AI is sucking in all the oxygen in the room, and investors who once rode on the software bandwagon are now abandoning it for faster horses that have AI pixie dust sprinkled on them. Not only is it taking attention and capital away from traditional software players, but AI is also making many of them obsolete. The global software market topped US$736 billion last year.
Median revenue growth for listed software firms has decelerated from 29% per year in 2021 to an expected 13% this year and 11% next year. The rise and rise of AI is increasingly being seen as an existential crisis for the software industry. There is a growing concern that incumbent software firms will eventually be overtaken by AI players or large language models (LLMs) that can create apps.
Look no further than the plummeting software stocks for evidence that the curtain is falling on the once-beloved sector. The benchmark iShares Expanded Tech-Software ETF (IGV) has been trailing the broader market barometer S&P 500 Index for over a year. This year, the S&P 500 is up about 10%, the Nasdaq 100 index has risen 12% while the IGV is up just 7.2%. If you take out the three largest components of the software index — Microsoft, Palantir Technologies and Oracle Corp, all of whom are now seen as AI icons rather than software firms — the rest of the software sector index would be down 14.5% year to date compared with the broader market, which is up 10%. “The growing fear that AI will fundamentally change how enterprise software is traditionally built and used has weighed heavily on investor sentiment,” notes Brent Thill, software analyst for investment bank Jefferies in San Francisco.
AI is still in an early buildout phase. Top players building LLMs will spend US$400 billion ($515 billion) next year, up from US$364 billion this year. Not surprising, then, most of the value has accrued to pick-and-shovel players like Nvidia, which supply graphics processing units or AI chips to software giant Microsoft, search supremo Google, e-commerce pioneer Amazon, Elon Musk’s xAI, Oracle and social media behemoth Meta Platforms. Although the next AI growth phase is likely to be AI application software, we are still several years away from that point.
See also: Singapore Academy of Law brings AI to 35-year-old LawNet platform
Let me first explain how the software business got to where it is today, so you can understand how AI is killing traditional software. Though the word “software” was first used by American mathematician and statistician John Wilder Tukey in 1958 to distinguish computer programs from hardware, software-like stored programs were first written by Tom Kilburn 10 years earlier, in 1948, for a small-scale experimental machine called Manchester Baby.
Two decades ago, people or companies used to buy shrink-wrapped physical packages of software on compact discs or CDs, or digital video discs or DVDs. Users would break the packaging seal and install the software on a single desktop computer. With ubiquitous high-speed internet access, anyone can download software over the internet from a cloud server.
Software firms these days essentially sell a subscription, bundling the software with ongoing services like delivery, upgrading and maintenance over the internet. Instead of selling a new version of the software once a year, these days, software firms just update the software every few weeks or months over the internet. The business model of enterprise software or software sold mostly to companies, or SaaS, is different from the old “shrink-wrap” model because software firms now offer continuous service rather than a one-time sale.
See also: Mainframes stage an AI comeback
Corporate customers pay a recurring fee for each user in the company, or “seat”, who has access to a SaaS product. You can pay a monthly, quarterly or annual fee for each user who needs access to the software. The total cost for a company increases with the number of users, providing a simple, predictable and scalable way for both vendors to ensure revenue and customers to manage costs for a growing workforce.
The SaaS model became much more entrenched as cloud-based applications began to roll out, which allowed easy scaling of user licences as businesses grew. Over the years, the model has evolved. Growing companies can opt for either a seat-based model, or how many of its employees are using the platform, or a usage-based model or how much a company’s staffers use the software firm’s product. AI is seen as a key driver of productivity enhancement, which means fewer humans will be working alongside AI agents, which in turn makes it a threat to SaaS’s seat-based model.
This is not something far out in the future. Buy now, pay later firm Klarna has reported that it is employing fewer people because of its widespread use of generative AI, or Gen AI. Several large users of the Adobe software have reported that they now have far fewer people using it as they roll out Gen AI.
Vibe coding fears
Here is how AI disruption of software is unfolding: AI advancements are accelerating the pace of decline in knowledge workers or total headcount, putting pressure on seats, which in turn reduces revenue for software firms. Moreover, software companies that have traditionally relied on price hikes, particularly for growing companies, will more likely see increasing competition from AI start-ups as it becomes easier to build custom software through vibe coding, a software development approach in which a human guides an LLM to generate, test and refine code through conversational, natural language prompts rather than writing code line by line.
Vibe coding helps software developers focus on high-level goals and iterative feedback, while the AI handles much of the code’s creation and debugging, making software development faster and more accessible. But more than that, vibe coding also enables non-technical users to create basic applications. Vibe coding makes you and me, and everybody else, a software engineer. Not everybody thinks that’s necessarily bad for the software companies. “Vibe coding fears are overdone,” says Jefferies’ Thill.
The fears may be overdone, but vibe coding is starting to hurt outsourcing players. Citigroup analysts in a Sept 2 report argued that Gen AI was starting to impact services offshoring in India and the Philippines, given their exposure to lower value-added skills such as code development, customer service and knowledge summarisation. The investment bank cited a Stanford University study that showed GenAI is already materially impacting the employment of early-career American workers in AI-exposed sectors like software development and customer service. Citigroup also cited a Massachusetts Institute of Technology study that AI gains are being found in back-office automation, replacing BPO or business process outsourcing, where companies hire external service providers to perform non-core business functions or processes such as customer service, accounting or IT to achieve cost savings and gain efficiencies so they can focus on their core competencies.
Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends
The bear case for software goes something like this: Over time, software as we know it will vanish and be replaced by AI agents or autonomous software programs 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. These systems are a network of individual AI agents that work collaboratively, communicate and coordinate to solve complex problems or perform large-scale tasks that would be beyond the capabilities of a single agent.
Rishi Jaluria, software analyst for RBC Capital, notes that the definition of software has evolved, “from mainframe systems to packaged CD-ROMs to downloadable software and now SaaS”. He believes the ability to create software from natural language through advanced AI models like OpenAI’s latest GPT-5 will eventually benefit the software industry, as lowering barriers to entry helps create more innovation.
While AI is driving real change, Thill argues that “the pace of disruption will be evolutionary, not existential”. He noted in a recent report titled AI Is Not Eating Software that enterprise software is deeply layered with meaningful complexity spanning across security, governance, compliance, integrations and custom workflows that AI cannot yet replicate end-to-end. The intricacies of enterprise architecture make full AI disintermediation of software unlikely”. Thill argues that infrastructure and security software players will still do well in a world dominated by AI, but horizontal apps could struggle. RBC’s Jaluria doesn’t believe that the large incumbents will automatically be the beneficiaries given data and distribution, but argues that companies that innovate around AI and build AI-native solutions, even if they cannibalise their existing businesses, can be AI beneficiaries over the long term.
Software M&A deals
As stock prices of traditional software companies are hit, investment bankers are pounding on their doors with deal proposals. Because there is little appetite for buying other beleaguered peers, a lot of the deals are software firms buying into AI firms, or more specifically, AI agents and application start-ups. American software companies have made US$33.8 billion in AI acquisitions so far this year. Earlier this year, automated business workflows software firm ServiceNow bought Moveworks, which sells AI assistants, for US$2.85 billion.
Two months ago, NiCE, which sells software for call routing, bought conversational AI chatbot maker Cognigy for US$955 million. Salesforce bought sales lead generator Bluebirds for US$1.1 billion and supply chain automation provider Regrello, while Workday paid US$1 billion for recruiting assistant Paradox in August. Privately held cloud-based platform for data analytics, Databricks, which last month raised over US$1 billion from VCs at US$100 billion valuation, bought Neon for US$1 billion and machine-learning application start-up Tecton for US$900 million earlier this year. “Software M&A is on pace for its second-best year in 2025” with total software sector merger and acquisition activity for the year possibly reaching over US$150 billion, investment bank Jefferies noted in a recent report.
So, what’s next? We may not yet be witnessing the end of software as we know it, but it is clear that software’s SaaS model will need a lot of tweaking if it is to remain relevant in the new AI environment. Ultimately, “AI is a transformational wave, not a destructive hurricane,” notes Thill.
The way I see it, the AI boom will more likely help accelerate growth in the software ecosystem by creating new opportunities and transforming how software is developed and used, rather than destroying it.
Assif Shameen is a technology and business writer based in North America
