The software barometer shed another 1.9% on Feb 4, down 17% over the past week and over 29% since it peaked last October. “We expect volatility to persist in the coming days as investors weigh the impact of potential business model disruptions from AI,” UBS analysts wrote in a Feb 4 note.
Last September, just weeks before the software sector index hit its peak, I wrote a piece titled “AI is eating software, and the world”. The headline was a playful dig at prominent Silicon Valley venture capitalist Marc Andreessen’s essay on software-as-a-Service (SaaS), dubbed “Why software is eating the world”, which argued that firms like Adobe, ServiceNow and Intuit Inc were helping businesses digitise workflows using cloud computing, digital signatures and data analytics to improve efficiency and customer service.
Anthropic’s Cowork agent
The software Armageddon was triggered by Anthropic’s release of new capabilities for its Claude Cowork assistant. It’s a tool that can autonomously build spreadsheets, browse the web, draft reports and organise files. AI start-ups are challenging incumbent enterprise software firms with agentic assistants that can perform tasks in an array of job functions with minimal instruction, operating with relative autonomy on a user’s computer.
Anthropic’s legal tool, for example, can review contracts, legal briefings and perform other industry-specific functions. Claude’s creator also released plug-ins for finance, customer service and other businesses. Little wonder, then, that corporate customers are asking: Why pay licensing fees for third-party tools when anyone can cheaply develop customised apps in-house with the help of AI?
See also: Thailand moves to seize US$420 mil assets linked to scammers
The irony is that until now, software businesses had seen AI as a tool they could use to enhance their own products. Every software firm was talking about “leveraging” AI to stay competitive and move to the next level. Now they regard Anthropic, OpenAI and peers as serious challengers. Anthropic doesn’t see itself as a mere upstream player that develops large language models (LLMs) for training, inference and reasoning. It offers a full-stack, agentic AI platform that spans from low-level infrastructure to specialised, ready-made enterprise applications. It builds its own models that can be customised for an industry’s specific needs.
Anthropic’s release sparked a sell-off in publishing firms that cater to law firms. London Stock Exchange Group stock fell 13%, while Thomson Reuters shares plunged as much as 21%. CS Disco, a litigation technology firm, sank as much as 14%, and LegalZoom.com declined 19%. The sell-off then spread to the broader software sector. Shares of Salesforce Inc, ServiceNow, Adobe and Workday Inc fell 7% while Intuit plunged 11%. Software players apart, investment banks such as Goldman Sachs, private credit giant Blue Owl Capital Inc and private equity (PE) firm TPG Inc, which have huge exposure to the software sector, were also hit.
Over the past decade, over 1,900 software firms were bought out by PE players in transactions valued at $440 billion. PE firms paid top dollar to buy software firms. For years, they have been eagerly awaiting the reopening of the IPO window. Now, as IPOs are ready to hit the market, software valuations have collapsed.
See also: Samsung claims lead in race to ship AI memory chips to Nvidia
Here is why this past week’s sharp plunge in software matters: The software sector makes up just 7.5% of the S&P 500 — down from nearly 14% at the peak. Software developers went from selling shrink-wrapped software on CDs to selling subscriptions and bundling the software with services like delivery, upgrading and maintenance over the internet. Instead of a traditional one-time sale, SaaS was built on a recurring-revenue model, with corporate customers paying a recurring fee for each individual user in the company, or “seat”, with access to a software product. For 15 years, SaaS valuations relied on a simple assumption: as customers hired more employees, they purchased more software seats. Revenue scaled with headcount.
AI improvements have accelerated the decline in knowledge-worker roles and total headcount, thereby putting pressure on seats, which in turn reduced revenue for software firms. Now, new AI tools are threatening to supplant many of the better-known enterprise software firms. We are now moving towards a world where outcomes scale without adding humans. If an AI agent can perform the work of an entire marketing team or half the human resources team, the software users are no longer focused on the number of licences. The focus has since moved to the results delivered.
Over 110 enterprise software stocks in the IGV soared a whopping 1,030%, or more than 11-fold, between October 2010 and October 2025. During that period, Microsoft’s shares surged 2,000%, Salesforce was up over 1,100%, and Oracle shares rose 780%. ServiceNow, which was only listed on Nasdaq in mid-2012, was up nearly 2,900%. A lot of investors made a ton of money betting on software stocks over the last 15 years.
Here’s what differentiates AI’s agentic model from the SaaS model that’s now fading. AI is mostly built on LLMs that can perform tasks independently by using tools, designing plans and adapting to dynamic environments to achieve specific goals as well as multi-agentic systems. Under the SaaS model, software creates value by organising human labour through tools like customer relationship management (CRM) and project management, while under the AI agentic model, software creates value by executing the work itself. As corporate users evaluate return on investment based on outputs rather than seats, per-user pricing of the SaaS model became increasingly harder to justify. So, investors who don’t yet see how AI will generate returns across its own value chain are forced to aggressively sell shares of Salesforce, Adobe, Twilio Inc and others. Revenue that once appeared predictable now carries more long-term uncertainty.
Human staffers have, until recently, acted as the connective tissue between software applications that corporates use. They checked Salesforce for pipeline data, opened NetSuite to review budgets and sent emails to move work forward. Now AI agents are trying to do all that and more. They can pull data from one system, update another and trigger actions across multiple tools without human intervention. That creates a challenge for traditional systems of record. Salesforce, ServiceNow and Workday provide a suite of software that corporate customers use, but Salesforce tools cannot easily orchestrate workflows across HR and finance. To better compete with AI agents, rival software firms threatened by Anthropic and OpenAI would either need to merge with each other or collaborate in ways they have not done so far.
Shift in business models
AI is accelerating a shift in business models in software and IT services. Analysts say companies will use new tools to address their technology needs, including those from OpenAI’s Codex, Anthropic’s Claude Cowork or vibe coding providers such as Lovable, Replit, Bolt and Cursor AI that use AI to write computer code, allowing the developer to focus on the “vibe” or higher-level goals. Lower barriers to entry in coding and creating software will likely generate more competition, which, in turn, leads to lower prices that the software firms can charge.
As value shifts upwards, underlying SaaS products risk becoming commoditised storage layers. Last year, Microsoft CEO Satya Nadella warned that legacy software could be reduced to basic CRUD (Create, Read, Update, Delete) databases that store information while agents take over the more complex, intelligent layers, capturing user attention and economic value. He argued that the era of relying on standalone, static applications for business processes is ending, to be replaced by AI-driven systems. Nadella also believes the “age of apps is over”. He predicts that AI agents will become the primary orchestrators of workflow, making the traditional user interface-heavy apps obsolete. Instead of users jumping between different tools such as Salesforce, Slack, ERPs or enterprise resource planning to perform CRUD tasks, intelligent agents will seamlessly handle these actions across various systems.
Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends
Software firms must be mindful that AI is a real threat to many of them, and unless they have scale, they will be disrupted sooner or later. The most vulnerable are those whose applications provide an interface on top of a database. When an agent can query the data directly, the interface loses its value because Claude Code makes it easier to meet requirements for small tasks. Also vulnerable are smaller and medium-sized firms and younger start-ups without network effects or a large user base. Larger, complex platforms with substantial codebases and deep workflows — such as Microsoft, Oracle, Salesforce, ServiceNow and Intuit, which focuses on accounting and tax software — are insulated. Also safe are Palantir, Snowflake Inc and privately owned Databricks, which is reportedly seeking an IPO later this year. Others in the same category are security-based software firms like CrowdStrike Holdings Inc and Palo Alto Networks Inc.
For now, though, the market may be overlooking the baby that’s being thrown out with the bathwater. In the aftermath of OpenAI’s unveiling of its chatbot ChatGPT in late 2022, Microsoft became a close partner and the principal funder of the world’s fastest-growing AI start-up, pouring US$13 billion into the firm, which at the time was a non-profit AI lab. That catapulted its shares to US$555 ($700) last October. Software giants’ shares plunged to under US$408 on Feb 3, or down 30% from 52-week highs. Oracle, whose stock touched an all-time high of US$345 last September when it unveiled its vision to become a hyperscaler alongside big AI LLM builders Microsoft, Amazon and Google, saw its stock plunge over 61% to US$144 this past week.
ServiceNow’s shares peaked at US$234 in January last year. It is now down 55% from those highs. Salesforce stock peaked in December 2024 at around US$365. This past week, they traded at around US$187, or down 49%. Adobe’s stock peaked in August 2021 and is now hovering around 61% below its all-time high. Intuit’s shares are 44% from its peak last June. Atlassian Corp, whose software is used in project management and team collaboration, is down 32% since the start of the year and 79% from its 2021 peak.
The widespread sell-off in SaaS stocks, says Derek Hernandez, an analyst for VC-focused Pitchbook, “is a fundamental repricing of software, not its death.” SaaS, he notes, was once “perceived as cash-flowing with durable moats, without much question about its sustainability. Those moats now appear vulnerable to AI disruption.”
Still, it takes time for companies to rip up their existing systems. Microsoft, Oracle, ServiceNow and peers have a “sticky” advantage in following corporate governance procedures, compliance protocols and dealing with procurement systems. True, security and compliance matter more than just cost, and adoption of a radically new approach takes time, but corporates like Goldman Sachs are working with Anthropic on an in-house AI agent, which is already successful in handling complex workflows like accounting and compliance.
Beaten-down software is ripe for a reboot, says Jefferies analyst Brent Thill. Catalysts include clarity on “coopetition” with AI model providers, a meaningful increase in enterprise AI deployments, acceleration in revenue growth due to AI monetisation, and finally, moderating capex as accruing value shifts from infrastructure to software, he adds.
Assif Shameen is a technology and business writer based in North America
