AI vs SaaS: Sell first, ask later, the market only "got it right half the time"?

The new products from Anthropic have triggered a nearly trillion-dollar sell-off in enterprise software stocks, exposing the market's excessive panic over AI threats.

Barclays pointed out that investors have overlooked a key technological distinction: AI tools are indeed encroaching on the application layer business of SaaS companies, but they have yet to shake the underlying "system of record" infrastructure—which is precisely the core moat of enterprises like Salesforce and SAP.

The products released by Anthropic last week, such as Claude Cowork, became the last straw that broke the camel's back for enterprise software stocks. Customer relationship management software stock Salesforce and financial management software stock Workday have cumulatively fallen over 40% in the past 12 months.

Behind this panic sell-off is investors' blurred understanding of the boundaries of AI capabilities. The market generally believes that next-generation AI tools like Anthropic and OpenAI will completely replace traditional SaaS software, leading to the value of established companies dropping to zero.

According to a report released by Barclays on February 10 titled "Software Is Not Dead, Just Changing," this "one-size-fits-all" simplistic logic does not apply to most enterprise software companies.

What AI Can and Cannot Do

The essential advantage of generative AI lies in pattern recognition and "draft generation," but its probabilistic nature also constitutes a fundamental limitation. AI excels at tasks that require extracting patterns from vast amounts of data, such as processing natural language and coding, but it struggles in scenarios that require absolute accuracy.

According to the Barclays report, traditional software operates based on deterministic rules, where the same input will inevitably produce the same output. In contrast, AI software is fundamentally probabilistic, operating by learning behaviors rather than hard-coded logic, and cannot guarantee consistency in every output.

This means that AI works at a higher level of abstraction and is not a replacement for traditional software.

This technical characteristic determines the applicable boundaries of AI. In fault-tolerant scenarios such as knowledge work and content generation, AI can replace or even surpass traditional SaaS applications; but in areas requiring a "single correct answer," such as billing processing, compliance auditing, and business rule execution, AI is still not up to the task.

Independent analyst Benedict Evans pointed out that successful SaaS products stem from mapping unique organizational problems into workflows and then encoding them into software. These long-accumulated customized business rules form the infrastructure of banks, hospitals, retailers, and are also the foundation of companies like Epic Systems and Oracle

Half Misjudged, "System Records" Layer Hard to Replace

Barclays report clearly states that three types of enterprise software companies have been mispriced in the sell-off and are worth re-evaluating by investors.

First are system record companies. For example, Salesforce, as a customer relationship management system, holds the "single truth" about a company's customers and revenue—key data such as transaction progress, discount approvals, sales commissions, and revenue forecasts all require definitive answers.

SAP's position is even more solid. As an enterprise financial system record, SAP CEO Christian Klein emphasized during the January earnings call that advanced generative AI models cannot handle the critical business data and workflows that companies rely on for survival.

Barclays believes SAP is stickier than Salesforce because the irreplaceability of financial truths is stronger. Workday holds a similar position in the human resources and payroll sector.

AI will not only fail to replace these systems but will actually increase their importance. AI agents will create more data touchpoints, and the complexity that system records need to handle will rise accordingly. Barclays' research points out, "This means the importance of these systems increases rather than their value going to zero, contrary to market views."

Data Tools and AI Computing Sector Also Misjudged by the Market

In addition to system record companies, the Barclays report identifies two other types of investment opportunities misjudged by the market.

The second category is beneficiaries of AI agents. AI will bring more demand for code and data, tools like JFrog (FROG) that manage software versions and security, as well as data vendors like Snowflake (SNOW) and MongoDB (MDB), may see increased usage due to AI expansion.

The third category is AI computing providers. Here lies the biggest logical contradiction in the market. If AI is powerful enough to disrupt the entire software industry, the demand for computing power should logically surge, yet companies like Oracle and CoreWeave have suffered severe blows in the sell-off. "There must be an issue here that warrants deeper investigation; market sentiment is overly pessimistic," wrote Barclays analysts.

Half Sold Right, Application Layer Profits Squeezed

Market panic is not without reason. SaaS companies built on database infrastructure have long performed poorly: interfaces are clunky, not intuitive enough, prices are inflated, and there are sometimes security vulnerabilities. Worse still, customers are often locked into inferior systems due to high migration costs.

Matt Stoller, research director of the American Economic Freedom Project, wrote: "The American software industry model is shaped around monopolization, providing low quality and poor security at high prices." He described a meeting in 2016 with community bankers, who criticized their niche software vendors as "expensive" and "terrible."

Swedish fintech company Klarna will stop using Salesforce and Workday software in 2024, opting instead for products from smaller SaaS companies like Deel and Neo4j, and building more modern application layers on top using AI coding tool Cursor This reveals the true threat path of AI to SaaS: customers are not simply replacing SaaS software with AI tools, but are building their own applications with AI, squeezing out expensive interface layers while retaining underlying data.

The repricing of the software sector will continue

This market correction is necessary for the enterprise software application layer. SaaS companies have long enjoyed high valuation multiples because they control both the infrastructure and the interface. If the technologies from Anthropic and OpenAI can overlay on system records, it will begin to erode the pricing power of SaaS companies.

Barclays' research report summarizes: "This means that the era of easily obtaining high profits for bloated enterprise software application layers may be coming to an end." However, this does not equate to the end of the entire industry. The key is to distinguish which companies rely on application layer profits and which companies' value is rooted in irreplaceable system record layers.

SAP's statements during the January earnings call represent the confidence of system record vendors. Other SaaS executives are also pushing back against bearish views. But the market needs time to digest these technical details and distinguish between genuine disruptive threats and exaggerated panic.

The indiscriminate nature of the current sell-off indicates that investors, such as those in the credit market who previously had limited understanding of the software industry, are making decisions based on the most extreme viewpoints.

As the understanding of AI's capabilities and the business models of SaaS companies deepens, the market may begin to reprice those companies that have been incorrectly categorized as "AI victims." However, for those companies that have long relied on charging high fees for inferior application layers, the valuation squeeze may just be beginning.

Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account individual users' specific investment objectives, financial conditions, or needs. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk