaper to create should increase the quantity of software produced, just as falling computing and storage costs expanded rather than contracted the software industry during previous technology cycles. The global software market has grown more than fivefold over the past two decades as cloud infrastructure, open-source tools and better development frameworks progressively reduced the cost of building applications. AI represents another potentially enormous step down that cost curve.
More importantly, writing code isn’t the same thing as building enterprise software. It is estimated that writing code represents only 15% to 20% of the work required to build an enterprise system. Designing workflows, structuring data, controlling access, maintaining security and integrating systems represent much of the remaining complexity. Corporate customers also care about reliability, accountability, regulatory compliance, and having someone responsible when a mission-critical system fails. AI can dramatically reduce development costs without eliminating these requirements.
Incumbents possess additional advantages that the market arguably discounted too aggressively. Systems of record contain years or decades of proprietary data and organizational context. Vertical software is embedded deeply in specialized workflows. Cybersecurity businesses collect data through networks and endpoint agents that cannot simply be recreated by asking a language model to write code. Scale, customer relationships, distribution and domain expertise remain valuable even when code becomes cheaper.
None of this means the bears were entirely wrong. Some narrow applications can undoubtedly be replicated or absorbed into broader AI platforms. Seat-based pricing may come under pressure as agents replace portions of human workflows, incumbents unable or unwilling to redesign their products will lose share, and AI-native competitors will create substantial new value. This is creative destruction, not immunity from destruction.
This was the investment decision we were faced with earlier this year – which companies have defensible moats, and which companies are most at risk of disruption? We culled our portfolios and trimmed the weaker players, while holding on to the software stocks we felt would endure and were being unduly punished by the market.
As we hoped, the numbers have begun to contradict the narrative. Even during the first quarter selloff, fourth-quarter 2025 software earnings were growing at a double-digit rate, industry margins were at record highs, and near-term earnings estimates were being revised upward. More recent results have strengthened that case. IGV (a software ETF) has gained roughly 39% since the April lows. Almost every S&P 500 software company reporting in the latest earnings season beat profit estimates, with the average surprise exceeding 13%, while more than 80% beat revenue expectations. Microsoft rallied sharply following accelerating cloud growth, Salesforce jumped more than 20% following its results and forecast, and ServiceNow has risen roughly 45% since reporting in late July. More importantly, many companies are not just holding the line, but are showing meaningful traction on integrating and monetizing AI into their software and data both internally and externally.
Perhaps that is the real lesson of 2026. Markets are very good at identifying disruption but considerably less reliable at estimating its timing, magnitude, and ultimate distribution of winners and losers. The investment question, therefore, may no longer be whether software is dead. It is which companies can use AI to deepen their competitive advantages, expand their markets and improve their economics, and which ones become casualties of the transition. After indiscriminate selling earlier this year, investors increasingly appear willing to make that distinction again, and we are excited to continue on this roller coaster ride.
The Not Quite Death of Software
by Jonathan McAdams, CFA®
Monty Python and The Holy Grail is considered by many to be one of the top big-screen comedies of all time. One of my favorite scenes in the movie is when, during the plague of the Middle Ages, a man can be seen pushing a cart through the streets yelling “Bring out yer dead!” A villager approaches carrying an old man over his shoulder and throws him on the cart. The old man pops his head up and weakly protests “I’m not dead!” and “I’m getting better,” as the other two men argue about the prognosis of the obviously “not dead” man.
One of the more remarkable investment stories of 2026 has been the roller coaster trip in the oft-proclaimed “dead” software stocks. Earlier this year, investors appeared to reach a remarkably simple conclusion about a remarkably complicated technological transition: artificial intelligence was going to kill software. The catalyst was real. New agentic products from Anthropic and OpenAI launched in January 2026 demonstrated that AI was moving beyond chatbots and into enterprise workflows previously dominated by traditional software. If AI could write code, analyze spreadsheets, conduct legal research and operate other applications autonomously, investors wondered whether corporations would continue paying premium prices for hundreds of individual software applications.
The resulting carnage in the market was substantial. By early April, the iShares Tech-Software Sector ETF (IGV) had fallen roughly 30% year-to-date and 36% from its late-2025 high. Valuations compressed from roughly 35 times forward earnings late last year to 18 times, the lowest level since 2014. The drawdown was the largest non-recessionary 12-month decline in software equities in more than 30 years, with roughly $2 trillion erased from the sector’s peak market value.
As often happens in markets, however, a reasonable question rapidly became an extreme conclusion. History is littered with industries that were supposedly facing extinction, only to emerge larger, albeit substantially changed. The PC was expected to eliminate mainframes. The internet was supposed to kill physical retail. Digital distribution was supposed to destroy traditional media. Cloud computing was subsequently expected to marginalize incumbents like Microsoft, Oracle and SAP. In virtually every case, disruption was real, individual companies disappeared, and enormous value migrated toward new winners. But the ultimate market became larger rather than smaller. That distinction between an industry dying and an industry changing seems particularly relevant today.
There were (and are) several reasons to question the software-is-dead thesis. The most obvious is almost tautological: AI is software. Making software dramatically cheaper to create should increase the quantity of software produced, just as falling computing and storage costs expanded rather than contracted the software industry during previous technology cycles. The global software market has grown more than fivefold over the past two decades as cloud infrastructure, open-source tools and better development frameworks progressively reduced the cost of building applications. AI represents another potentially enormous step down that cost curve.
More importantly, writing code isn’t the same thing as building enterprise software. It is estimated that writing code represents only 15% to 20% of the work required to build an enterprise system. Designing workflows, structuring data, controlling access, maintaining security and integrating systems represent much of the remaining complexity. Corporate customers also care about reliability, accountability, regulatory compliance, and having someone responsible when a mission-critical system fails. AI can dramatically reduce development costs without eliminating these requirements.
Incumbents possess additional advantages that the market arguably discounted too aggressively. Systems of record contain years or decades of proprietary data and organizational context. Vertical software is embedded deeply in specialized workflows. Cybersecurity businesses collect data through networks and endpoint agents that cannot simply be recreated by asking a language model to write code. Scale, customer relationships, distribution and domain expertise remain valuable even when code becomes cheaper.
None of this means the bears were entirely wrong. Some narrow applications can undoubtedly be replicated or absorbed into broader AI platforms. Seat-based pricing may come under pressure as agents replace portions of human workflows, incumbents unable or unwilling to redesign their products will lose share, and AI-native competitors will create substantial new value. This is creative destruction, not immunity from destruction.
This was the investment decision we were faced with earlier this year – which companies have defensible moats, and which companies are most at risk of disruption? We culled our portfolios and trimmed the weaker players, while holding on to the software stocks we felt would endure and were being unduly punished by the market.
As we hoped, the numbers have begun to contradict the narrative. Even during the first quarter selloff, fourth-quarter 2025 software earnings were growing at a double-digit rate, industry margins were at record highs, and near-term earnings estimates were being revised upward. More recent results have strengthened that case. IGV (a software ETF) has gained roughly 39% since the April lows. Almost every S&P 500 software company reporting in the latest earnings season beat profit estimates, with the average surprise exceeding 13%, while more than 80% beat revenue expectations. Microsoft rallied sharply following accelerating cloud growth, Salesforce jumped more than 20% following its results and forecast, and ServiceNow has risen roughly 45% since reporting in late July. More importantly, many companies are not just holding the line, but are showing meaningful traction on integrating and monetizing AI into their software and data both internally and externally.
Perhaps that is the real lesson of 2026. Markets are very good at identifying disruption but considerably less reliable at estimating its timing, magnitude, and ultimate distribution of winners and losers. The investment question, therefore, may no longer be whether software is dead. It is which companies can use AI to deepen their competitive advantages, expand their markets and improve their economics, and which ones become casualties of the transition. After indiscriminate selling earlier this year, investors increasingly appear willing to make that distinction again, and we are excited to continue on this roller coaster ride.
Be Mindful of your “Surprise Raise”
What Higher Bond Yields Mean for Investors
Employee Spotlight: Blake Crosby
Proactive Planning: Laying the Foundation for the Next Generation
Featured Recipe: Fire Roasted Three-Meat Game Day Chili
Crude Oil and Chokepoints: The Growing Global Impact