For nearly a decade, Software as a Service (SaaS) companies attracted a market premium due to their asset-light business model. SaaS moats, or economic competitive advantages, not only derived from coding an advanced software, but also included factors such as recurring subscription revenue; sophisticated interfaces requiring training time; high switching costs; and embedded company data such as client histories, workflows, and pipeline projects. This premium was best captured in the iShares Expanded Tech-Software Sector ETF (IGV), Wall Street’s prominent basket tracking North American SaaS equities. Trading at consistently large price-to-earnings (PE) multiples, this ETF reached a Covid-era peak at 51x PE, nearly twice the S&P500 PE at the time.

However, the narrative shifted when the IGV experienced a 36% drop from its high in late October 2025 to a multi-year low in April 2026 – 24% at the time of publication, and valuations compressed to 22x PE in March 2026 – 35x PE at the time of publication. The sell-off was attributed not to worsening fundamentals, but rather to forward-looking repricing driven by the impact of large language models (LLMs). Financial publications argue that LLMs are undermining investor confidence in SaaS moat durability. LLMs can replicate future SaaS digital services faster and cheaper, and they can reduce per-seat licences as companies replace employees with AI agents, decreasing subscriptions needed. This uncertainty is impacting equity valuations even as current earnings remain intact. So what can SaaS entities do when their brand reputation has to contend with a disruptive market force?

The HALO litmus test 

The investor rotation out of SaaS was first explained by Josh Brown, CEO of Ritholtz Wealth Management, who coined the term ‘HALO’ (Heavy Assets, Low Obsolescence) to describe companies insulated from AI disruption. The thesis is if an LLM can replicate what a company makes or sells then the moat is weaker than expected. Brown compared SaaS with asset-heavy industries such as energy, materials, industrials, and consumer staples, and argued that physical infrastructure such as equipment and facilities cannot be digitised. The HALO narrative was subsequently widely adopted in investment newsletters, including CNBC, FT, WSJ, Barron’s, and J.P. Morgan.

In this context, J.P. Morgan Private Bank argued that investors are expecting AI agents to become the dominant interface by becoming a one-stop shop for workflows, lowering barriers to entry. For example, users can input simple prompts into Anthropic’s models to generate code and complete complex tasks from data aggregation and analysis to formatting and expense tracking, cutting time and costs. This automation affects recurring subscriptions of dedicated software tools for workflows which legacy SaaS companies have built their moats around.

The cybersecurity industry was one subsector of SaaS initially caught in the LLM crosshairs. CrowdStrike and Palo Alto Networks were oversold in late February 2026 when Anthropic debuted Claude Code Security in limited preview. This tool can scan codes for vulnerabilities and suggest patches for human review. Cybersecurity equities fell 10-25% in a few days, and industry-specific ETFs such as iShares Cybersecurity and Tech ETF (IHAK) and WisdomTree Cybersecurity Fund (WCBR) fell to their lowest levels since December 2023. In response, the Bank of America published guidance that Claude only posed a significant threat to exclusive code scanning platforms, not comprehensive cybersecurity vendors.

Decline in per-seat licences

Market observers assert that the AI productivity boom impacts the traditional pricing model for SaaS because the number of licences required will decrease as LLMs replace workers. This is also likely compelling in other industries where AI agents can complete tasks with little to no human interference, rendering per-seat-pricing misaligned or obsolete as clients require fewer humans to operate software, according to global consultancy firm Bain & Company (B&C).

This assessment was echoed by Jason Lemkin, former co-founder and CEO of Adobe, a multimedia cloud-based platform, who claimed that “seat compression” is a core fear for SaaS investors. He stated that legacy SaaS companies relied on seat-based pricing because it was a predictable source of revenue growth as clients increase headcounts. This model is now threatened by AI agents doing the work of multiple employees, leading to fewer licences or more price renegotiations. Similarly, Sanjay Poonen, CEO of data security firm Cohesity, said that some SaaS companies may need to lower seat-based pricing given LLM competition. This impacts revenue growth and terminal value, posing a risk for investors.

The power of narrative

Investors need to be selective with their SaaS investments as companies respond to AI disruption. Analysts argue that industry recovery is possible if earnings stabilise for the foreseeable future; if companies possess defensible moats that can capture the upside opportunity of AI, which primarily exist in the infrastructure layer; and if executives successfully adapt pricing models to deal with seat compression.

First, SaaS valuations will only stabilise when companies post positive earnings for several quarters to dispel uncertainty around their moats. Brown noted SaaS fundamentals will not save companies from lower multiples despite posting solid earnings in Q1 2026. This is because investors are focused on a three-year outlook instead of quarterly as they doubt whether companies can survive current margins and growth rates. For example, shares of ServiceNow, a cloud-based platform automating enterprise workflows, dropped 10% in late January, despite topping Wall Street’s Q4 2025 earnings expectations and issuing better than expected guidance. Morgan Stanley noted that the earnings were“good, but not good enough” because there was no narrative shift due to increased scepticism over future margins. The narrative has to be sustained to make the needed impact.

Second, some SaaS companies have more defensible moats which could be leveraged to capture LLM developments and deliver better outcomes for investors. LLMs cannot seamlessly replicate complex workflows which require deep domain expertise, access private client data that is resistant to third-party scraping, or match distribution and scale built over time. These aspects buy incumbents time to adapt, according to Rick Sherlund, the founder of AI investment advisory firm Sherlund Partners. However, Sherlund argued that these advantages are not insurmountable and will erode over time if SaaS does not quickly reposition.

Adapting and evolving

This is why selective positioning is key. Gabriela Borges, US software equity research analyst at Goldman Sachs, and Sherlund asserted that certain layers of SaaS are more vulnerable to LLMs, specifically applications – such as enterprise resource planning (ERP) or customer relationship management (CRM) – where end-users directly interact with products. They argued the infrastructure layer is less likely to be disrupted because it is more difficult to replicate physical computing, networking, and storage hardware of cybersecurity and cloud infrastructure firms, which aligns with Brown’s HALO thesis. For example, CrowdStrike and Palo Alto Networks collect threat data through physical hardware deployed where clients’ internal systems connect to external networks. Therefore, they control proprietary datasets of endpoint behaviour across millions of devices, which no LLM can replicate without deploying at scale.  Indeed, cybersecurity leaders such as CrowdStrike, Palo Alto Networks, Cloudflare, and Datadog are now proving the durability of their moats as they are back to trading at all-time highs at the time of publication.

Some SaaS companies are also already considering other pricing models. B&C has already identified a trend of more than 30 SaaS vendors introducing AI capabilities in hybrid pricing models, bundling per-seat with AI-based usage or feature access. Salesforce, a cloud-based CRM platform, and Adobe now use an AI meter on top of seat-based pricing. Borges argued that SaaS companies seeking to monetise AI could also opt for outcome-based pricing, charging for example per successfully completed customer service call, instead of per customer service representative. However, B&C noted that SaaS firms will need to offer clients flexibility, clarity, and time regarding their new models as they navigate which use case will return the most value.

SaaS companies can also potentially respond to LLM disruption by positioning themselves as AI beneficiaries or highlighting defensible moats. For example, cybersecurity firms would have benefited from amplifying favourable Bank of America guidance regarding their moat durability as multifaceted security providers, in contrast to exclusive vulnerability scanning platforms. Similarly, firms could seek to highlight positive outlooks regarding SaaS from reputable market research firms such as Fundstrat, whose co-founder Tom Lee argued in May 2026 that investors are underestimating how SaaS can adjust to and leverage the oncoming AI threat. When predictions become turbulent, firms cannot afford to miss out on opportunities to leverage their strengths.