I recently observed a very interesting phenomenon. Since February, the entire SaaS sector has been crushed, with Salesforce dropping nearly 40%, and ServiceNow plummeting 11% in a single day, just because management mentioned that AI agents might impact seat growth. The market logic is very simple and crude: AI can replace manual work, so companies won’t need as many software seats, and the pay-per-user model should die. This wave of panic has been dubbed the "SaaS Doomsday" by the media.

But just as this pessimistic sentiment is overwhelming, HSBC’s US tech research head released a provocative report titled "Software Will Devour AI." His core argument directly counters the entire market narrative: AI will not eliminate software; instead, it will be absorbed by software, becoming an embedded capability layer within enterprise systems.

Honestly, this logical reversal is quite fascinating. The market is panicking over "AI replacing software," while this HSBC research report at the code level is saying "software will tame AI." He draws an analogy to the internet era: when the internet exploded, early value was in infrastructure, but the ultimate winners were software companies built on that infrastructure. The same logic applies to AI now—2024 and 2025 are the foundational period for computing power and models, and 2026 will be the year software monetization kicks off.

I find his defensive reasoning most convincing. Why can’t AI model companies replace enterprise software giants like Oracle and SAP? Three reasons are quite compelling. First, foundational models have inherent flaws; LLMs trained on public internet data simply cannot learn the private architectures and business logic accumulated over decades in enterprise systems. Second, the so-called vibe coding (natural language code generation) capability is seriously overestimated; it merely shifts design responsibility to developers, and those AI model companies lack enterprise-level software operation and maintenance experience. Third, the high switching costs for enterprises are real barriers—99.999% uptime isn’t achieved by code alone, but through time and trust.

What’s most interesting is his characterization of AI agents. The market views agents as disruptors of software, but this report believes agents must operate within parameters and permissions defined by software. In other words, what enterprises need isn’t omnipotent AI, but AI that can be governed, audited, and operate within compliance frameworks. Only deeply embedded intelligent agents within enterprise software systems can achieve this. Software is the key pathway for enterprises to control and utilize AI—this is the core judgment of the entire report.

Based on this logic, HSBC’s investment advice is very clear: the current valuation of the software sector has fallen to historic lows, and the dawn of AI monetization is imminent—this is an entry opportunity, not a signal to exit. Their buy list includes core players like Oracle, Microsoft, Salesforce, ServiceNow, Palantir, CrowdStrike, Alphabet, etc. The key is that only those software companies with deep data moats, embedded AI capabilities, and not relying solely on per-user billing models will truly benefit from this wave of monetization.

However, I think there’s a question this report didn’t directly answer: if AI agents can truly operate efficiently within software frameworks, will enterprise demand for software "seats" continue to shrink? The value logic of software as an AI carrier might hold, but whether the "pay-per-user" business model can sustain current valuations remains uncertain. Every financial report in 2026 will provide new evidence for this debate.