[Software] Apocalypse Now? Separating Existential Risk from Investable Reality

Why Software Is at the Center of the AI Sell-Off

If you own software stocks, you may be wondering whether artificial intelligence (AI) is about to upend the software-as-a-service (SaaS) business model. You’ve likely seen headlines declaring “Di-SaaS-ter,” “SaaS-pocalypse,” “Software A(I)pocalypse” and “Software A(I)rmageddon.” The reality, however, is more nuanced.

The question is whether AI can write code and automate tasks, and, if so, whether software companies are still solid investments. This uncertainty helps explain the brutal sell-off in software stocks in late 2025 and 2026.

Investors focused on the capabilities of AI-native enterprise productivity tools, such as Anthropic’s November 2025 Claude Code update and January 2026 launch of Claude Cowork. These and similar tools crystallized the risk that AI systems could undermine or commoditize established software categories.

The challenge for software companies is threefold:

• First, the latest versions of AI models appear capable of replacing core applications.

• Even if AI tools don’t outright replace vital enterprise software solutions, AI models in the hands of fewer, more productive workers could slash the number of seat-based software licenses required.

• Third, companies may choose to build certain additional functions or modules instead of buying them. For example, rather than purchasing a plug-in for a content management system, a company might develop a custom solution. If this trend continues, it could negatively impact the sales and long-term growth of enterprise software providers.

Investors reacted to these new AI tools by triggering sharp drawdowns across companies, including ServiceNow, Salesforce, Intuit, HubSpot and Atlassian. The concern is that if AI can duplicate much of these platforms’ capabilities at a lower cost, long‑standing SaaS economics could be at risk.

Build vs. Buy in the Age of AI

Companies have traditionally been able to develop software internally rather than buy it. However, with the advent of AI, the cost-effectiveness of building versus buying is evolving.

We think it’s not just the code but also the data and expertise that will determine who wins and who loses in software. So, AI might simplify coding, but data differentiates systems, companies and providers. AI can’t replicate that uniqueness.

Think of it in terms of switching costs. If your software provider has a deep data moat and years of integration and experience with mission-critical business systems, the cost to switch is very high.

Let’s use our own industry as an example. In asset management, financial liability, audit, compliance, accuracy and high-stakes decisions all depend on these services and software systems. Is a company likely to risk all that by trying to “vibe code” its own solution?

Bloomberg vs. Perplexity: AI Disruption Meets Data and Switching Costs

Consider the financial data providers Bloomberg and FactSet. These companies have unique data, expertise and relationships that are hard to replicate. High-quality, real-time data is their competitive advantage. But so is the fact that they are parts of business-critical systems and processes that no corporate operations or tech person wants to touch for fear of breaking them.

In March, Perplexity introduced a financial research tool priced at $200 per month. The product is arguably aimed at Bloomberg, which charges $30,000 per terminal per year. Perplexity’s offering may well capture individual users who are unwilling or unable to pay Bloomberg’s price.

But for financial institutions worldwide, Bloomberg’s proprietary data and trading functions are deeply integrated into their systems. The reality is that their per-user seat license business peaked years ago. It’s Bloomberg’s position in the index and data businesses that keep it relevant and thriving.

AI as a Force Multiplier, Not a Universal Replacement

Some context also helps understand AI’s potential for disruption. Historically, technology has acted as a force multiplier for knowledge workers by improving productivity, enabling new workflows, and ultimately creating new job categories and businesses.

AI fits this pattern in many respects. An academic study on generative AI’s effect on the labor force from the 2022 public launch of ChatGPT to December 2025 found productivity and wage gains in occupations with significant AI exposure.¹ What’s more, the authors found little evidence that AI exposure significantly affected job openings or employment levels.

The point is that we should be skeptical of unquestioningly adopting the worst-case estimate of AI’s effects on jobs, companies or entire industries. In our view, what’s required is a clear-eyed analysis of the technology’s capabilities, opportunities and impacts for each use case.

Where AI Is Already Delivering Measurable Software Value

AI performs best when outcomes are deterministic, meaning identical inputs consistently yield the same results. Coding illustrates this well, as a computer program either functions correctly or it doesn’t. Anthropic has pointed out that Claude was developed using AI, highlighting the significant potential of these tools in software creation.

AI has also delivered measurable gains in digital advertising, with companies like Google and Meta reporting more than 20% improvements in return on advertising spend.²

Cybersecurity is another area where AI has already demonstrated its ability to add value. Companies such as Zscaler and CrowdStrike benefit from the growing adoption of “never trust, always verify” frameworks and AI-enhanced threat detection.

In cloud infrastructure, we believe monitoring and system visibility platforms like Datadog are well-positioned as AI increases system complexity and drives higher monitoring needs.

Electronic design automation leaders Cadence and Synopsys remain largely unaffected by traditional SaaS disruption. Their operations are closely aligned with the increasing complexity of semiconductor design, a field experiencing sustained growth driven by rising demand for AI chips.

But these successes don’t mean AI can seamlessly replace entire software platforms or business processes. For instance, call centers and customer service automation were once viewed as easy wins for AI. However, many users will go to great lengths to bypass bots to reach human operators.

What’s more, AI adoption is uneven, and trust matters, particularly in high-stakes environments involving financial liability, compliance and auditability.

AI Adoption Is Uneven: Identifying Resilient Software Companies

We believe caution is warranted when considering investments in software companies with limited competitive advantages, elevated valuations and slower AI adoption. This is especially true for firms whose products aren’t deeply integrated into customers’ workflows and can be readily replicated by competitors. Examples in this category include Workday, Atlassian and Salesforce.

In today’s landscape, we find the most compelling software businesses to be those with durable moats and established enterprise relationships — where AI enhances the product’s value rather than diminishes it.

It can also be misleading to compare current software valuations to historical levels, as past prices were driven by high growth that now looks less certain for many companies across the sector.

Why AI Sentiment May Exceed Software Reality

While AI is a powerful tool, it doesn’t replicate proprietary datasets, deep domain knowledge or years of established workflows. As a result, software differentiation will increasingly hinge on more than code alone. Data, expertise, integration and strong customer relationships are becoming critical advantages.

This underscores our guiding principle that fundamental, company-by-company research remains essential to investment success. We believe management quality, competitive advantage, pricing power and the ability to earn a return on investment will ultimately determine which software businesses compound value — and which don’t — in the age of AI.