The SaaSpocalypse: is Software dead?

What happened

On 3 February 2026, Anthropic released specialized plug-ins for its Claude Cowork agent targeting legal, compliance, and professional services workflows. The announcement triggered what traders at Jefferies dubbed the "SaaSpocalypse."

On an intraday basis, Thomson Reuters Corp plunged -21%, Relx PLC -15%, Wolters Kluwer -13%, and LegalZoom cratered -20%. But the contagion spread far beyond legal software: the XSW index has recorded six straight sessions of losses, erasing approximately $830 billion in market capitalization since 28 January. The sell-off went global: India's Nifty IT fell 6.3%, China's CSI Software dropped 3%, Hong Kong's Kingdee lost 13%. 

The fear can be summarized in a single question posed by John Zito of Apollo Global Management last fall: "Is software dead?"

Impact on our Investment Case

Consistency of our thesis — and its limits

We have argued consistently since mid-2024 that value in AI investing would migrate from infrastructure to applications, and specifically toward vertical, industry-specific software. The event that triggered this sell-off, namely Anthropic releasing a vertical AI tool for legal workflows, is actually a confirmation of this directional argument. AI's value creation is indeed happening at the application layer.

The companies building and deploying these AI solutions are themselves software companies. Their products run on cloud infrastructure, consume data services, require monitoring tools, and need cybersecurity protection; all capabilities provided by the names in our portfolio.

Our thesis has evolved from "software beats hardware" to "vertical software beats horizontal" to "enablement software is essential infrastructure". If the enablement layer itself comes under pressure (e.g., through vertical integration by AI platforms, open-source commoditization of tooling, or customer base contraction), we will need to adapt again. We prefer to state this explicitly rather than present our current positioning as the final answer.

The Software industry is not a monolith

The sell-off has been largely indiscriminate. Yet AI's impact on software companies varies enormously depending on where they sit in the value chain. We see three distinct categories:

Software genuinely at risk.

Pure-play SaaS vendors whose core value is simple workflow automation (e.g., contract review, document management, basic data entry) face legitimate disruption. When an AI agent can perform the same task without a dedicated application, the per-seat licensing model comes under existential pressure. Thomson Reuters, LegalZoom, and CS Disco were hit hardest for good reason.

Software that enables AI deployment.

Companies like Datadog (observability), Snowflake (data infrastructure), and Mongodb (database) provide the plumbing that every AI agent and agentic workflow requires. Beyond that, a significant portion of their revenue comes from existing enterprise clients who have not yet begun their AI transition, and when they do, structural shifts of this magnitude require substantial capex and time. This provides an additional layer of resilience that the market is currently ignoring. We believe the conflation with genuinely vulnerable names is excessive, but we must also be candid about the risks this category faces (see below).

Software enhanced by AI.

Cybersecurity companies like Palo Alto Networks, and CrowdStrike are integrating AI to strengthen their competitive moats. AI-enhanced attacks increase the threat surface (driving demand), while AI-powered defense creates premium products. These names have been caught in the broader sector de-grossing, but in our view their fundamental exposure to AI remains more as a tailwind than a threat.

Risks we are monitoring

There are genuine risks to the enablement layer that we are actively evaluating:

Vertical integration by AI platforms.

Anthropic's acquisition of Bun (which we analyzed in December) demonstrates that AI companies are willing to internalize capabilities rather than rely on third-party tools. If this pattern extends to observability, data infrastructure, or security (i.e., if AI platforms build rather than buy these capabilities), the enablement layer's addressable market narrows. We do not see evidence of this happening at scale today, but the precedent is established.

Customer base transition risk.

Our enablement-layer holdings currently derive the bulk of their revenue from existing enterprise software companies, not from AI platforms. If legacy customers contract faster than AI-native customers ramp, there is a revenue trough even if the long-term thesis is correct. Being right about the destination but wrong about the journey is financially painful for any investor with a finite horizon.

Concentration of buying power.

If the software ecosystem consolidates from thousands of SaaS companies into a handful of dominant AI platforms, the enablement layer faces fewer, larger customers with substantially greater pricing leverage. This is structurally different from today's fragmented customer base.

On the "paradox" in the bear narrative

Bank of America's Vivek Arya identified a tension in the current market narrative: investors appear to be simultaneously pricing in weak AI infrastructure ROI and AI powerful enough to make software obsolete. If AI is truly that capable, the infrastructure supporting it should remain in demand. This is an intellectually appealing argument, but the reality is more nuanced than a clean paradox. The bear case is not that AI fails and succeeds simultaneously. It is that the value capture is shifting away from both legacy SaaS and the current generation of infrastructure tools, and toward the model providers themselves (Anthropic, OpenAI, Alphabet's DeepMind) who are vertically integrating. In this scenario, AI succeeds spectacularly, but neither traditional software vendors nor their current enablers capture the economics. We do not fully subscribe to this view, but we take it seriously.

Valuations: lower, but not yet compelling

While software multiples have compressed from their peaks, they have not reached levels that historically signal attractive entry points. The AI-related software segment (where much of our exposure is concentrated) still trades at forward P/E ratios in the mid-to-high 20s. This is below pandemic-era excesses, but it is not distressed territory, and there remains meaningful downside before valuations become truly compelling on an absolute basis.

The market has reached oversold levels in just a couple of weeks. But oversold does not mean cheap. For aggressive positioning, we need either significantly lower valuations or clear evidence that the market is beginning to differentiate between software categories. We see neither yet.

Our Takeaway

The "SaaSpocalypse" reflects a genuine risk for legacy SaaS companies whose value proposition can be replicated by AI agents. On that front, the market's repricing has logic behind it. Where we see a potential disconnect is in the indiscriminate nature of the selling: the failure to distinguish between software that AI replaces and software that AI requires. Our portfolio is positioned in the latter category. But we also acknowledge, following our own analysis of vertical integration trends, that this "safe" category carries risks we actively monitor.

We do not claim superior timing. The market is always right in the moment, and ebb and flows do not always align with fundamentals. What we do know is that such mismatches, when they occur, have historically created opportunities for patient investors with a medium-to-long-term horizon.

We are maintaining our positioning, monitoring developments closely, particularly the risks outlined above, and remain ready to adjust as visibility improves.